Release stability improvements as version 2.0.1

Adoption returns to the repaint mask (option A) and layer blending stays opt-in via ◐. Right-click/Escape now flattens any remaining layer and releases the region in one action, so the separate ✓ commit button is gone.

.github/workflows/ci.yml

@@ -16,6 +16,9 @@ jobs:
       - uses: actions/setup-python@v7.0.0
         with:
           python-version: '3.12'
+      - uses: actions/setup-node@v7.1.0
+        with:
+          node-version: '22'
       - name: Install CPU test dependencies
         run: |
           python -m pip install torch --index-url https://download.pytorch.org/whl/cpu
@@ -32,7 +35,11 @@ jobs:
           for source in pathlib.Path('web').glob('*.js'):
               subprocess.run(['node', '--input-type=module', '--check'], input=source.read_bytes(), check=True)
           PY
-      - name: Check exclusive routing behavior
-        run: node --experimental-default-type=module tests/test_exclusive_routes.mjs
-      - name: Check comparison result history
-        run: node --experimental-default-type=module tests/test_compare_history.mjs
+      - name: Run all frontend regression tests
+        run: |
+          python - <<'PY'
+          import pathlib, subprocess
+          for test in sorted(pathlib.Path('tests').glob('test_*.mjs')):
+              print(f'Running {test}', flush=True)
+              subprocess.run(['node', '--experimental-default-type=module', str(test)], check=True)
+          PY

CHANGELOG.md

@@ -1,5 +1,146 @@
 # Changelog
 
+## 2.0.1 · stability fixes (2026-11-10)
+
+- Release late preview object URLs after sibling failures, discard corrupt cached PNGs, bound decoding and cancel stale preview reads. Downloads and both ordinary/crop adoption use bounded transfers too.
+- Add upload inactivity detection and cancellation without limiting the overall duration of advancing XHR transfers; keep authentication fallback and real byte progress.
+- Dispose workspaces safely during initialization/validation, coalesce background checks, and hand submitted task IDs to a reopened workspace before result recovery. Recovery selects sliding comparison even when the result was already cached.
+- Share one exact-task history poller between node and workspace; cancel disposed recovery requests and release discarded pending states.
+- Keep the last good source usable after an invalid replacement. Protect source rollback from intervening completed strokes/crop changes and release partially decoded masks and retired blend canvases.
+- Bound preview-cache batch metadata to 512 batches as well as the existing byte limit. Empty/invalid batches leave cached images unchanged.
+- Find dense mask bounds from row/column occupancy instead of allocating coordinates per painted pixel; sanitize into a private tensor with one less full-image allocation.
+- Reject malformed workflow input mappings before core replacement hooks can crash. Existing node IDs, workflows, masks, crop layers, presets and sampling behavior are retained.
+
+## 2.0.0 · 2026-11-09
+
+- Ship the standalone Adachi Workbench with English default labels and Simplified Chinese translations following ComfyUI's language setting.
+- Add one official text-to-image / inpainting workflow with group bypass presets and automatic exclusive routing, using neutral prompts and a user-selected checkpoint.
+- Use a transparent title treatment with semi-transparent black lettering instead of a white title background.
+
+- Rename the page brand to Adachi Workbench and add a remembered, keyboard-accessible canvas/settings divider.
+- Remove personal-template node IDs and preset/model-name matching. Discover components from the native execution graph, including Use Everywhere and subgraphs.
+- Preserve intermediate model, conditioning, image and output processing. Add explicit pipeline/controller selection for ambiguous workflows.
+- Distinguish inherited LoRA chains from custom overrides; preserve split CLIP paths and allow execution-only custom LoRAs without a loader in the editable graph.
+
+- Keep the canvas and editing controls available during generation; submit an independent snapshot and defer automatic view switching while a pointer stroke is active.
+- Show actual execution stages and sampler steps above workflow settings, including queued, cached, reconnect and completion states.
+- Return to mask editing after adopting a crop, preserving its region, result layer and repaint mask.
+- List missing model components and their model folders, and expose separate text encoder / VAE choices for Anima-style profiles.
+
+- Detect missing backend routes before submitting and support API-prefixed proxy paths.
+- Add remembered light/dark themes, compact edit and before/after icons, hover help and default sliding comparison after generation.
+- Follow sampler conditioning through Get/Set and exclusive routes; derive preset buttons from the enabled controller and warn when unavailable.
+- Keep crop and result controls above the measured source-history dock on narrow screens.
+- Recover lost result notifications by queued task ID in both node and workspace; add streamed image progress, inactivity timeouts, cancellation, automatic retries and reconnect recovery.
+
+## 1.0.1
+
+- Retry comparison image loading after network errors or timeouts and add a manual reload button. Expired preview references still require a new generation.
+
+## 1.0.0
+
+- First stable release of the inline inpainting workspace, comparison, crop blending and workflow switching tools.
+- Includes in-memory comparisons and source upload progress from 0.9.0. Update the bilingual crop instructions and example to match the current editing controls.
+
+## 0.9.0
+
+- Keep new comparison originals and results in a bounded in-memory cache instead of creating preview files. Existing disk-backed preview references remain supported; adopted sources and layers remain persistent input files. The cache defaults to 512 MiB, adjustable with `ADACHI_COMPARE_MEMORY_MIB` (64–4096 MiB); restart or eviction expires old previews.
+- Show actual browser upload progress for source images, followed by a loading indication. Authentication fallback uses an indeterminate indicator rather than inventing a percentage.
+
+## 0.8.2
+
+- Actually display a newly uploaded image when its dimensions differ; the incompatible old mask is reset. Same-size replacement continues to preserve the mask.
+- Automatically fill connected inpaint dimensions from an uploaded image in Mask mode and from a locked crop in Crop & Blend mode. Existing saved workflow dimensions and linked parameter inputs are respected.
+
+## 0.8.1
+
+- Expire only this plugin's comparison preview PNGs after 24 hours, checking hourly even in a long-running ComfyUI. Adopted input images and unrelated temporary files are preserved. Cached comparisons refresh before the expiry deadline.
+- Add a compact download button for the latest generated image, including the last image in a batch, independently of the history selection.
+
+## 0.8.0
+
+- Restructure adoption in Crop & Blend: after "Adopt this result" the layer is baked into the source at full quality and the editor returns to the repaint mask immediately, so you can generate again right away. The ◐ button now enters layer blending explicitly instead of auto-entering after adoption.
+- Remove the ✓ merge step entirely: right-click (or Escape) now releases the selected region and flattens any remaining layer in one action, so there is no intermediate commit button between adopting a result and picking the next region.
+- Speed up inline mask change detection: the source image is fingerprinted by size and mtime instead of hashing its full high-resolution bytes on every change/execution.
+
+## 0.7.2
+
+- Adopting a result no longer decodes the full-size image a second time: the already-decoded image is reused for the crop layer, cutting one download/decode round trip.
+- Show a visible message when adoption is skipped because the preview is still loading, no target is selected, or the crop/layer state changed, instead of failing silently.
+- Bump the asset versions for the crop, comparison and result-adoption modules.
+
+## 0.7.1
+
+- Uploading a new source image now keeps the painted mask, undo/redo history and viewport, the same way adopting a result does; only the source pixels change. A first upload still starts with an empty mask.
+- When the new image has a different size the mask cannot be kept, so the image loads fresh and a notice explains that the mask was reset instead of failing the upload.
+
+## 0.7.0
+
+- Add a download button next to the inline upload button: it saves the current source image as a PNG at its own resolution. It stays disabled until an image is loaded.
+- Change the pin toggle so it freezes the view exactly as it is drawn instead of re-fitting the image: panning and zooming stay available while pinned, and the pinned view survives canvas resizes.
+- Keep routing the plain wheel to the workflow canvas while pinned, as before.
+
+## 0.6.9
+
+- Place the Clear button directly after the swap width/height control in Crop & Blend; in Mask mode it is the only visible button of that bar and stays at the bottom left of the canvas.
+
+## 0.6.8
+
+- Add a Clear button to the bottom-left canvas bar: one click empties the mask of the active target, in both Mask and Crop & Blend modes, and it stays disabled until an image is loaded.
+- Move the selected-region painting help onto the repaint mask button, and remove hover text from the painting canvas entirely: hovering the image no longer shows any hint in any mode. The bottom controls and their button tooltips keep the instructions.
+- Remove the now-unused canvas and repaint mask hint entries from the locale dictionary.
+- Bump the inline crop module's asset version so the updated file is fetched instead of a cached copy.
+
+## 0.6.7
+
+- Before merging an adopted crop layer, hold the right mouse button to preview the original; release, pointer cancellation or window blur restores the workspace. Preview changes display only. After merge, right-click again releases the crop selection.
+- Add English/Chinese hover instructions for bottom controls and state-specific canvas gestures, including size/hardness shortcuts, masks, adoption, merging and Undo.
+
+## 0.6.6
+
+- Right-click releases a locked crop without painting; the next plain click selects a new region. Adopted layer effects are merged before reselection. Right-click no longer paints or erases in either mode; Alt remains the temporary eraser.
+- Remove the redundant reselect button. Move the existing mask Undo button from the inpaint controls to the inline toolbar, immediately after Merge, using a return-arrow icon. Undo also remains accessible in ordinary mask mode.
+
+## 0.6.5
+
+- Keep crop size presets, width/height swap and custom dimensions available after locking a region. Changing dimensions automatically unlocks it for a new selection.
+- Merge an adopted layer before changing the region, and retain the old selection if merging fails.
+
+## 0.6.4
+
+- Use a pure black dashed crop outline and automatically synchronize connected inpaint resolution when locking a crop. Linked resolution inputs remain under their upstream node's control.
+- Remove the Inline Mask result socket and obsolete feedback connections. Both modes now receive results exclusively through Image Compare adoption.
+- Crop adoption keeps the full source, crop location and repaint mask; the accepted patch remains a separate editable blend layer. Generation alone does not update it. Tag comparison previews with their originating crop to reject stale adoption.
+
+## 0.6.3
+
+- Check oversized embedded sources before workflow configuration, so the first autosave cannot disable browser draft storage before the mode-change guard runs. Upload exact source bytes before changing the file reference, including sources in subgraph definitions.
+- Keep the array and optional named widget snapshots synchronized when serializing Inline Mask; stale embedded copies no longer survive in the named snapshot. Keep repaint masks, fusion masks and layer references unchanged.
+- Apply the source-capacity check after uploading/replacing an embedded image too. Respect disabled browser persistence and retain embedded data if upload fails.
+- A page that already encountered the native draft quota error needs one reload to reset ComfyUI's disabled-storage flag: save the workflow with Ctrl+S first, then use Ctrl+F5. No browser drafts are cleared by the plugin.
+
+## 0.6.2
+
+- Remove magic wand and quick selection, including their toolbar menu and background workers.
+- Locking a crop dims the outside and starts an empty repaint mask, ready for manual painting. Brush, shape and mask actions stay inside the locked crop.
+- Reduce presets to six sizes with 64-aligned dimensions from 512 through 2048; retain the width/height swap button. Custom sizing steps by 64 pixels.
+
+## 0.6.1
+
+- Add a collapsed painting-tools/mask-settings section in WebUI Inpaint and Inpaint Prepare. Keep the complete Undo/Redo/Invert/Fill/Clear/Import/Export row visible, and save the fold state in the workflow.
+- Reproduce browser draft quota failures with a large embedded source. Before a mode change or enabling embedding, probe available draft capacity using a temporary plugin-owned key; credit the existing current draft and preserve all other browser storage.
+- If the embedded source does not fit, upload an exact PNG copy first, then use the input-file reference without reloading pixels or clearing repaint masks, fusion masks, layers or undo commands. A failed upload retains the original embedded data and mode. Small embeddings remain embedded.
+- Share in-flight recovery across embedding/mode changes, respect disabled ComfyUI browser persistence, and keep the recovery notice above the crop toolbar.
+
+## 0.6.0
+
+- Merge rectangular cropping and two-layer blending into Inline Mask, with a compact mode selector; ordinary mask mode ignores the result input.
+- Add 32-aligned crop presets, pointer-following selection, fixed-region repeat generation, exact full-image stitching and a third full_image output. Feedback wiring is split into an acyclic execution graph before validation.
+- Replace the proposed crop version history with original/result layers and a separate editable fusion mask. Brush reveals the new layer; Alt erases back to the original. Reselecting merges visible layers first.
+- Add a compact selection-tool menu: contiguous magic wand (W), brush-guided color quick selection (Q), tolerance and selection brush size. Alt subtracts, and both tools support undo/redo in either mode. Selection calculations run in a Web Worker.
+- Use arrow/fit/pin icons, Ctrl+wheel for size and Shift+wheel for hardness. Use source size follows the locked crop in crop mode.
+- Retain existing image/mask output indices, old workflows, comparison adoption and mask preservation. Restart ComfyUI and hard-refresh after updating.
+
 ## 0.5.13
 
 - Restore a missing enabled execution input by resolving its saved workflow route through Get/Set, reroutes and subgraph output slots to an existing flattened execution node.

README.md

@@ -1,227 +1,279 @@
+<div align="center">
+
 # Adachi-Inpaint-Workbench
 
-Load an image, paint a mask and inpaint **directly inside the workflow node** — no modal mask editor and no save/load round-trip on every stroke.
+**Paint. Inpaint. Compare. Keep going.**
 
-![Inline mask painting with WebUI-style inpainting](preview/nodes2-inline.png)
+An inline inpainting workspace for ComfyUI, with WebUI-style controls and tools for switching workflow branches.
 
-MIT licensed, by Adachi.
-Supports whole-image and masked-area sampling, an Alt temporary eraser, Ctrl+wheel brush size, image comparison and one-click adoption of a result as the next source image.
+**English** · [简体中文](README.zh-CN.md)
 
-## How this differs from other nodes
+[![License: MIT](https://img.shields.io/badge/License-MIT-38966b)](LICENSE)
+[![ComfyUI custom nodes](https://img.shields.io/badge/ComfyUI-Custom_Nodes-4776b4)](https://github.com/Comfy-Org/ComfyUI)
+[![Python 3.10+](https://img.shields.io/badge/Python-3.10%2B-4776b4)](pyproject.toml)
+[![Validate plugin](https://github.com/adachi0130/Adachi-Inpaint-Workbench/actions/workflows/ci.yml/badge.svg)](https://github.com/adachi0130/Adachi-Inpaint-Workbench/actions/workflows/ci.yml)
 
-| | ComfyUI's built-in mask editing | This node pack |
-|---|---|---|
-| Where you paint | A **modal editor** opens over the graph (`MaskEditor`); you close it to get back to the workflow | You paint **on the node's own canvas**, and the mask lives in the workflow |
-| Per-stroke cost | Commit the mask in the dialog, then re-run | Every stroke updates the mask in the browser and is stored with the workflow; no disk save/load per stroke |
-| Sampling | Left to the workflow you build | The connected **WebUI Inpaint** node carries the sampling controls (`area`, `width/height`, `padding`, `mask_blur`, `mask_grow`, fill mode, sampler, Refiner) |
+[Features](#features) · [Installation](#installation) · [Quick start](#quick-start) · [Node reference](#node-reference) · [Examples](#example-workflows)
 
-The built-in `Painter` node draws an image and exposes a mask, but it stops there — it has no inpaint pipeline. The built-in `MaskEditor` edits masks in a modal dialog. `ComfyUI-Inpaint-CropAndStitch` does crop-and-stitch well but does not paint. This pack connects the whole loop in one place: **paint → sample → composite at original resolution → compare → adopt the result as the next source and keep painting.**
+![Inline painting, WebUI-style inpainting controls and image comparison in one ComfyUI workflow](preview/nodes2-inline.png)
 
-## Install
+*The original workbench preview. Earlier toolbar screenshots are retained below; v0.6 uses arrow / fit / pin icons.*
 
-Copy the whole `Adachi-Inpaint-Workbench` folder into `ComfyUI/custom_nodes/`, keeping `web/`, `backend.py` and `__init__.py`.
-Restart ComfyUI, then hard-refresh the browser with `Ctrl+F5`.
-Search for `Inline Mask` or `WebUI Inpaint`; the category is **Adachi-Inpaint-Workbench**.
-The node pack follows the ComfyUI language setting: node names, sockets and tooltips come from `locales/`, and the canvas buttons and parameter panel from `web/i18n.js`, so a Chinese ComfyUI shows the Chinese labels. English is the built-in default, and any missing entry falls back to English.
+</div>
 
-On Windows you can instead run the installer from the unzipped plugin folder, pointing it at your ComfyUI directory (the one containing `main.py`).
+## Why this workbench?
 
-```powershell
-powershell -ExecutionPolicy Bypass -File .\install.ps1 -ComfyUIPath "D:\ComfyUI_windows_portable\ComfyUI"
-```
+Inpainting usually means moving between the workflow, a mask editor and the generated result. This pack keeps that loop on the graph: paint directly on a node, generate, compare, then adopt a result as your next source **without losing the painted mask**.
 
-The installer only copies this plugin; on update it zips the previous installation into `user/adachi-workbench-backups/` and does not restart the server.
-The plugin uses ComfyUI's existing PyTorch, NumPy and Pillow — no extra pip or npm packages.
-To uninstall, close ComfyUI and delete `custom_nodes/Adachi-Inpaint-Workbench`.
+| Tool | What it adds |
+| --- | --- |
+| **Inline Mask** | Paint masks or crop, inpaint and blend two layers in the same node. |
+| **Adachi Inpaint Workbench** | Whole-image or masked-area sampling, familiar inpainting controls and compositing back at the source resolution. |
+| **Image Compare** | Hover comparison, result history and one-click adoption for another editing round. |
+| **Group Bypass Presets** | Save combinations of enabled/bypassed groups and apply them with one click. |
+| **Auto Exclusive Switch** | Route the only enabled branch to a shared output, without a manual switch widget. |
 
-### ComfyUI Manager and Comfy Registry
+English and Simplified Chinese interfaces follow the ComfyUI language setting. The plugin uses the PyTorch, NumPy and Pillow already available in ComfyUI; no extra frontend build is required.
 
-Manager **Git URL** installation can use `https://github.com/adachi0130/Adachi-Inpaint-Workbench`, or clone it into `ComfyUI/custom_nodes`:
+## Features
+
+### Adachi Workbench · light and dark themes
+
+Click **Open Adachi Workbench** on an Adachi Inpaint Workbench node. Workflows need the plugin’s Inline Mask, inpaint and Image Compare nodes; their IDs, titles, preset names and model filenames can be different. The workspace follows the selected node’s actual connections to its editor and result. If several candidates are ambiguous, choose them under **Workflow connections**. An enabled Group Bypass Presets node supplies its own preset buttons; missing or ambiguous controllers show a reminder.
+
+Models, VAE, text encoders and prompts are discovered from the native execution graph, including Get/Set, Use Everywhere broadcasts and subgraph execution IDs. Intermediate model patches, ControlNet, multiple conditioning encoders, image processing and result postprocessing remain connected. Recognizable components get editing controls; other custom processing remains under the workflow’s control. No personal template IDs or model-family name matching are used.
+
+New profiles inherit the workflow’s LoRA chains, including different positive/negative CLIP paths. Clear **Inherit workflow LoRAs** to use a custom list; an empty custom list disables recognized LoRAs. Custom LoRAs can also be applied when the editable graph has no LoRA loader. Refiner use is available in the model menu. Existing explicit workspace selections remain stored per preset.
+
+Drag the divider between the image and settings to resize them. The width is remembered; double-click resets it, and arrow keys make small adjustments. Narrow screens keep the stacked layout. The theme icon remembers the light/dark preference. Hover controls for help.
+
+Generation shows actual workflow stages and sampler steps above settings. You can keep painting, zooming and changing parameters; the queued task uses its submitted snapshot. Generated results open sliding comparison without replacing the source. Use the single before/after icon to toggle images, the pencil to edit, and **Adopt** to adopt. Crop adoption returns to editing with its region and mask intact. Source history can restore previous images, masks and crop state for the loaded browser session.
+
+The comparison node and workspace recover lost execution notifications by queued task ID. Image loading reports actual received bytes, a percentage when size is known, then decoding; transient failures retry and can resume after reconnecting. Expired server-side previews require another run. Recovery does not create preview files. Updating backend code requires a full ComfyUI restart; frontend-only updates need a hard refresh.
+
+The [official workflow](examples/Adachi-Workbench.json) is a starting point, not a required template. Validation also covers independently built multi-pipeline workflows, changed node IDs and renamed presets. Labels default to English; choose Simplified Chinese in ComfyUI settings for the Chinese interface. Other locales use the English fallback.
+
+### Paint directly on the graph
+
+Upload, drop or paste a source image and paint on **Inline Mask** itself. In Mask mode, uploading an image automatically fills the connected inpaint dimensions; a replacement of the same size keeps the mask, while a different-sized image starts with a fresh mask. Mask strokes stay in the browser and are saved with the workflow; painting does not require a disk save/load round-trip for every stroke.
+
+While a source image uploads, a progress bar reports the bytes actually sent and then shows when the image is loading. On platforms requiring a special authentication fallback, the indicator stays indeterminate rather than showing an estimated percentage.
+
+- Brush, soft edges, hardness, opacity, pen pressure, temporary erasing and shape tools.
+- A compact **↑ upload / ⛶ fit / pin** toolbar, with a mode selector at the upper right. Enable the pin (**Keep**) to keep the image fitted and centered; plain wheel then navigates the workflow, while `Ctrl` + wheel still changes brush size and `Shift` + wheel changes hardness.
+- Resize the editor horizontally or vertically. The pin state and node dimensions survive workflow reload.
+- The connected **Adachi Inpaint Workbench** node hosts the painting controls and sampling parameters, leaving more space for the picture. Its painting tools, mask mode and fill content fold behind an arrow; **Undo / Redo / Invert / Fill / Clear / Import / Export** remain visible. Fold state is saved with the workflow.
+- Bounded display buffers and cached image/mask compositing reduce redraw cost at large graph zoom; the saved mask retains its source resolution.
+
+<details>
+<summary>View the compact toolbar</summary>
+
+<p align="center">
+  <img src="preview/inline-mask-toolbar.png" alt="Inline Mask with Upload, Fit and the active Keep pin" width="300">
+</p>
+
+*Toolbar reference supplied by the author; the source preview is unloaded in this screenshot.*
+
+</details>
+
+### Crop and blend in the same node
+
+Choose **Mask** or **Crop & Blend** at the upper right of Inline Mask. Both modes receive images through **Adopt this result** in Image Compare; Inline Mask has no result input socket.
+
+In Crop & Blend mode, choose one of six 64-aligned presets spanning 512–2048 pixels per side, including 1024 × 1536 (or custom dimensions in multiples of 64). The dashed rectangle follows the pointer by its upper-left corner; left-click locks it. The outside is darkened. A new region starts with an empty repaint mask. Paint inside the rectangle; brush and shape operations stay within its bounds. Use **◌ repaint mask** to return to repaint-mask editing. The position stays fixed for repeated passes. Size presets remain editable while locked; changing the preset, orientation or custom dimensions cancels the current selection and returns to pointer placement, merging any adopted layer first. The outline is a pure black dashed rectangle, and locking it automatically fills the connected Inpaint node's width and height from the crop, using the same rules as Use source size.
+
+| Connection | Purpose |
+| --- | --- |
+| `Inline Mask.image` + `mask` → `Adachi Inpaint Workbench.image` + `mask` | Send the selected crop and its repaint mask. |
+| `Inline Mask.image` → `Image Compare.original`; `Adachi Inpaint Workbench.image` → `Image Compare.result` | Compare the crop before adopting it. |
+| `Inline Mask.full_image` → `Save Image.images` | Save the current full image with the **adopted** layer stitched in. |
+
+Generate, inspect the result in **Image Compare**, then click **Adopt this result** to return it. There is no feedback wire and generation does not automatically replace the editor image. Use the inpaint node's **image** output for comparison; its `crop` output is the sampling canvas and can have a different size. In the [official workflow](examples/Adachi-Workbench.json), select **Inpaint**, then switch Inline Mask to **Crop & Blend**.
+
+After adoption, the result appears in the selected area and the editor returns to the repaint mask, ready for another pass. Select **◐ blend layers** to paint the layer mask: brush keeps new pixels, **Alt** restores original pixels, and soft edges blend seams. The blend mask is independent of the repaint mask. While editing the blend mask, **hold the right mouse button** to preview the original and release it to return to the workspace. To select a new region, switch to **◌ repaint mask** and right-click (or press Escape); the visible result is merged before the selection is released. Left-click places the next region. The **↶ Undo** button undoes strokes in the active mask. Hover the bottom controls for instructions. For a generated batch, select the desired result in Image Compare before adopting it. A result tagged with an older crop is rejected after selecting another region.
+
+### Inpaint the whole image or just the masked region
+
+| Sampling area | Behaviour |
+| --- | --- |
+| **Whole image** · `whole_image` | Resize the full source to the sampling canvas, generate, then composite through the mask. |
+| **Only masked** · `only_masked` | Crop the mask bounds with padding and aspect-ratio context, generate at the sampling size, then paste the region back into place. |
+
+Both modes return a final `image` at the **source image's resolution**. Sampling `width` and `height` are independent controls; **Use source size** fills them from the uploaded image, rounded to the sampler's required multiple of 8.
+
+The panel includes mask inside/outside, original/fill/noise/zero-latent content, denoise, sampler, scheduler, steps, CFG and seed. Advanced controls cover blur, growth, extra feathering, batch size and optional Refiner staging.
+
+### Compare, adopt and refine
+
+Move the pointer over **Image Compare**: **result on the left, original on the right**. Leave the picture to see the full result. The divider stays thin, the image stays fitted, and scrolling navigates the workflow background.
+
+Loading errors and timeouts are retried automatically; **↻** reloads the comparison manually. If the temporary memory preview has expired after server restart or cache eviction, run the workflow again.
+
+Results accumulate within an editing round, with the newest selected automatically. Each result retains its matching original; replayed preview notifications are not counted twice.
+
+Use the **↓** button to download the latest generated result at its original resolution, even when an older result is selected in the list. For a batch, it downloads the last image in that batch.
+
+Click **Adopt this result** to replace the inline source while preserving its mask and current-session undo/redo history. The comparison list resets to **1/1**; the next generation starts a new round. Failed adoption leaves the list intact. Adoption never queues a generation automatically.
+
+### Switch entire workflow setups with presets
+
+Save group states for your own setups: text-to-image, inpainting, model variants, ControlNet, upscaling, or any other grouped branches.
+
+![Group Bypass Presets with a preset list on the left and per-group states on the right](preview/group-bypass-presets.png)
+
+Choose a preset on the left, configure **Enabled / Bypassed / Keep current** on the right, or apply it from the node's right-click menu. Presets are stored in the workflow and need no wires or third-party group controller. Missing groups are reported; overlapping groups give the smaller group's setting priority.
+
+### Automatically use the remaining branch
+
+Connect two alternatives to **Auto Exclusive Switch** and bypass one. The remaining enabled branch flows to the shared output, with no `switch` toggle to maintain.
+
+![Two Auto Exclusive Switch nodes routing positive and negative conditioning into Adachi Inpaint Workbench](preview/auto-exclusive-switch.png)
+
+Use it for conditioning, strings, images, masks, models or latents; the two alternatives must use the same data type. Both enabled, or neither available, is an error rather than a silent preference for A.
+
+KJNodes Get/Set routes and reroutes are supported. **Both Gets may stay enabled**: bypassing a branch's corresponding Set or source disables that branch. Workflow-aware server checks also handle saved subgraph output routes when the source node exists in the submitted execution graph.
+
+## Installation
+
+### Git clone
+
+Run from your `ComfyUI/custom_nodes` directory:
 
 ```bash
 git clone https://github.com/adachi0130/Adachi-Inpaint-Workbench.git
 ```
 
-A Manager **listing** submission is open at [ComfyUI-Manager PR #3378](https://github.com/Comfy-Org/ComfyUI-Manager/pull/3378) and still needs the maintainer to merge it and refresh the index, so search visibility is not guaranteed yet. A Comfy Registry publisher is not configured; see [the publishing notes](registry/README.md).
+**Restart ComfyUI completely, then hard-refresh the browser with `Ctrl+F5`** (`Cmd+Shift+R` on macOS). Search for `Inline Mask`, `Adachi Inpaint Workbench` or `Adachi` in the node menu.
+
+### Other installation options
+
+- **ComfyUI Manager:** search for `Adachi-Inpaint-Workbench` in the Registry-backed Manager. Version availability depends on Registry validation; check the [Registry listing](https://registry.comfy.org/nodes/adachi-inpaint-workbench). You can also use the Manager's Git URL installation option with this repository's URL.
+- **ZIP:** [download the repository](https://github.com/adachi0130/Adachi-Inpaint-Workbench/archive/refs/heads/main.zip), extract it, and place the plugin folder at `ComfyUI/custom_nodes/Adachi-Inpaint-Workbench/`. Keep `web/`, `backend.py` and `__init__.py` together.
+- **Windows installer:** from the extracted plugin folder, run the following, pointing to the ComfyUI directory containing `main.py`:
+
+```powershell
+powershell -ExecutionPolicy Bypass -File .\install.ps1 -ComfyUIPath "D:\ComfyUI_windows_portable\ComfyUI"
+```
+
+The installer backs up an existing installation to `user/adachi-workbench-backups/`. It does not restart ComfyUI.
+
+For a Git installation, update with `git pull` inside the plugin folder, then restart ComfyUI and refresh the browser. Keep only one copy of this node pack in `custom_nodes`.
 
 ## Quick start
 
-1. Drag `examples/01-inline-inpaint.json` into ComfyUI and pick your checkpoint.
-2. In **Inline Mask**, upload, drop or paste an image and paint the areas to repaint directly on the canvas.
-3. Edit the positive/negative prompts, then set the painting tools, sampling area, sampling resolution, sampler and denoise in the **WebUI Inpaint** node, and run.
-4. `image` outputs the full composited picture, `crop` outputs the actual sampling canvas, and `mask` outputs that canvas's mask.
+1. Drag [Adachi-Workbench.json](examples/Adachi-Workbench.json) into ComfyUI. Replace **YOUR_CHECKPOINT.safetensors** with your own compatible checkpoint; the template does not download models or require a LoRA.
+2. Click **Text to image** in Group Bypass Presets, edit the neutral example prompts, and queue to create an image.
+3. Click **Inpaint**. In **Inline Mask**, upload or drop a source image and paint the region to change. **White repaints; black preserves.** Orange is only the on-screen overlay.
+4. Set the sampling area, dimensions and denoise in **Adachi Inpaint Workbench**. Queue on the graph, or click **Open Adachi Workbench** to edit in the separate workspace.
+5. Compare the output. Adopt a result when you want it to become the next source, and continue editing with the mask intact.
 
-**White repaints, black preserves; orange is the painted area on the canvas.**
-With `mask_mode=outside` this inverts and everything outside the painted area is repainted.
-`mask_blur`, `mask_grow` and extra feathering change the effective mask boundary, so blend pixels at that boundary change too; pixels where the effective mask is zero are preserved exactly.
+```mermaid
+flowchart LR
+    Source[Inline Mask] -->|image + mask| Inpaint[Adachi Inpaint Workbench]
+    Inputs[Model · VAE · Conditioning] --> Inpaint
+    Source -->|original| Compare[Image Compare]
+    Inpaint -->|result| Compare
+    Inpaint --> Save[Save Image]
+    Compare -. Adopt this result .-> Source
+```
 
-## The two sampling areas
+### Essential connections
 
-| `area` | What it does | Behaviour |
-|---|---|---|
-| `whole_image` | Repaint at the source extent | Scales the whole image to `width × height`, samples, then composites back through the mask |
-| `only_masked` | Repaint at the mask extent | Takes the mask bounding box, adds `padding` context, expands toward the target ratio, upscales that region to the sampling size, and pastes the result back in place |
+| From | To |
+| --- | --- |
+| `Inline Mask.image` / `mask` | `Adachi Inpaint Workbench.image` / `mask` |
+| Your model / VAE / positive / negative conditioning | The matching inputs of `Adachi Inpaint Workbench` |
+| `Inline Mask.image` | `Image Compare.original` |
+| `Adachi Inpaint Workbench.image` | `Image Compare.result` and/or `Save Image.images` |
 
-`width/height` are the sampling canvas size and must be multiples of 8; the final `image` keeps the uploaded image's resolution in both modes.
-That differs from the WebUI whole-image mode, which usually outputs at the target size — take `crop`, or add a resize node after the full output, if you want a target-size image.
-`padding`, `mask_blur` and `mask_grow` are all measured in source pixels.
-When the mask touches an image edge, masked-area mode uses aspect-preserving fill to keep the entire mask instead of cropping it to match the ratio.
-`resize_mode` (stretch, center crop, aspect fill) applies to `whole_image`; masked-area mode automatically uses the transform that keeps the whole mask.
+The Refiner inputs are optional. Leave them disconnected unless you are using a compatible second-stage model.
 
-## Canvas controls
+## Node reference
 
-| Action | How |
-|---|---|
-| Brush / eraser | Toolbar toggle, `B` / `E`; hold `Alt` or use the right mouse button to erase temporarily and release to restore the previous tool; the eraser cursor and a pale pink dashed outline show the erase mode |
-| Rectangle / freehand lasso | `R` / `L`, drag out the region to repaint |
-| Brush size / hardness / brush opacity | Source-pixel brush size, edge hardness and painting opacity |
-| Orange overlay | Display opacity only; it never changes the real mask |
+All nodes live under **Adachi-Inpaint-Workbench**. The group preset controller is frontend-only; the other six are user-facing backend nodes.
+
+| Node | Purpose | Outputs |
+| --- | --- | --- |
+| **Inline Mask** | Mask painting or crop/inpaint/layer blending | `image`, `mask`, `full_image` |
+| **Adachi Inpaint Workbench** | Prepare, sample and composite in one node | `image` = full composite; `crop` = sampling canvas; `mask` = sampling mask |
+| **Image Compare** | Preview, compare and adopt results | `image` = unchanged result batch |
+| **Group Bypass Presets** | Apply saved group states | No connections needed |
+| **Auto Exclusive Switch** | Forward exactly one enabled alternative | `output` of the selected type |
+| **Inpaint Prepare** | Crop/resize pixels and mask for a custom sampler workflow | `pixels`, `mask`, `context` |
+| **Inpaint Composite** | Return a decoded local result to source coordinates | `image` |
+
+For modular sampling: `Prepare.pixels → VAE Encode → Set Latent Noise Mask → sampler → VAE Decode → Composite`. Connect `Prepare.mask` to the latent mask node and **always connect `Prepare.context` to `Composite.context`**.
+
+## Canvas shortcuts
+
+| Action | Shortcut / control |
+| --- | --- |
+| Brush / eraser / rectangle / lasso | `B` / `E` / `R` / `L` |
+| Temporary eraser | Hold `Alt`; release to return to the previous tool |
+| Brush size | `Ctrl` + wheel, or `[` / `]` |
+| Brush hardness | `Shift` + wheel |
+| Preview original while blending | In **◐ blend layers**, hold right mouse button; release to restore the workspace |
+| Release crop selection | In **◌ repaint mask**, right-click or press Escape; left-click to select a new region |
 | Undo / redo | `Ctrl+Z` / `Ctrl+Shift+Z`, or `Ctrl+Y` |
-| Zoom / pan | Wheel to zoom; drag with Space, the middle button or `H` to pan; **Fit** to reset |
-| Brush size keys | `[` / `]` |
-| Fast brush size | With the pointer over the canvas, hold `Ctrl` and wheel up/down; the plain wheel still zooms the image |
-| Editor height | Drag the `↕` bar at the bottom of the editor; the size is saved with the workflow |
-| Invert / fill / clear | Act on the current painted mask and are undoable |
-| Import / export mask | Full-resolution PNG; greyscale images import by brightness, images with alpha import by alpha |
-| Image / mask / overlay | Preview toggles only; they do not change the output |
+| Image zoom / pan, when unpinned | Wheel / Space + drag, middle-button drag, or the pan tool (`H`) |
+| Keep the image fitted | Pin beside **Fit**; click again to unlock |
+| Adjust editor height | Drag the bottom `↕` handle |
 
-Brush pressure and soft brushes are supported; the mouse uses a fixed pressure.
-Inline Mask keeps the image editing area and small **Upload** / **Fit** / pin buttons in the corner, and adjusts the canvas height to the image ratio after loading to reduce empty margins.
-The brush, eraser, rectangle, lasso, pan, four brush sliders, undo/redo, mask actions, view toggles, fit and the embed option all live in the connected **WebUI Inpaint** node and drive that same inline canvas.
-You upload the image once; every stroke then updates the mask in the browser and is stored as PNG data in the workflow and in the executed parameters, so there is no saving and reloading of files per stroke.
-Wheel events over the picture zoom it; over the toolbar, buttons, sliders or the parameter panel they follow ComfyUI's canvas navigation settings.
-A continuous wheel gesture that starts on the ComfyUI canvas keeps navigating the outer canvas when it moves over the picture; scrolling inside the picture again after stopping zooms the image.
-Very large images still use browser memory, and PNG encoding time depends on resolution and hardware, so instant completion is not guaranteed for every image.
+The painting panel also provides invert, fill, clear, PNG mask import/export, image/mask/overlay views and **Embed source in workflow**. Overlay opacity affects display only. Scrolling over buttons and parameter controls navigates the outer workflow; a gesture starting on the workflow keeps navigating it when the pointer crosses the picture.
 
-## Painting tools and inpainting parameters
+## Example workflows
 
-The Inline Mask header has **Upload**, **Fit** and a pin toggle (**Keep** / **保持**). Enable the pin to keep the picture fitted and centered: plain wheel navigates the workflow, Ctrl+wheel still adjusts the brush, and image panning is disabled. Disable the pin to resume image zoom and pan. The pin state is saved with the workflow. Display buffers are bounded during workflow zoom; the saved mask retains its original resolution.
+Use the single [official template: Adachi-Workbench.json](examples/Adachi-Workbench.json). It includes text-to-image and inpainting branches, a shared checkpoint, Group Bypass Presets, automatic exclusive conditioning/image selection, comparison and saving. The preset controller stays outside the groups it controls. Select one preset at a time; enabling both branches intentionally fails exclusive selection.
 
-Once Inline Mask is connected to the `image` input of **WebUI Inpaint**, the original painting controls appear in that inpaint node, and the parameter panel below drives the node's real sampling parameters.
-Width and height each get their own row with both a slider and a number field; denoise is a slider too, and **Use source size** fills in the source dimensions, rounded to the nearest multiple of 8 the sampler requires.
-Mask mode, fill, sampling area, sampler, scheduler, steps, CFG and seed are shown directly; mask processing, batch and Refiner switching live under **Mask, batch and advanced settings**.
-If one inline source feeds several inpaint nodes, the painting controls go to the first WebUI Inpaint node and each node keeps its own parameter panel; disconnecting or deleting that node moves the tools to the next available one.
-Parameters driven by an external link disable the matching panel control; a modular workflow puts the painting tools in **Inpaint Prepare** and keeps the sampling parameters in the separate sampler node.
+All example labels are English. Prompts are newly written neutral examples; the checkpoint is a placeholder for your own model. Inpaint requires uploading a source first. Crop & Blend uses the same connected editor, so no second workflow is needed. For model families with separate model, text encoder and VAE loaders, replace the checkpoint block with compatible loaders and connect the matching outputs.
 
-## Comparison and chained inpainting
+## Compatibility and important behaviour
 
-Add an **Image Compare** node and connect Inline Mask's `image` to `original` and the final full-size generated `image` to `result`, then run.
-With the pointer over the picture, the thin divider shows the generated result on the left and the original on the right; leaving the picture hides the divider and shows the full result, and the image always remains fitted and centered in its canvas. Scrolling over the comparison image navigates the ComfyUI workflow background using the current navigation settings; comparison images do not zoom or pan independently. Batch selection, dimensions and the fit button sit below the picture.
-When you are happy, click **Adopt this result**: the plugin uploads the selected result as the new source image and writes it back into the upstream inline editor while preserving the painted mask, current undo/redo history and viewport. Use **Clear** when you want to start a new mask; a normal **Upload** still starts a fresh mask.
-To preserve mask coordinates, an adopted result must match the source dimensions; a mismatched result is rejected without changing the current source or mask. The mask remains embedded in saved workflow data; command-level undo history is retained during adoption in the current session.
-If the original is routed through intermediate IMAGE nodes the plugin walks upstream to find the inline editor; when several candidates exist you can pick one explicitly under **Adopt into**.
-Adoption only happens when you click the button and never starts the next generation automatically; the `image` output still passes the whole result batch, so you can feed a Save Image node.
-The result list accumulates images across runs and selects the newest result automatically. Each result keeps its original from that run. Successful adoption clears the old list and displays the accepted image as 1/1; the next generation starts a fresh list using the new source. Failed adoption leaves the old list intact. Saving the workflow also saves preview references, selection and the pending new-round state.
+- **Static images:** ordinary 4D `[B,C,H,W]` latents and single-frame 5D `[B,C,1,H,W]` image latents are supported, including the Wan-derived VAE path used by Anima. Supply the model family's matching VAE and conditioning. This is not multi-frame video inpainting.
+- **Anima:** for ordinary masked img2img, leave **inpaint_model** disabled. That option is for dedicated inpainting architectures, not a requirement for painting a mask.
+- **Source size:** the final composite keeps the source resolution in both sampling areas. Use `crop` or a separate resize node for a target-size output.
+- **Crop layers:** the fusion mask and crop coordinates are saved in the workflow; the latest layer is a relative image reference under `input/adachi-inpaint-layers/`. Include that file when moving an unmerged edit between machines, or merge layers and embed the source first.
+- **Sharing:** masks are embedded in saved workflows. Source images use a relative `input/` path by default; enable **Embed source in workflow** to include the picture too. Embedded images increase JSON size. If an embedded source would exceed browser draft storage when loading a workflow, uploading an image, changing modes or enabling embedding, the plugin first uploads an exact source copy and switches to an input-file reference, preserving masks and layers. To intentionally keep a very large source embedded, disable ComfyUI browser workflow persistence before embedding; exported files are not subject to the browser draft quota. If the page already reports draft failures on every edit, save with **Ctrl+S**, then reload with **Ctrl+F5** to reset ComfyUI's disabled draft-storage state.
+- **Adoption:** source and result dimensions must match to preserve mask coordinates. Undo/redo commands survive adoption in the current session; saving/reloading retains the mask, not the command history.
+- **Comparison previews:** new originals and results are held in server memory instead of temporary image files. The default cache is 512 MiB and at most 512 comparison batches; set `ADACHI_COMPARE_MEMORY_MIB` to 64–4096 before starting ComfyUI to change it. Cache eviction or server restart expires old comparisons and their download references; rerun the workflow to refresh them. A single batch larger than the limit reports an error. Previous disk-based preview references remain supported while available, and expired old plugin preview files are still cleaned up. Images adopted into the editor are saved separately under `input/adachi-inpaint/` or `input/adachi-inpaint-layers/`.
+- **Scope:** this is an original WebUI-style implementation. It does not promise identical WebUI pixels, scripts or extension behaviour. Extra soft-edge feathering is not WebUI's full Soft Inpainting algorithm.
 
-Comparison previews live in ComfyUI's temp directory and may need a re-run after cleanup or restart; an adopted source image is stored in `input/adachi-inpaint/` and does not depend on temp previews.
+<details>
+<summary>Sampling, fill and batch details</summary>
 
-## Group bypass presets
+- Sampling dimensions use multiples of 8. Padding, blur and mask growth are measured in source pixels; effective mask edges can change after blur/growth, while pixels with zero effective mask remain unchanged.
+- `mask_mode=outside` inverts the painted region before mask processing.
+- `resize_mode` controls whole-image resizing. Masked-area mode uses an aspect-preserving transform that retains the full mask, including at source edges.
+- Fill modes: **original**, surrounding-colour **fill**, **latent noise**, and **zero latent**. The latent modes belong to Adachi Inpaint Workbench; the modular Prepare node handles pixels and masks.
+- An empty mask or `denoise=0` skips VAE/sampling. `batch_size` is variants per input image, not ComfyUI queue count.
+- For modular multi-source batches, group variants per source (`A,A,B,B`), rather than interleaving them (`A,B,A,B`).
+- Optional Refiner uses `refiner_model` and `refiner_switch`; `0.8` runs the last 20% with the Refiner. Compatible Refiner conditioning is recommended. A switch of `1` uses only the base model; `0` uses only the Refiner.
+- Input limits: 64 MiB per file, 32 megapixels and 16384 pixels per image edge. Large-image painting and PNG encoding still depend on browser memory and hardware.
 
-Search for **Group Bypass Presets** (Chinese: **组忽略预设**) under Adachi-Inpaint-Workbench. It needs no connections and does not require rgthree or Danbooru.
+</details>
 
-1. Click **+ New preset**, name it and save. The preset captures the current groups' enabled/bypassed state.
-2. Choose **Enabled**, **Bypassed** or **Keep current** for each group on the right. Changes update and apply the selected preset immediately.
-3. Add more presets. Clicking a name on the left applies that preset to all of its configured groups at once. You can also right-click the node and choose **Apply preset**.
-4. Use **Save current group states** to capture changes you made directly in the workflow, or rename/delete presets as needed.
+## Troubleshooting and development
 
-Presets are stored in the workflow. Group renames retain their preset mapping; newly added groups stay unmanaged until configured. Loading a workflow restores its saved node modes without automatically applying a preset. If groups overlap, the smaller group's setting takes priority; the preset controller itself stays enabled.
-Bypass uses ComfyUI's standard bypass mode, so a bypassed node still needs compatible input/output types when it sits on an active data path.
-Try [examples/03-group-presets.json](examples/03-group-presets.json), which uses only empty images and colour inversion and needs no model.
+**Old UI after updating?** Fully restart ComfyUI and hard-refresh the browser. Backend Python changes need a server restart; refreshing the page alone is insufficient.
 
-## Fill, batch and Refiner
+**Exclusive switch reports no enabled input?** Check the real source and corresponding Set, not only the visible Get nodes. Keep the selector outside the groups it controls. If an integration still drops a subgraph output, placing the CLIP encoder outside that subgraph is a practical workaround. For a report, include the workflow, plugin/frontend versions and full error text; current errors show frontend, workflow and submitted-input flags.
 
-| `fill` | Behaviour |
-|---|---|
-| `original` | Encode the source pixels into the latent |
-| `fill` | Multi-scale fill from the colours surrounding the mask, then encode |
-| `latent_noise` | Fill the masked latent with seed-dependent noise |
-| `latent_nothing` | Zero the masked latent |
+**An image is missing after opening someone else's workflow?** Supply the source at its relative input path, or have the sender save with source embedding enabled. Expired comparison previews require regeneration.
 
-The two latent fills are implemented by the all-in-one **WebUI Inpaint** node; standalone **Inpaint Prepare** only handles pixels and masks, so choose `original` or `fill` in a standard-node pipeline.
-`denoise=0` and an empty mask return the source image directly and skip VAE and sampling; `batch_size` is the number of variants per input image, and the same seed reproduces the same batch.
-For repeated batches use ComfyUI's queue count together with the seed's after-generate control; `batch_size` controls a single run.
-Keep `inpaint_model=false` for ordinary SD/SDXL checkpoints; enable it only for dedicated inpainting models, which switches to ComfyUI's native `InpaintModelConditioning`.
+CPU regression tests:
 
-You can optionally attach a compatible `refiner_model` and use `refiner_switch` for the share of base sampling steps; `0.8` means roughly the first 80% of steps run on the base model.
-Encoding with the Refiner's own CLIP and wiring `refiner_positive`/`refiner_negative` is recommended; if left unconnected the original conditioning is reused and the VAE still comes from the base side.
-`refiner_switch=1` uses only the base model and `0` uses only the Refiner; a known-incompatible latent format raises an error.
-
-## Auto Exclusive Switch
-
-Search for **Auto Exclusive Switch** / **自动二选一** under **Adachi-Inpaint-Workbench**.
-Connect two alternative branches to `input_a` / `input_b` and connect `output` to the common downstream input.
-For example, connect the CONDITIONING outputs of two positive CLIP Text Encode nodes to A/B, then connect `output` to the sampler's positive input.
-Bypass one encoder (Ctrl+B), or use a Group Bypass Preset to enable one group and bypass the other: the remaining branch is selected automatically.
-Both enabled, or neither connected/enabled, produces an error instead of silently preferring A.
-
-The node accepts conditioning, strings, images, masks, models, latents and other ComfyUI data types; both alternatives and the downstream socket should use the same data type.
-The model-free `examples/04-exclusive-switch.json` demonstrates two branches with two group presets.
-It has no switch widget. A small internal value is computed from the graph when queuing, including when a bypassed image/model transform would otherwise pass its input through.
-Place the selector outside the two controlled groups and connect each group's final node to it.
-KJNodes Get/Set routes and reroutes are supported: an ignored/muted Get, its corresponding Set, or the source behind those routing nodes disables that selector branch. Missing setters and virtual routing cycles also count as unavailable. Ordinary processing nodes remain independent active sources; the selector does not guess the meaning of arbitrary processing chains.
-The ignored branch is not requested by this selector; any separate preview/save outputs on that branch may still execute independently.
-For raw API prompts, omit the ignored input and omit the internal `active_inputs` field; exactly one of `input_a` / `input_b` must remain.
-Without workflow metadata, the backend checks actual submitted connections after virtual routing: when only one input survives, it uses that branch even if cached frontend flags still report both enabled. With both connections present and no workflow metadata, the frontend flags remain necessary to distinguish bypassed transform pass-through from an enabled branch. Updating this backend logic requires a full ComfyUI restart as well as a browser refresh.
-When workflow metadata accompanies the request, the server independently follows Get → named Set → source node. Both Gets can stay enabled: bypassing one branch's Set or encoder disables that branch. The server normalizes flags and removes ignored execution inputs before validation and cache lookup, while the saved workflow retains its connections. Without workflow metadata the submitted connections/frontend flags remain the fallback.
-If the enabled input was dropped during frontend execution serialization, the server can restore its saved connection, including a subgraph output mapped to a flattened node ID. It only reconnects an existing submitted node whose type matches the saved source; it does not create missing model/encoder nodes.
-
-## Using a standard sampling workflow
-
-`examples/02-modular-inpaint.json` shows a wiring you can swap samplers into.
-
-```text
-Inline Mask ── image/mask ──> Inpaint Prepare
-                                pixels ──> VAE Encode ──> Set Latent Noise Mask ──> KSampler
-                                mask ──────────────────────────┘                      │
-                                context ──────────────────────────────────────┐      │
-                                                                             │      ▼
-                                                                        Inpaint Composite <── VAE Decode
-                                                                             │
-                                                                          Save Image
-```
-
-Always wire `Prepare.context` back into **Inpaint Composite** so the local result returns to the source coordinates and size.
-The modular example targets ordinary checkpoints; a dedicated inpainting checkpoint should use the conditioning nodes it needs, or the all-in-one node directly.
-When several input images each produce several variants, Composite needs the variants grouped per source image (for example `A,A,B,B`); the all-in-one node already orders them that way, while an external latent-repeat node that outputs `A,B,A,B` needs its batch order adjusted first.
-
-## Sharing and compatibility
-
-The painted mask is written into the workflow automatically; by default only a relative path inside ComfyUI's `input/` is stored for the source image.
-To hand someone a workflow that opens on the same picture, tick **Embed source in workflow** below the canvas before saving the JSON; they install this plugin, pick their own model and can keep editing.
-Without embedding you have to supply the source image as well and keep the same relative path under `input/`.
-Embedding grows the workflow file, and the source and mask then travel with the shared workflow, so check the picture content before sharing.
-
-This is an original WebUI-style implementation and does not promise pixel-perfect agreement with AUTOMATIC1111/Forge results.
-`soft_edge` is extra Gaussian feathering with `soft_strength` controlling the extra radius; it **does not reimplement WebUI's Soft Inpainting algorithm or its full parameter set**.
-Samplers, schedulers and Refiner use the current ComfyUI implementations, and the fill algorithms are this plugin's own.
-Static-image pipelines support ordinary 4D `[B,C,H,W]` latents and single-frame 5D `[B,C,1,H,W]` latents, without hardcoding a model name or channel count. This includes the Wan-derived VAE used by Anima and Qwen-image pipelines. Independent images/variants stay in the batch dimension, with their own masks; they are not encoded as consecutive video frames.
-Use the matching model, VAE and text conditioning for the model family. For ordinary Anima img2img, keep **inpaint_model** disabled; that option invokes native conditioning for dedicated inpainting architectures, and is not required merely because a mask is painted.
-Actual multi-frame video, audio/3D latents, model-specific control pipelines, WebUI scripts/extensions and automatic WebUI plugin configuration are not implemented.
-Limits are 64 MiB per file, 32 megapixels per image and 16384 px per side; no models are bundled and nothing is downloaded.
-Node loading, canvas editing, workflow/executed-parameter save and restore have been verified in an isolated CPU ComfyUI service on the development machine, and the backend is verified with mock VAE/samplers; real-checkpoint image quality and compatibility across every model family and Refiner have not been tested.
-
-## Tests
-
-Run the CPU tests from this plugin directory with your ComfyUI Python:
-
-```powershell
+```bash
 python -m unittest discover -s tests -v
+node --experimental-default-type=module tests/test_exclusive_routes.mjs
+node --experimental-default-type=module tests/test_compare_history.mjs
 ```
 
-## 中文简介
+See [verification notes](VERIFICATION.md) for tested configurations and limits, [changelog](CHANGELOG.md) for updates, and [rollback instructions](ROLLBACK.md) for returning to an earlier version. Model-family and image-quality coverage is not exhaustive.
 
-本插件把蒙版绘制放在工作流节点内部，**不需要弹出遮罩编辑器**，也省去每画一笔就保存、加载磁盘文件的往返。
-连接 **WebUI 局部重绘** 节点后，画笔工具与采样参数都在该节点内，支持整图范围与蒙版范围重绘、原尺寸回贴、Alt 临时橡皮擦、Ctrl+滚轮笔径、图像对比，以及把满意的结果一键回填为新的原图继续重绘。
-代码里的默认文案是英文，以便 Manager 用户直接使用；节点搜索可输入 `Inline Mask`，分类为 **Adachi-Inpaint-Workbench**。
+## Credits and license
 
-想用中文界面时，把 ComfyUI 的语言切成中文即可，插件会自动跟随：
-在 ComfyUI 设置里找到 **Locale / 语言**（设置项 `Comfy.Locale`）选择 `zh`，然后刷新浏览器（`Ctrl+F5`）。
-节点名称、分类、输入输出名与提示由 `locales/zh/` 提供，通过 ComfyUI 的 `/i18n` 机制加载；
-画布按钮与参数面板由 `web/i18n.js` 读取同一个语言设置翻译，因此「宽度」「高度」「上传」等也会一起变回中文。
-英文用户看到的一直是英文（英文是内置默认值）；某个词条若没有对应翻译，会自动回退为英文。
-两套语言共用同一个节点类 ID 和参数键名，所以已保存的工作流不受语言切换影响。
+Created by **Adachi / 安达**. Licensed under [MIT](LICENSE).
 
-新增「自动二选一」：把两个正面条件（或图像、模型等同类型数据）接到输入 A/B，输出接共同下游；忽略其中一个分支，就自动使用另一个。两路都启用或都无输入会报错，不需要 switch。切换节点放在两个被控制分组之外，也能配合组忽略预设使用。
+Built for [ComfyUI](https://github.com/Comfy-Org/ComfyUI), with familiar inpainting controls inspired by WebUI and hover-comparison interaction inspired by [rgthree-comfy](https://github.com/rgthree/rgthree-comfy). This repository contains its own implementation; those projects are not bundled dependencies.
 
-
-Anima 兼容：支持其 Wan/Qwen VAE 的单帧五维潜空间，也保留普通四维 VAE；会同步适配蒙版、填充、批次和解码回贴。普通 Anima 图生图请保持“专用重绘模型”关闭，并使用配套模型、VAE 与文本条件；这不是多帧视频重绘功能。
+Bug reports and workflow examples are welcome in [Issues](https://github.com/adachi0130/Adachi-Inpaint-Workbench/issues).

README.zh-CN.md

@@ -0,0 +1,279 @@
+<div align="center">
+
+# Adachi-Inpaint-Workbench
+
+**绘制蒙版，局部重绘，对比结果，继续精修。**
+
+直接在 ComfyUI 工作流内操作的重绘工作台，提供 WebUI 风格参数，以及分组预设、自动分支选择工具。
+
+[English](README.md) · **简体中文**
+
+[![MIT 许可证](https://img.shields.io/badge/License-MIT-38966b)](LICENSE)
+[![ComfyUI 自定义节点](https://img.shields.io/badge/ComfyUI-Custom_Nodes-4776b4)](https://github.com/Comfy-Org/ComfyUI)
+[![Python 3.10+](https://img.shields.io/badge/Python-3.10%2B-4776b4)](pyproject.toml)
+[![插件检查](https://github.com/adachi0130/Adachi-Inpaint-Workbench/actions/workflows/ci.yml/badge.svg)](https://github.com/adachi0130/Adachi-Inpaint-Workbench/actions/workflows/ci.yml)
+
+[功能介绍](#功能介绍) · [安装](#安装) · [快速开始](#快速开始) · [节点一览](#节点一览) · [示例工作流](#示例工作流)
+
+![在同一个 ComfyUI 工作流中绘制蒙版、设置重绘参数并对比结果](preview/nodes2-inline.png)
+
+*保留工作台早期预览图；下方保留旧工具栏截图；v0.6 使用上传箭头、适应、图钉。*
+
+</div>
+
+## 为什么做这个工作台？
+
+局部重绘时，常常需要在工作流、蒙版编辑器和生成结果之间来回切换。本插件把这段操作放回画布：直接在节点内涂蒙版，生成并对比结果，再把满意的图片作为下一轮原图，**保留已经画好的蒙版**继续精修。
+
+| 工具 | 用途 |
+| --- | --- |
+| **内联蒙版 · Inline Mask** | 同一节点内涂蒙版，或裁切、重绘、双图层融合。 |
+| **安达重绘工作台 · Adachi Inpaint Workbench** | 整图或蒙版区域采样，使用熟悉的重绘参数，按原图尺寸回贴结果。 |
+| **图像对比 · Image Compare** | 鼠标划线对比、查看结果历史，一键采用后继续重绘。 |
+| **组忽略预设 · Group Bypass Presets** | 保存各组的打开/忽略组合，一键切换工作流配置。 |
+| **自动二选一 · Auto Exclusive Switch** | 自动输出唯一启用的分支，不用手动设置 switch。 |
+
+界面支持英文和简体中文，跟随 ComfyUI 的语言设置。使用 ComfyUI 已有的 PyTorch、NumPy 和 Pillow，无需额外构建前端。
+
+## 功能介绍
+
+### Adachi Workbench · 黑白主题
+
+在任意安达重绘工作台节点中点击「打开安达工作台」。工作流需包含本插件的内联蒙版、重绘和图像对比节点，但节点编号、标题、预设名称及模型文件名均不需与示例一致。页面沿所选重绘节点的实际连线寻找图像和结果；多个候选无法唯一判断时，在「工作流连接」中选择。已启用的组忽略预设节点提供预设按钮，缺失或存在多个控制器时会提醒。
+
+模型、VAE、文本编码器和提示词从原生执行图读取，支持 Get/Set、Use Everywhere 广播及子图执行编号。模型采样设置、ControlNet、多编码器条件合并、图像处理和结果后处理会保留；可识别的组件显示编辑控件，其余自定义处理按原工作流执行，不按私人节点编号或模型家族名称重新接线。
+
+新方案默认「沿用工作流 LoRA」，保留原有 LoRA 和正负 CLIP 分叉。取消勾选后使用自定义列表，空列表会禁用已识别的 LoRA；原工作流没有 LoRA 节点也可添加自定义 LoRA。Refiner 可在模型菜单选择，旧版已明确设置的方案仍会保留。
+
+画布与配置之间的分隔条可以左右拖动，宽度会记住；双击恢复默认，方向键可微调。窄屏保持上下排列，右上角可切换并记住黑白主题。操作说明放在悬停提示中。
+
+生成时可以继续涂蒙版、缩放、移动和调整参数，已提交任务使用当时的快照。工作流设置上方显示真实阶段和采样步数；结果默认滑动对比，前后图共用一个切换图标，点击「采用」才替换原图。裁切融合采用后返回编辑，保留选区与蒙版。下方可回退或选择之前的原图，历史限当前已加载的浏览器会话。
+
+图像对比节点和网页按本次任务编号找回丢失的结果通知。图片加载显示真实接收字节、已知总大小时的百分比及解码状态；弱网失败自动重试，重连后可恢复，不产生预览文件。服务器重启或缓存淘汰造成的预览过期仍需重新运行。后端升级需重启 ComfyUI，仅前端改动强刷网页即可。
+
+[官方模板](examples/Adachi-Workbench.json) 仅作为起点，不是必需的固定工作流。已验证不同编号和预设名的副本，以及完全新建的多链路工作流。界面默认英文，在 ComfyUI 设置中选择简体中文后显示中文，其他语言暂时回退英文。模型权重须与原工作流匹配；缺失文件会列出文件名和目标目录。
+
+### 直接在工作流里画蒙版
+
+在**内联蒙版**上传、拖入或粘贴原图，然后直接在节点画布上涂抹。普通蒙版模式上传图片时会自动填写已连接的局部重绘宽高；替换同尺寸图片时保留蒙版，尺寸不同则载入新图并重置蒙版。蒙版在浏览器内更新，随工作流保存，不需要每画一笔都保存、加载磁盘文件。
+
+上传原图时，进度条显示浏览器实际发送的数据比例，上传完成后提示正在加载图片。需要特殊认证回退的平台会显示不定进度，不会虚构百分比。
+
+- 支持画笔软边、硬度、不透明度、数位笔压感、临时擦除和形状工具。
+- 顶部使用**↑ 上传、⛶ 适应、图钉**，右上角切换模式。图钉的悬停提示为「保持」：开启后图片持续适应并居中，普通滚轮控制工作流背景，`Ctrl` + 滚轮调整笔径，`Shift` + 滚轮调整硬度。
+- 支持横向、纵向调整编辑区大小；图钉状态和节点尺寸都能随工作流保存。
+- 连到 **安达重绘工作台**后，画笔工具和采样参数显示在重绘节点内，让图片编辑区保持简洁。画笔工具、蒙版内外和填充内容收进箭头折叠区，**撤销、重做、反转、全选、清空、导入蒙版、导出**整行始终显示；折叠状态随工作流保存。
+- 放大工作流时限制显示画布的分配尺寸，并缓存原图/蒙版合成，减少反复重绘；实际保存的蒙版仍保持原图分辨率。
+
+<details>
+<summary>查看简洁工具栏</summary>
+
+<p align="center">
+  <img src="preview/inline-mask-toolbar.png" alt="内联蒙版的上传、适应和已开启的保持图钉" width="300">
+</p>
+
+*作者提供的工具栏示意图；截图中的原图预览尚未载入。*
+
+</details>
+
+### 一个节点内裁切、重绘和融合
+
+在内联蒙版右上角选择**蒙版**或**裁切融合**。两种模式都通过图像对比节点的**采用此结果**回传图片；内联蒙版不再提供结果输入口。
+
+裁切融合模式保留六组常用预设，宽高均为 64 的倍数，范围 512～2048（含两端）；可用交换按钮切换横竖方向，自定义尺寸按 64 像素递增。虚线框的左上角跟随鼠标，左键点击锁定，框外压暗。新选区的蒙版为空，需要自己在框内涂抹；画笔和形状操作限制在选区内，点击 **◌ 重绘蒙版**可返回编辑重绘范围。选区框为纯黑虚线；点击锁定时，相连 Inpaint 节点会自动填写裁片宽高，等同于点一次「使用原图宽高」。选区位置固定，支持反复重绘同一处。选定后仍可切换预设、交换横竖或修改自定义宽高；修改会自动取消当前选区，重新跟随鼠标定位，已有采用图层会先合并保留。
+
+| 连线 | 用途 |
+| --- | --- |
+| `内联蒙版.image`、`mask` → `安达重绘工作台.image`、`mask` | 把裁片和重绘蒙版送去生成。 |
+| `内联蒙版.image` → `图像对比.original`；`安达重绘工作台.image` → `图像对比.result` | 先对比裁片，再选择采用。 |
+| `内联蒙版.full_image` → `保存图像.images` | 保存包含**已采用图层**的完整图像。 |
+
+先生成，在**图像对比**中查看，满意后点击**采用此结果**，才会嵌回当前裁切位置；没有回接输入线，也不会生成后自动替换。比较结果使用重绘节点的 **image** 输出；`crop` 是采样画布，尺寸可能不同。在[官方模板](examples/Adachi-Workbench.json)中选择 **Inpaint**，再将内联蒙版切换到**裁切融合**即可。
+
+采用后结果显示在选区中，编辑器回到重绘蒙版，可直接继续生成。点击 **◐ 图层融合**可编辑融合蒙版：画笔保留新图，按住 **Alt** 擦回原图，软笔刷可融合接缝；融合蒙版与重绘蒙版相互独立。编辑融合蒙版时，**按住右键看原图，松开恢复工作区**。要选择新区域，请切回 **◌ 重绘蒙版**后右键，或按 Escape；当前可见结果会先合并，再取消选定，接着左键选择新区域。**↶ 撤销**可撤销当前蒙版的涂抹，鼠标悬停底部控件可查看说明。批量生成时，在图像对比中选中想要的结果再采用；换到其他选区后，带有旧选区标记的结果会拒绝回贴。
+
+### 整图重绘与蒙版区域重绘
+
+| 采样区域 | 工作方式 |
+| --- | --- |
+| **整图范围** · `whole_image` | 将整张原图调整到采样尺寸，生成后通过蒙版合成回原图。 |
+| **仅蒙版区域** · `only_masked` | 按蒙版边界截取区域，加入 padding 和宽高比所需的上下文，采样后回贴到原位置。 |
+
+两种模式最终输出的 `image` 都保持**原图尺寸**。采样宽、高独立设置；**使用原图宽高**按钮可一键填写，并按采样器要求对齐到 8 的倍数。
+
+面板提供蒙版内/外、原图像素/填充/潜空间噪声/空白潜空间、重绘幅度、采样器、调度器、步数、CFG 和种子。高级设置包括蒙版模糊、扩张、额外羽化、批次数量和可选 Refiner。
+
+### 对比结果，采用后继续精修
+
+鼠标移入**图像对比**：**左边是重绘结果，右边是原图**；移开后显示完整结果。分隔线保持细线，图片持续适应编辑区，滚轮作用于工作流背景。
+
+网络错误或加载超时时会自动重试，也可点击 **↻** 手动重新加载；如果临时预览已因服务器重启或缓存淘汰而失效，请重新运行工作流。
+
+同一轮编辑中的结果会累计保存并默认选中最新一张，每张结果对应当时的原图；重复的预览通知不会重复计数。
+
+点击 **↓** 可以按原始分辨率下载最新生成的结果，即使列表中正在查看旧图也会下载最新一张；批量生成时下载该批最后一张。
+
+点击**采用此结果**，把选中的图片回填为内联蒙版的新原图，同时保留蒙版和当前会话中的撤销/重做记录。结果列表立即回到 **1/1**，下一次生成开启新一轮。采用失败时保留原列表；采用操作不会自动开始生成。
+
+### 一键切换整套分组配置
+
+为自己的工作流保存多套组合，例如文生图、修图、不同模型、ControlNet、放大等分支的打开/忽略状态。
+
+![组忽略预设：左侧选择预设，右侧设置每个组的状态](preview/group-bypass-presets.png)
+
+左侧点选预设，右侧配置各组**打开 / 忽略 / 保持当前**，也能通过节点右键菜单应用。预设随工作流保存，不需要连线，也不依赖第三方组控制节点。已删除的组会显示提示；分组重叠时，较小组的设置优先。
+
+### 自动使用剩下的一路输入
+
+将两路备选输入连接到**自动二选一**，忽略其中一路后，另一条有效分支自动流向共同输出，无需手动切换 `switch`。
+
+![两个自动二选一节点分别把正面、负面条件连接到 安达重绘工作台](preview/auto-exclusive-switch.png)
+
+可用于提示词条件、字符串、图像、蒙版、模型、潜空间等数据；两路输入应为同一类型。两路都启用或都不可用时会报错，不会擅自优先选择 A。
+
+支持 KJNodes 的 Get/Set 和重路由节点。**两个 Get 可以一直开启**：忽略其中一路对应的 Set 或上游源节点，就会停用那条分支。请求带有工作流信息时，后端也会核对已保存的子图输出；执行请求中仍存在相应源节点时，可恢复丢失的有效连接。
+
+## 安装
+
+### Git 安装
+
+在 `ComfyUI/custom_nodes` 目录运行：
+
+```bash
+git clone https://github.com/adachi0130/Adachi-Inpaint-Workbench.git
+```
+
+**完整重启 ComfyUI，再按 `Ctrl+F5` 强制刷新网页**，macOS 可用 `Cmd+Shift+R`。在节点菜单搜索 `Inline Mask`、`Adachi Inpaint Workbench` 或 `Adachi`。
+
+### 其他安装方式
+
+- **ComfyUI Manager：**可搜索 `Adachi-Inpaint-Workbench`；各版本的可用性取决于注册表验证结果，参见[注册表页面](https://registry.comfy.org/nodes/adachi-inpaint-workbench)。也可以使用 Manager 的 Git URL 安装方式填写本仓库地址。
+- **ZIP：**[下载仓库压缩包](https://github.com/adachi0130/Adachi-Inpaint-Workbench/archive/refs/heads/main.zip)，解压后放到 `ComfyUI/custom_nodes/Adachi-Inpaint-Workbench/`，保留完整的 `web/`、`backend.py` 和 `__init__.py`。
+- **Windows 安装脚本：**在解压后的插件目录运行以下命令，路径指向包含 `main.py` 的 ComfyUI 目录：
+
+```powershell
+powershell -ExecutionPolicy Bypass -File .\install.ps1 -ComfyUIPath "D:\ComfyUI_windows_portable\ComfyUI"
+```
+
+脚本会将已有安装备份到 `user/adachi-workbench-backups/`，不会自动重启 ComfyUI。
+
+Git 安装用户可在插件目录运行 `git pull` 更新，然后重启 ComfyUI 并刷新网页。`custom_nodes` 中只保留一份本插件，避免重复加载。
+
+## 快速开始
+
+1. 将 [Adachi-Workbench.json](examples/Adachi-Workbench.json) 拖入 ComfyUI，把 **YOUR_CHECKPOINT.safetensors** 换成自己兼容的 checkpoint；模板不下载模型，也不要求 LoRA。
+2. 在组忽略预设中点击 **Text to image**，修改中性的示例提示词，运行生图。
+3. 点击 **Inpaint**，在**内联蒙版**上传或拖入原图，涂抹需要修改的区域。**白色重绘，黑色保留**，橙色仅为显示叠加。
+4. 在**安达重绘工作台**设置重绘范围、宽高和重绘幅度，直接运行，或点击**打开安达工作台**进入独立界面编辑。
+5. 对比结果，满意后采用为新原图，保留蒙版继续修改。
+
+```mermaid
+flowchart LR
+    Source[内联蒙版] -->|image + mask| Inpaint[安达重绘工作台]
+    Inputs[模型 · VAE · 文本条件] --> Inpaint
+    Source -->|original| Compare[图像对比]
+    Inpaint -->|result| Compare
+    Inpaint --> Save[保存图像]
+    Compare -. 采用此结果 .-> Source
+```
+
+### 必要连线
+
+| 从哪里 | 连到哪里 |
+| --- | --- |
+| `内联蒙版.image` / `mask` | `安达重绘工作台.image` / `mask` |
+| 模型 / VAE / 正面条件 / 负面条件 | `安达重绘工作台`对应输入 |
+| `内联蒙版.image` | `图像对比.original` |
+| `安达重绘工作台.image` | `图像对比.result`，以及需要时的 `保存图像.images` |
+
+Refiner 输入是可选项。只有需要兼容的第二阶段模型时才连接，否则留空即可。
+
+## 节点一览
+
+分类为 **Adachi-Inpaint-Workbench**。组忽略预设是前端控制节点，其余六个为用户使用的后端节点；
+
+| 节点 | 用途 | 输出 |
+| --- | --- | --- |
+| **内联蒙版 · Inline Mask** | 普通蒙版或裁切重绘、双图层融合 | `image`、`mask`、`full_image` |
+| **安达重绘工作台 · Adachi Inpaint Workbench** | 一体化准备、采样、合成 | `image`：完整合成图；`crop`：采样画布；`mask`：采样蒙版 |
+| **图像对比 · Image Compare** | 预览、对比、采用结果 | `image`：原样传递结果批次 |
+| **组忽略预设 · Group Bypass Presets** | 应用已保存的组状态 | 无需连线 |
+| **自动二选一 · Auto Exclusive Switch** | 传递唯一启用的输入 | `output`：被选中分支的数据类型 |
+| **重绘准备 · Inpaint Prepare** | 为自定义采样流程裁剪、调整图像和蒙版 | `pixels`、`mask`、`context` |
+| **重绘合成 · Inpaint Composite** | 将局部解码结果回贴到原图 | `image` |
+
+模块化流程可使用 `Prepare.pixels → VAE 编码 → 设置潜空间噪声蒙版 → 采样器 → VAE 解码 → Composite`。`Prepare.mask` 连接噪声蒙版节点，**务必把 `Prepare.context` 连到 `Composite.context`**。
+
+## 画布快捷操作
+
+| 操作 | 快捷键或控件 |
+| --- | --- |
+| 画笔 / 橡皮擦 / 矩形 / 套索 | `B` / `E` / `R` / `L` |
+| 临时橡皮擦 | 按住 `Alt`，松开后恢复之前的工具 |
+| 调整笔径 | `Ctrl` + 滚轮，或 `[` / `]` |
+| 调整硬度 | `Shift` + 滚轮 |
+| 融合时临时看原图 | 在 **◐ 图层融合**中按住右键，松开恢复工作区 |
+| 取消裁切选定 | 在 **◌ 重绘蒙版**中右键或按 Escape，再左键选择新区域 |
+| 撤销 / 重做 | `Ctrl+Z` / `Ctrl+Shift+Z`，或 `Ctrl+Y` |
+| 未固定时的图片缩放 / 移动 | 滚轮 / 空格拖动、中键拖动，或移动工具 `H` |
+| 保持图片适应 | 点击「适应」旁的图钉，再点一次解除 |
+| 调整编辑区高度 | 拖动底部 `↕` 手柄 |
+
+画笔面板还提供反转、全选、清空、PNG 蒙版导入/导出、原图/蒙版/叠加视图和**原图嵌入工作流**。叠加不透明度只影响显示。鼠标在按钮、参数区滚动时控制外部工作流；从工作流背景开始的连续滚轮手势，移入图片后仍控制工作流。
+
+## 示例工作流
+
+只提供一个[官方模板：Adachi-Workbench.json](examples/Adachi-Workbench.json)，包含生图与修图两条分支、共用模型、组忽略预设、条件／图像自动二选一、图像对比和保存。预设控制器位于受控组外；每次选择一个预设，同时打开两路会触发二选一的检查。
+
+模板标题使用英文，提示词为重新编写的中性示例，模型为占位名称，需自行选择。修图前请先上传原图；裁切融合直接切换同一个内联蒙版即可。使用分离模型、文本编码器和 VAE 的模型家族时，请用匹配的加载节点替换 checkpoint 区块并连接对应输出。
+
+## 兼容性与重要行为
+
+- **静态图片：**支持常规四维 `[B,C,H,W]` 和单帧五维 `[B,C,1,H,W]` 图像潜空间，包括 Anima 使用的 Wan 衍生 VAE 路径；需提供该模型配套的 VAE 和文本条件，不是多帧视频重绘。
+- **Anima：**普通带蒙版图生图请关闭 **inpaint_model / 专用重绘模型**。该选项用于专用 inpaint 架构，并非画了蒙版就需要开启。
+- **原图尺寸：**两种采样区域最终都按原图尺寸合成；如果要目标采样尺寸的图片，使用 `crop` 或另接调整尺寸节点。
+- **裁切图层：**选区和融合蒙版会保存到工作流，最新重绘图层引用 `input/adachi-inpaint-layers/` 中的图片。跨机器移动尚未合并的编辑时需带上这张图，或先合并图层再嵌入原图。
+- **分享工作流：**蒙版会写入 JSON。原图默认只存 `input/` 下的相对路径，开启**原图嵌入工作流**才会把图片也写进去，文件会变大。加载工作流、上传图片、切换模式或勾选嵌入时，如果浏览器草稿空间不足，插件先保存一份完全相同的原图到 input 图片目录，再改用文件引用，保留蒙版、图层和笔画记录。若确实要把大图嵌入 JSON，可先关闭 ComfyUI 的浏览器工作流持久化，再勾选嵌入；导出文件不受浏览器草稿限额影响。若当前页面已经每次操作都提示草稿失败，请先按 **Ctrl+S** 保存，再按 **Ctrl+F5** 刷新一次，重置 ComfyUI 本页已停用的草稿状态。
+- **采用结果：**结果尺寸必须与原图一致，才能保留蒙版坐标。撤销/重做记录在本次会话中保留；保存重开保留蒙版，不保存逐笔命令历史。
+- **对比预览：**新的原图和重绘结果保存在服务器内存中，不再生成对比临时图片。默认缓存上限为 512 MiB，同时最多保留 512 批对比，可在启动 ComfyUI 前设置 `ADACHI_COMPARE_MEMORY_MIB` 为 64～4096 调整字节上限。内存淘汰或服务器重启后，旧对比和下载引用会失效，重新运行工作流即可恢复；单批超过上限会提示错误。旧版磁盘预览文件存在时仍可读取，其过期清理也继续保留。采用到编辑器的图片另存于 `input/adachi-inpaint/` 或 `input/adachi-inpaint-layers/`。
+- **功能范围：**这是独立实现的 WebUI 风格工具，不保证与 WebUI 像素、脚本和扩展行为完全一致；额外软边羽化并非 WebUI 的完整 Soft Inpainting 算法。
+
+<details>
+<summary>采样、填充与批次细节</summary>
+
+- 采样宽高使用 8 的倍数。padding、模糊和蒙版扩张以原图像素计量；模糊/扩张会改变有效蒙版边界，有效蒙版为零的像素保持不变。
+- `mask_mode=outside` 会先反转涂抹区域，再进行蒙版处理。
+- `resize_mode` 控制整图采样的调整方式；仅蒙版区域使用保留完整蒙版的比例变换，也适用于蒙版碰到图像边缘的情况。
+- 填充模式包括**原图像素、周边颜色填充、潜空间噪声、空白潜空间**。潜空间模式属于一体化重绘节点；模块化 Prepare 只处理图像和蒙版。
+- 空蒙版或 `denoise=0` 会跳过 VAE 和采样。`batch_size` 表示每张原图生成的变体数，不是 ComfyUI 队列次数。
+- 模块化多原图批次需按原图分组变体，例如 `A,A,B,B`，而非交错的 `A,B,A,B`。
+- 可选 Refiner 使用 `refiner_model` 与 `refiner_switch`；`0.8` 表示最后 20% 步骤使用 Refiner，建议提供兼容的专用文本条件。设为 `1` 只用基础模型，`0` 只用 Refiner。
+- 输入限制为单文件 64 MiB、单图 3200 万像素、单边 16384 像素。大图绘制与 PNG 编码速度仍取决于浏览器内存和硬件。
+
+</details>
+
+## 排错与开发
+
+**更新后还是旧界面？** 完整重启 ComfyUI，再强制刷新网页。Python 后端修改不能仅靠刷新网页加载。
+
+**自动二选一提示没有有效输入？** 检查真正的上游源节点和对应 Set，不能只看 Get。将二选一放在受控分组之外。若其他集成仍会丢掉子图输出，可以先把 CLIP 编码器放到子图外。反馈时请附上工作流、插件/前端版本和完整错误；当前错误会显示前端、工作流与执行输入的状态标记。
+
+**打开别人的工作流后图片丢失？** 按相对路径补齐原图，或请对方开启原图嵌入后再保存。临时对比预览失效则需要重新生成。
+
+运行 CPU 回归测试：
+
+```bash
+python -m unittest discover -s tests -v
+node --experimental-default-type=module tests/test_exclusive_routes.mjs
+node --experimental-default-type=module tests/test_compare_history.mjs
+```
+
+已测试配置和范围见[验证记录](VERIFICATION.md)，版本变化见[更新日志](CHANGELOG.md)，恢复旧版本见[回滚说明](ROLLBACK.md)。模型家族和实际画质测试并非全覆盖。
+
+## 致谢与许可证
+
+由 **Adachi / 安达** 制作，采用 [MIT 许可证](LICENSE)。
+
+基于 [ComfyUI](https://github.com/Comfy-Org/ComfyUI) 构建，重绘交互参考 WebUI，鼠标划线对比体验参考 [rgthree-comfy](https://github.com/rgthree/rgthree-comfy)。代码由本仓库独立实现，这些项目不作为捆绑依赖。
+
+欢迎通过 [Issues](https://github.com/adachi0130/Adachi-Inpaint-Workbench/issues) 反馈问题、分享工作流用法。

ROLLBACK.md

@@ -1,9 +1,36 @@
 # Git versions and rollback
 
-The current local version is **v0.5.13**. Both the installed plugin and the source repository retain local commits and tags. This update is also pushed to the GitHub main branch.
+The stability fixes in 2.0.1 start from `local-before-stability-20261110` (the published `v2.0.0` commit). A complete copied-installation ZIP is stored in `ComfyUI/user/adachi-workbench-backups/` with prefix `Adachi-Inpaint-Workbench-before-stability-20261110-`. Restore that backup or switch a clean source checkout to the baseline tag, then reinstall, restart ComfyUI and hard-refresh. The published 2.0.0 archive remains unchanged.
+
+The current version is **v2.0.1**. Git history lives in the source repository. The installed custom-node folder may be a ZIP/file copy without its own `.git`; do not run rollback commands in ComfyUI's parent repository.
+
+Before installing, the installer saves the previous plugin directory under `ComfyUI/user/adachi-workbench-backups/`. Restore that ZIP to roll back a copied installation. For a Git clone, use the tags below.
 
 | Tag | Contents |
 |---|---|
+| `v2.0.1` | Stability, cancellation, recovery and dense-mask memory fixes |
+| `v2.0.0` | Standalone workbench, workflow discovery, cloud recovery, localization and unified template |
+| `v1.0.1` | Previous published version with comparison retry |
+| `v1.0.0` | First stable release with current bilingual crop instructions |
+| `local-before-v1.0.0` / `v0.9.0` | Source before stable-release documentation and metadata |
+| `v0.9.0` | In-memory comparison previews and real upload progress |
+| `local-before-v0.9.0` / `v0.8.2` | Source before memory previews and upload progress |
+| `v0.8.2` | Refresh different-sized uploads and synchronize inpaint dimensions |
+| `local-before-v0.8.2` / `v0.8.1` | Source before upload and dimension synchronization fixes |
+| `v0.8.1` | Expire old comparison previews and download the latest result |
+| `local-before-v0.8.1` / `v0.8.0` | Source before comparison-preview cleanup and latest-result download |
+| `v0.6.7` | Hold right mouse to preview the original before merging; contextual hover help |
+| `local-before-v0.6.7` | Source before original-layer preview and expanded hover instructions |
+| `v0.6.6` | Right-click releases crops; move mask Undo beside Merge |
+| `local-before-v0.6.6` | Source before right-click reselection and inline Undo |
+| `v0.6.5` | Change crop dimensions while selected, automatically return to selection |
+| `local-before-v0.6.5` | Source before editable locked-crop dimensions |
+| `v0.6.4` | Black crop frame, automatic dimensions and comparison-only result adoption |
+| `local-before-v0.6.4` | Source before the comparison-adoption update, including v0.6.3 draft preflight fixes |
+| `v0.6.2` | Draft recovery, collapsible controls, retired selection tools and revised crop behavior |
+| `local-before-v0.6.2` | Exact baseline before the combined draft/fold and crop updates |
+| `v0.6.0` | Merged crop/blend editor, magic wand and quick selection |
+| `local-before-v0.6.0` | Exact source baseline before the merged editor (includes bilingual README) |
 | `v0.5.13` | Restore dropped enabled inputs through saved subgraph output routes |
 | `v0.5.12` / `local-before-v0.5.13` | Trace Get/Set source states server-side before prompt validation and caching |
 | `v0.5.11` / `local-before-v0.5.12` | Resolve exclusive selection from actual submitted connections despite stale frontend flags |
@@ -25,14 +52,23 @@ To undo only this update, run in the plugin folder:
 
 ```powershell
 git status --short
-git switch --detach local-before-v0.5.13
+git switch --detach v2.0.0
 ```
 
 Return to this version with:
 
 ```powershell
-git switch --detach v0.5.13
+git switch --detach v2.0.1
 ```
 
 Restart ComfyUI and hard-refresh the browser (`Ctrl+F5`) after switching versions.
-Commit or back up any local modifications before switching. A release ZIP does not contain `.git`; local history is kept in the source and installation repositories.
+Commit or back up any local modifications before switching. A release ZIP does not contain `.git`; use the installer backup for a copied installation.
+
+
+Local workbench UI / network preview (2026-11-09) baseline: `local-before-workspace-ui-network-20261109` (`392149f`). To undo this local work on a clean checkout, use `git switch --detach local-before-workspace-ui-network-20261109`, reinstall those files, restart ComfyUI and hard-refresh. The installed directory also has a full ZIP backup in `user/adachi-workbench-backups/` whose name starts with `Adachi-Inpaint-Workbench-before-workspace-ui-network-20261109-`.
+
+
+The editable-generation / execution-stage update starts at `local-before-workspace-run-progress-20261109` (`8ec5281`). A copied-installation ZIP backup named `Adachi-Inpaint-Workbench-before-run-progress-20261109-*.zip` is in the same backup directory. This update changes frontend files only; hard-refresh after restoring them.
+
+
+Generic routing / resizable workspace baseline: `local-before-generic-workbench-20261109` (`02719c3`). The copied-installation backup starts with `Adachi-Inpaint-Workbench-before-generic-layout-20261109-` in `user/adachi-workbench-backups/`. This update changes frontend files; hard-refresh after restoring them.

VERIFICATION.md

@@ -1,5 +1,51 @@
 # Verification record
 
+## 2.0.1 · stability audit (2026-11-10)
+
+- 68 Python regression tests and all 20 JavaScript test files cover new cancellation, stalled decoding/upload, corrupt-cache retry, dense/sparse/edge masks, bounded cache metadata, partial rollback failures, concurrent completed edits and malformed input mappings.
+- Real ComfyUI browser checks pass on frontends 1.42.11 and 1.52.7 in classic and Nodes 2.0 rendering. Each configuration performs 20 mount/close cycles with zero remaining workspace API listeners, delayed initialization disposal, a delayed submit response across close/reopen, and recovery with execution notifications suppressed.
+- Actual CPU image queues cover normal adoption/undo and crop-origin validation, crop/mask retention, layer adoption, automatic return to editing, protected exterior pixels and blend clear/undo. Invalid source replacement retains the previous usable image. Model sampling algorithms are unchanged; no new model-quality claim is based on these CPU queues.
+- Browser resource checks recover a corrupt PNG, reload 25 times and overlap 30 cancellation attempts. At settled checkpoints exactly two preview object URLs remain, destruction releases all tracked URLs, and no unhandled promise rejection is observed. Network fault tests exhaust immediate retries, then recover on reconnect with byte progress and no duplicate history entries.
+- Isolated CPU benchmark: a full 2048×4096 mask, four threads, four iterations per process averages 203.4 ms before and 14.6 ms after. Additional measured process peak working set is 521.0 MiB before versus 9.2 MiB after. This measures mask-bound computation only, not overall generation time or workflow frame rate.
+- Scope: local fault injection is not a long-duration cloud soak test, and cannot establish compatibility with every third-party plugin/model combination. The public 2.0.0 Registry version remains flagged pending external review; this local audit does not establish Registry approval.
+
+## 2.0.0 · 2026-11-09
+
+- All 63 Python regression tests and 19 JavaScript test files pass. Frontend modules pass syntax checks; locale JSON, release metadata and local links in both READMEs validate.
+- The single official template imports into a real ComfyUI frontend. Both presets compile through the native graph serializer and pass backend validation with no missing links. Text-to-image excludes the uploaded-image branch; inpainting excludes the latent-generation branch. The three exclusive selectors consistently select the only enabled input.
+- Both template presets complete actual 512×512, two-step WAI generation without LoRA, using newly authored neutral prompts and a synthetic input. Inpainting changes the nonempty masked region while pixels outside the protected region remain exact. The distributed template keeps its checkpoint placeholder and contains no model names, local paths, source images or personal prompts.
+- Real frontend settings expose the locale at `comfyAPI.app.app.ui.settings`. The translation reader supports that path, older frontend shapes and the English fallback; tests cover interpolation, context keys and current execution messages.
+- GitHub publication, Registry upload and Registry approval are checked separately. See the Registry review notes for public scan findings; local testing does not constitute Registry security approval.
+
+- Local white workspace preview: real `gpt test` WAI v160 generation with no LoRA, 512×512, two sampler steps and a nonempty mask succeeds. Protected pixels remain exact. The UI run is also verified: immediate opening, background backend validation, invalid-model rejection without queueing, result preview without adopting, before/after, explicit adoption, mask-preserving rollback and source selection. Classic and Nodes 2.0 browser tests cover actual mouse painting, comparison, clean close and reopen; 768/390 layouts have no horizontal overflow. Source history tests include crop/layer/fusion-mask restoration. No remote cloud or Anima model inference is claimed by these checks.
+
+- 1.0.0: all 54 CPU backend tests and all seven JavaScript regression suites pass; every frontend module passes syntax checking and all bundled examples parse as JSON. The release updates documentation and metadata on top of 0.9.0; no full model generation was run for this release.
+
+- 0.9.0: 54 backend CPU tests cover in-memory PNG pixels, atomic batch storage, LRU eviction, cached execution invalidation and oversized batch rejection. Actual CPU image queues in frontend 1.52.7 and both node renderers verify comparison history, download bytes, ordinary and crop adoption, mask retention, layer blending, workflow reload and legacy disk-backed references; no new comparison temp files are created. Real Chrome network throttling confirms intermediate upload percentages, loading indication, user header, failure and concurrent upload behavior in both renderers. Proxy paths and auth fallback were separately tested; the user's actual cloud environment and full model inference were not tested.
+
+- 0.8.2: Chrome interaction with an isolated ComfyUI and frontend 1.52.7 passes in classic nodes and Nodes 2.0. Actual uploads confirm different-size source pixels and canvas display update, same-size masks are preserved, connected inpaint dimensions follow ordinary uploads and crop selection, switching back restores image dimensions, a newly connected inpaint node receives the size, and reopening a saved workflow preserves manually saved dimensions. No model inference was run.
+
+- 0.8.1: 51 backend CPU tests pass, including expiry, preservation of unrelated/active/input files and comparison cache renewal. On an isolated local ComfyUI with frontend 1.52.7, actual lightweight image queues in classic nodes and Nodes 2.0 verify two comparison results, selection of an older result, downloading the newest PNG byte-for-byte, and cached notification deduplication. Browser checks also cover the Chinese tooltip, absent result and an expired-preview error. No full model generation was performed.
+
+- 0.6.7: real Chrome mouse interactions on frontend 1.52.7 pass in classic nodes and Nodes 2.0. Display pixel checks verify hold-right original preview and release restoration, including release outside the picture, pointer cancellation and window blur. Preview overrides mask-only display temporarily, leaves serialized image/mask/crop state unchanged, and does not affect composite export pixels. After Merge, right-click still releases the selection; normal-mode right-click does not erase. Bottom controls have hover instructions and canvas help follows the editing stage. These checks use solid-color image fixtures, not model generation.
+
+- 0.6.6: real mouse checks in classic nodes and Nodes 2.0 on frontend 1.52.7 verify right-click release without erasing, right-click while unselected, left-click relocking, preserved adopted pixels after merge, Alt temporary erase in both modes, and a functional live Undo button directly after Merge. The old reselect button and inpaint-panel Undo are absent. Canvas context menus are stopped from bubbling to the outer node. Frontend syntax and diff whitespace checks pass.
+
+- 0.6.5: classic and Nodes 2.0 browser checks on frontend 1.52.7 verify changing a locked preset, pointer-following replacement selection, automatic inpaint dimensions after relocking, preservation of an adopted patch by merging, retention of the old selection/resolution after an upload failure, and reselection through swap/custom controls. Frontend syntax and diff whitespace checks pass.
+
+- 0.6.4 comparison-adoption update: 48 backend tests pass, including removal of the result socket, ignoring obsolete feedback links, and generating-region metadata on comparison previews. Resolution-sync checks cover multiple connected targets, linked-input protection and source-size bounds. Real CPU comparison queues and browser interactions on isolated port 8196 with frontend 1.52.7 pass in classic nodes and Nodes 2.0: black dashed frame, automatic inpaint dimensions, no automatic stitching, manual crop adoption with retained repaint mask, independent fusion editing, reload, stale-result rejection, normal-mode mask preservation, and migration of the old wired example. Existing draft preflight/storage, comparison history, geometry and frontend syntax checks also pass. No complete model generation is claimed by these lightweight image probes.
+
+- 0.6.3: reproduced the sticky native draft failure on real frontend builds 1.42.11 and 1.52.7: after a 12 Mi-character draft fails, even a two-character draft returns false until the page reloads. Isolated lightweight API fixtures (no model queues or production writes) verify oversized workflow loading before configuration, exact source byte hashes, repaint mask pixels, stale named-snapshot cleanup, repeated mode changes and subsequent successful native draft writes in classic nodes and Nodes 2.0 on both builds. Unit checks additionally cover nested subgraph sources, named-only snapshots, fitting/blocked storage, failed uploads and unchanged masks/layers. This does not claim inspection of the user's live browser console or a complete model generation.
+
+- 0.6.1 fixes (included in 0.6.2): reproduced the exact browser “failed to save workflow draft” toast on frontend 1.42.11 with an approximately 10-million-character embedded source and a real QuotaExceededError. The capacity guard now converts only a source that does not fit, uploading it before clearing embedded data. SHA-256 checks confirm the input PNG bytes are identical; repaint/fusion masks, layers and undo commands survive. Failed uploads preserve both embedding and mode, existing browser storage is untouched, fitting sources remain embedded and disabled workflow persistence is respected.
+- Combined 0.6.2: the quota/fold browser regression passes in classic nodes and Nodes 2.0 on frontends 1.42.11 and 1.52.7 (1.52.7 also tested after the concurrent crop/tool changes). The collapsed tools hide painting controls and mask/fill options while all seven action buttons remain visible and operational. Heights update, open/closed states survive workflow reload, and 1440/768/390 layouts pass. Real CPU feedback queues use image inversion rather than model inference; no complete Anima generation is claimed. All 47 backend tests and JavaScript tests for draft guarding, crop geometry, exclusive routing and comparison history pass.
+
+- 0.6.2: all 47 backend tests, crop geometry checks and existing draft-guard checks pass; all frontend modules pass syntax checks. Real browser checks on an isolated 8194 CPU service with frontend 1.52.7 pass in classic nodes and Nodes 2.0: locked crops start empty, manual strokes paint normally, fill/rectangle/brush stay within the crop, undo/redo and reload preserve the mask, a fresh region clears previous strokes, ordinary mode remains unrestricted, and the toolbar contains no automatic-selection tools. Six presets include both 512 and 2048 endpoints and use only 64-aligned dimensions.
+
+- 0.6.0: 47 CPU backend tests pass, including exact crop/mask slicing, soft fusion masks, batch composites, untouched outside pixels, content-addressed layer files, invalid bounds/result sizes, normal-mode ignored results and acyclic feedback normalization. Geometry, magic-wand/quick-selection, exclusive routing and comparison-history JavaScript checks also pass. All frontend modules pass syntax checks; example/localization JSON and project metadata parse successfully.
+- 0.6.0: real Chrome interaction and real CPU queues on isolated port 8191 with ComfyUI 0.19.3 and frontend 1.52.7 pass in classic nodes and Nodes 2.0. Tests cover a visible Inline Mask → ImageInvert → Inline Mask loop with full-image saving, normal-mode result ignoring, magic-wand add/subtract, quick-selection drag/subtract, independent repaint/fusion masks, soft-brush undo/redo, fixed-region repeat generation, saved-workflow restoration, merge/reselect, stale-job rejection, Ctrl/Shift wheel gestures, pinned graph scrolling and 1440/768/390 viewport layouts. No plugin script errors were observed. These are image-processing probes, not a complete Anima generation.
+- 0.6.0: the connected WebUI panel mounts the existing live painting controls, source-size filling uses locked crop dimensions, Escape closes the selection menu without reselecting, and the internal layer sink is hidden from the classic node menu. The old example loads with existing image/mask wires intact; the new eight-node crop/blend example is generated and reloaded through ComfyUI.
+
 - 0.5.13: the remote history was read without modifying its workflow or queue. Three actual failed request snapshots contained selector 532 with a selected flag but no data input; local normalization restores its saved tag-subgraph connection to `349:412`, leaving selector 533's `16` source intact. Two real local CPU queue probes reproduce the missing-input shape and subgraph mapping with lightweight image sources, verify both selector outputs and upstream changes, and avoid model generation. All 39 backend unit tests pass, including Get/Set recovery, flattened subgraph output slots and refusal to reconnect absent or mismatched source types. The remote's complete Anima generation was not executed by these tests.
 
 - 0.5.12: all 37 CPU unit tests pass. The supplied remote workflow was used as request metadata with lightweight red/blue image probes replacing model generation: positive selector 532 chooses B, negative 533 chooses A, all four Gets remain enabled. Ten real queue cases cover reverse Set/source states, switching back, encoder-only and Set-only bypass, unchanged-workflow preservation, cache reuse, and both-enabled/both-ignored errors. No complete Anima model generation is claimed by these routing probes. Sixteen native KJ Get/Set browser queue cases pass on frontend 1.52.7 in both renderers, and all 15 metadata-free stale-state API queue cases still pass.
@@ -70,3 +116,33 @@ References used during implementation:
 - [rgthree image comparison interaction reference](https://github.com/rgthree/rgthree-comfy/blob/main/py/image_comparer.py)
 
 The plugin code is an independent implementation; see LICENSE for the release licence. No source from the projects above is bundled.
+
+
+## 2026-11-09 · local workspace UI and cloud recovery
+
+- 58 Python tests and all JavaScript test files pass, including new protocol, conditioning-route, preview stream and result-recovery cases.
+- Real `gpt test` WAI profile without LoRA validates against ComfyUI: positive #521, negative #520, model #518. The original graph modes and links are restored after compilation. Missing models are blocked before queueing.
+- Real CPU image processing verifies manual adoption, unchanged mask pixels, undo/original restoration and result-count reset. Generation defaults to sliding comparison without adopting.
+- Chrome interaction checks verify remembered light/dark themes, icon tooltips, one before/after toggle, enabled/missing controllers and dynamically added presets.
+- Classic and Nodes 2.0 crop controls pass at 1440, 768 and 390px in both themes, including wrapped history labels, custom dimensions, painting and dimension synchronization.
+- Cloud failure simulation drops `executed` alone, then all execution notifications. Exact-task history recovery succeeds in both the node and workspace. Three image-load attempts are exhausted before restoring connectivity; byte progress is visible and preview recovery succeeds. Replayed output does not increase the result count.
+- Existing source workflows and images are not modified. Runtime files are compared with the preceding Git checkpoint before installation; the installed package is backed up before copying changes.
+
+
+## 2026-11-09 · editable generation and execution stages
+
+- An actual WAI 512px / six-step generation reports node stages and sampler fractions. A real mouse stroke was added during sampling; the queued mask remained byte-identical to its submitted snapshot, and both the source and new mask edits survived result delivery.
+- Crop adoption was checked independently with a real processed image and crop-origin metadata. It returns to the editor with the locked region, result layer and repaint mask intact.
+- Progress rendering fits above workflow settings at 1440, 768 and 390px in both themes. Unit checks cover prompt isolation, early events before the HTTP response, cached nodes, reconnects and late progress packets after completion.
+- Both local Anima profiles compile their subgraph correctly and fail validation only for the three absent model components. The installed backend and model inventory confirm the missing checkpoint, text encoder and VAE. Model UI now lists filenames and target folders; missing-component checks never infer absent files from a network failure.
+
+- Legacy frontend scalar/null `executing` messages and a submit HTTP response deliberately delayed until after task completion are covered, so early completion cannot cause a false error or leave the workbench busy.
+
+
+## 2026-11-09 · generic workflow routing and resizable workspace
+
+- Real divider drag, keyboard resize, persisted width and reset pass; 768px/390px layouts keep stacked panels and the English brand without overflow.
+- All three existing example presets compile through Use Everywhere’s public native serialization API. MODEL and positive/negative CLIP branches retain their original LoRA routes in inherited mode. Bypassed routing is enabled only for the selected bypassed workbench group, and all editable modes are restored. Unique exclusive switches are resolved in the separate execution graph before backend metadata normalization.
+- A copy with every root node ID changed and all preset titles replaced with unrelated names still discovers and validates the correct components.
+- A newly built two-pipeline workflow validates with SD1.5, selects the connected editor/result instead of an unrelated first loader, and preserves ModelSamplingDiscrete, CLIPSetLastLayer, two-encoder conditioning, negative conditioning processing, image resizing and result postprocessing. A real WAI generation in that workflow confirms the final ImageInvert is executed by checking outside-mask pixels.
+- A custom execution-only LoRA validates without any editable LoRA nodes. Pure tests cover independent IDs/filenames, ControlNet positive/negative slots, SDXL text fields, inherited/custom split CLIP paths, untouched unrelated branches, Vue proxy JSON copying, explicit ambiguity, group activation safety and subgraph result recovery.

__init__.py

@@ -1,12 +1,33 @@
 """Adachi-Inpaint-Workbench custom-node package for ComfyUI."""
 
 from .backend import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS, normalize_exclusive_prompt
+from .preview_cleanup import start_preview_cleanup
+from .preview_store import PREVIEWS
+from .workspace_execution import register_workspace_routes
+from aiohttp import web
 from server import PromptServer
+import re
 
 if not getattr(PromptServer.instance, "_adachi_exclusive_normalizer", False):
     PromptServer.instance.add_on_prompt_handler(normalize_exclusive_prompt)
     PromptServer.instance._adachi_exclusive_normalizer = True
 
+if not getattr(PromptServer.instance, "_adachi_preview_route", False):
+    @PromptServer.instance.routes.get("/adachi-inpaint/compare/{filename}")
+    async def adachi_comparison_preview(request):
+        name = request.match_info["filename"]
+        if not re.fullmatch(r"[A-Za-z0-9_-]{32}\.png", name):
+            raise web.HTTPNotFound()
+        data = PREVIEWS.get(name[:-4])
+        if data is None:
+            raise web.HTTPNotFound(text="Comparison preview expired; run the workflow again.")
+        return web.Response(body=data, content_type="image/png", headers={"Cache-Control": "no-store"})
+
+    PromptServer.instance._adachi_preview_route = True
+
+start_preview_cleanup()
+register_workspace_routes(PromptServer.instance)
+
 WEB_DIRECTORY = "./web"
 
 __all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"]

backend.py

@@ -10,9 +10,11 @@ import base64
 import binascii
 import hashlib
 import io
+import json
 import math
 from pathlib import Path
 import re
+import tempfile
 import warnings
 
 import numpy as np
@@ -20,6 +22,11 @@ from PIL import Image, ImageFilter, ImageOps, UnidentifiedImageError
 import torch
 import torch.nn.functional as F
 
+if __package__:
+    from .preview_store import PREVIEWS
+else:  # Standalone CPU tests load this module directly.
+    from preview_store import PREVIEWS
+
 
 CATEGORY = "Adachi-Inpaint-Workbench"
 MAX_IMAGE_PIXELS = 32_000_000
@@ -93,7 +100,7 @@ def _image_tensor(image):
     value = torch.as_tensor(image)
     if value.ndim != 4 or value.shape[-1] not in (3, 4) or min(value.shape[:3]) < 1:
         raise ValueError("IMAGE must have shape [batch, height, width, 3 or 4].")
-    return torch.nan_to_num(value[..., :3].float(), nan=0.0, posinf=1.0, neginf=0.0).clamp(0.0, 1.0)
+    return torch.nan_to_num(value[..., :3].float(), nan=0.0, posinf=1.0, neginf=0.0).clamp_(0.0, 1.0)
 
 
 def _mask_tensor(mask, images):
@@ -110,7 +117,7 @@ def _mask_tensor(mask, images):
         value = value.expand(images.shape[0], -1, -1)
     if value.shape[0] != images.shape[0]:
         raise ValueError("Mask batch must contain one mask or match the image batch.")
-    value = torch.nan_to_num(value, nan=0.0, posinf=1.0, neginf=0.0).clamp(0, 1)
+    value = torch.nan_to_num(value, nan=0.0, posinf=1.0, neginf=0.0).clamp_(0, 1)
     if value.shape[-2:] != images.shape[1:3]:
         value = F.interpolate(value[:, None], size=images.shape[1:3], mode="bilinear", align_corners=False)[:, 0]
     return value
@@ -136,11 +143,15 @@ def _feather(mask, grow, blur):
 
 def _bbox(mask, padding, target_ratio):
     height, width = mask.shape
-    nonzero = torch.nonzero(mask > 0, as_tuple=False)
-    if not len(nonzero):
+    occupied = mask > 0
+    rows = torch.where(occupied.any(dim=1))[0]
+    if not len(rows):
         return (0, 0, width, height)
-    y0, x0 = nonzero.amin(0).tolist()
-    y1, x1 = (nonzero.amax(0) + 1).tolist()
+    # Dense masks previously allocated two int64 coordinates per painted pixel
+    # (512 MB for a 32 MP mask). Only row/column occupancy is needed for bounds.
+    columns = torch.where(occupied.any(dim=0))[0]
+    y0, y1 = int(rows[0]), int(rows[-1]) + 1
+    x0, x1 = int(columns[0]), int(columns[-1]) + 1
     x0, y0 = max(0, x0 - padding), max(0, y0 - padding)
     x1, y1 = min(width, x1 + padding), min(height, y1 + padding)
     crop_width, crop_height = x1 - x0, y1 - y0
@@ -310,31 +321,39 @@ class AdachiInlineMask:
             "image": ("STRING", {"default": "", "tooltip": "Input image path populated by inline upload."}),
             "mask_data": ("STRING", {"default": "", "multiline": False, "tooltip": "Editor mask PNG data URL; white repaints."}),
             "source_data": ("STRING", {"default": "", "multiline": False, "tooltip": "Embedded source image for portable workflows."}),
+        }, "optional": {
+            "edit_mode": (["mask", "crop"], {"default": "mask"}),
+            "crop_state": ("STRING", {"default": "{}"}),
+            "blend_data": ("STRING", {"default": ""}),
+            "layer_image": ("STRING", {"default": ""}),
         }}
 
-    RETURN_TYPES = ("IMAGE", "MASK")
-    RETURN_NAMES = ("image", "mask")
+    RETURN_TYPES = ("IMAGE", "MASK", "IMAGE")
+    RETURN_NAMES = ("image", "mask", "full_image")
     FUNCTION = "load"
     CATEGORY = CATEGORY
     DESCRIPTION = "Paint an inpaint mask directly in the workflow."
     SEARCH_ALIASES = ["inline mask", "webui inpaint", "局部重绘", "安达", "adachi"]
 
     @classmethod
-    def IS_CHANGED(cls, image="", mask_data="", source_data=""):
+    def IS_CHANGED(cls, image="", mask_data="", source_data="", **kwargs):
         digest = hashlib.sha256()
-        for value in (image, mask_data, source_data):
+        for value in (image, mask_data, source_data, json.dumps(kwargs, sort_keys=True, default=str)):
             digest.update(str(value).encode("utf-8"))
             digest.update(b"\x00")
         if not source_data:
+            # The source is a stable input file; hashing its size and mtime is
+            # enough to detect a replacement and avoids reading a full high-res
+            # PNG on every change/execution. Content-hash the bytes only when
+            # the metadata cannot be read (the loader will then raise anyway).
             try:
-                with _input_path(image).open("rb") as stream:
-                    for chunk in iter(lambda: stream.read(1024 * 1024), b""):
-                        digest.update(chunk)
+                stat = _input_path(image).stat()
+                digest.update(f"{stat.st_size}:{stat.st_mtime_ns}".encode("utf-8"))
             except (ValueError, OSError):
                 return float("nan")
         return digest.hexdigest()
 
-    def load(self, image="", mask_data="", source_data=""):
+    def load(self, image="", mask_data="", source_data="", edit_mode="mask", crop_state="{}", blend_data="", layer_image="", result=None):
         raster = _data_image(source_data) if source_data else _read_raster(_input_path(image))
         rgba = raster.convert("RGBA")
         pixels = torch.from_numpy(np.asarray(rgba.convert("RGB"), dtype=np.float32).copy() / 255.0)[None]
@@ -345,7 +364,14 @@ class AdachiInlineMask:
             mask = np.asarray(mask_image, dtype=np.float32).copy() / 255.0
         else:
             mask = 1.0 - np.asarray(rgba.getchannel("A"), dtype=np.float32).copy() / 255.0
-        return pixels, torch.from_numpy(mask)[None]
+        mask = torch.from_numpy(mask)[None]
+        if edit_mode == "crop":
+            state = _crop_state(crop_state)
+            if state.get("locked"):
+                x, y, width, height = _crop_box(state, pixels.shape[2], pixels.shape[1])
+                composite = _blend_layer(pixels, _load_layer(layer_image), blend_data, state)
+                return composite[:, y:y+height, x:x+width].clone(), mask[:, y:y+height, x:x+width].clone(), composite
+        return pixels, mask, pixels
 
 
 class AdachiInpaintPrepare:
@@ -505,7 +531,7 @@ class AdachiWebUIInpaint:
     FUNCTION = "sample"
     CATEGORY = CATEGORY
     DESCRIPTION = "WebUI-style inpainting, crop, latent fill, sampling and original-size compositing."
-    SEARCH_ALIASES = ["webui", "img2img", "inpaint", "局部重绘", "安达", "adachi"]
+    SEARCH_ALIASES = ["webui", "img2img", "inpaint", "局部重绘", "安达重绘工作台", "安达", "adachi"]
 
     def sample(self, model, positive, negative, vae, image, mask, seed=0, steps=20, cfg=7.0,
                sampler_name="euler", scheduler="normal", denoise=0.75, batch_size=1, inpaint_model=False,
@@ -589,7 +615,7 @@ class AdachiImageCompare:
         return {"required": {
             "original": ("IMAGE", {"tooltip": "Original image shown on the right."}),
             "result": ("IMAGE", {"tooltip": "Full-size generated image shown on the left; use image, not crop."}),
-        }}
+        }, "hidden": {"prompt": "PROMPT", "unique_id": "UNIQUE_ID", "dynprompt": "DYNPROMPT"}}
 
     RETURN_TYPES = ("IMAGE",)
     RETURN_NAMES = ("image",)
@@ -599,9 +625,13 @@ class AdachiImageCompare:
     DESCRIPTION = "Hover to compare; adopt a result into the upstream inline editor."
     SEARCH_ALIASES = ["compare", "before after", "图像对比", "adachi", "安达工作台"]
 
-    def compare(self, original, result):
-        import nodes
+    @classmethod
+    def IS_CHANGED(cls, original, result, **kwargs):
+        # All evictions invalidate old cached UI descriptors; the store also
+        # identifies its process so a restarted server cannot replay stale URLs.
+        return PREVIEWS.fingerprint()
 
+    def compare(self, original, result, prompt=None, unique_id=None, dynprompt=None):
         original, result = _image_tensor(original), _image_tensor(result)
         if original.shape[1:3] != result.shape[1:3]:
             raise ValueError("Image sizes must match. Connect the final image output, not crop.")
@@ -609,10 +639,21 @@ class AdachiImageCompare:
             raise ValueError("Result batch must be a multiple of the original batch.")
         if original.shape[1] * original.shape[2] > MAX_IMAGE_PIXELS or max(original.shape[1:3]) > 16384:
             raise ValueError("Comparison images exceed the inline editor's 32 megapixel / 16384px limit.")
-        preview = nodes.PreviewImage()
-        before = preview.save_images(original, filename_prefix="adachi_compare_original")["ui"]["images"]
-        after = preview.save_images(result, filename_prefix="adachi_compare_result")["ui"]["images"]
-        return {"ui": {"adachi_originals": before, "adachi_results": after}, "result": (result,)}
+        images, encoded_size = [], 0
+        for batch in (original, result):
+            for pixels in batch:
+                rgb = np.clip(255.0 * pixels.cpu().numpy(), 0, 255).astype(np.uint8)
+                stream = io.BytesIO()
+                Image.fromarray(rgb).save(stream, format="PNG", compress_level=1)
+                data = stream.getvalue()
+                encoded_size += len(data)
+                if encoded_size > PREVIEWS.limit:
+                    raise ValueError("The comparison batch exceeds the in-memory preview limit. Reduce batch size or increase ADACHI_COMPARE_MEMORY_MIB (64–4096 MiB).")
+                images.append(data)
+        refs = PREVIEWS.put_batch(images, unique_id)
+        before, after = refs[:original.shape[0]], refs[original.shape[0]:]
+        origin = _comparison_origin(prompt, unique_id, dynprompt)
+        return {"ui": {"adachi_originals": before, "adachi_results": after, "adachi_origin": [origin] if origin else []}, "result": (result,)}
 
 
 _MISSING_INPUT = object()
@@ -780,7 +821,7 @@ def normalize_exclusive_prompt(data):
                 inputs.pop(name, None)
             elif name not in inputs and name in sources:
                 inputs[name] = sources[name]
-    return data
+    return normalize_crop_prompt(data)
 
 
 class AdachiExclusiveSwitch:
@@ -856,6 +897,140 @@ class AdachiExclusiveSwitch:
         return (value,)
 
 
+def _crop_state(value):
+    try:
+        state = json.loads(value or "{}")
+    except (ValueError, TypeError) as exc:
+        raise ValueError("Invalid crop selection data.") from exc
+    if not isinstance(state, dict):
+        raise ValueError("Invalid crop selection data.")
+    return state
+
+
+def _crop_box(state, image_width, image_height):
+    values = [state.get(k) for k in ("x", "y", "width", "height")]
+    if any(type(v) is not int for v in values):
+        raise ValueError("Crop coordinates and dimensions must be integers.")
+    x, y, width, height = values
+    if min(width, height) < 32 or width % 32 or height % 32:
+        raise ValueError("Crop width and height must be positive multiples of 32.")
+    if x < 0 or y < 0 or x + width > image_width or y + height > image_height:
+        raise ValueError("The crop must fit entirely inside the source image.")
+    return x, y, width, height
+
+
+def _load_layer(filename):
+    if not filename:
+        return None
+    raster = _read_raster(_input_path(filename)).convert("RGB")
+    return torch.from_numpy(np.asarray(raster, dtype=np.float32).copy() / 255)[None]
+
+
+def _blend_layer(original, generated, blend_data, state):
+    if generated is None:
+        return original
+    x, y, width, height = _crop_box(state, original.shape[2], original.shape[1])
+    generated = _image_tensor(generated).to(device=original.device, dtype=original.dtype)
+    if generated.shape[1:3] != (height, width):
+        raise ValueError(f"Inpaint result must match the selected crop ({width} x {height}); connect the inpaint node's image output.")
+    alpha = torch.ones((1, height, width, 1), device=original.device, dtype=original.dtype)
+    if blend_data:
+        raster = _data_image(blend_data).convert("L")
+        if raster.size != (width, height):
+            raise ValueError("The layer mask dimensions must match the crop.")
+        alpha = torch.from_numpy(np.asarray(raster, dtype=np.float32).copy() / 255)[None, ..., None].to(original)
+    composite = original.expand(generated.shape[0], -1, -1, -1).clone()
+    old = composite[:, y:y+height, x:x+width]
+    composite[:, y:y+height, x:x+width] = old * (1-alpha) + generated * alpha
+    return composite
+
+
+def _store_layer(image):
+    import folder_paths
+    pixels = (image.detach().cpu().clamp(0, 1).numpy()*255).round().astype(np.uint8)
+    name = f"layer-{hashlib.sha256(str(pixels.shape).encode() + pixels.tobytes()).hexdigest()}.png"
+    folder = Path(folder_paths.get_input_directory()) / "adachi-inpaint-layers"
+    folder.mkdir(parents=True, exist_ok=True)
+    if not (folder / name).exists():
+        # Readers never see an incomplete PNG when concurrent prompts reuse a layer.
+        with tempfile.NamedTemporaryFile(dir=folder, suffix=".png", delete=False) as stream:
+            temporary = Path(stream.name)
+        try:
+            Image.fromarray(pixels).save(temporary)
+            temporary.replace(folder / name)
+        finally:
+            temporary.unlink(missing_ok=True)
+    return {"filename": name, "subfolder": "adachi-inpaint-layers", "type": "input"}
+
+
+class AdachiLayerCommit:
+    """Execution-only sink; keeps the visible feedback loop acyclic at runtime."""
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {"required": {
+            "result": ("IMAGE",), "image": ("STRING", {"default": ""}),
+            "source_data": ("STRING", {"default": ""}), "crop_state": ("STRING", {"default": "{}"}),
+            "blend_data": ("STRING", {"default": ""}), "target": ("STRING", {"default": ""}),
+        }}
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "compose"
+    OUTPUT_NODE = True
+    CATEGORY = "_Adachi/internal"
+    DESCRIPTION = "Internal layer sink. Use Inline Mask in Crop & Blend mode."
+
+    def compose(self, result, image="", source_data="", crop_state="{}", blend_data="", target=""):
+        raster = (_data_image(source_data) if source_data else _read_raster(_input_path(image))).convert("RGB")
+        original = torch.from_numpy(np.asarray(raster, dtype=np.float32).copy()/255)[None]
+        state = _crop_state(crop_state)
+        generated = _image_tensor(result).cpu()
+        composite = _blend_layer(original, generated, blend_data, state)
+        layer = _store_layer(generated[-1])
+        return {"ui": {"adachi_layer": [{"target": target, "job": state.get("job", ""), "image": layer}]}, "result": (composite,)}
+
+
+def normalize_crop_prompt(data):
+    """Ignore obsolete feedback links; adoption now happens only in Image Compare."""
+    prompt = data.get("prompt", {})
+    if isinstance(prompt, dict):
+        for node in prompt.values():
+            if isinstance(node, dict) and node.get("class_type") == "AdachiInlineMask":
+                node.setdefault("inputs", {}).pop("result", None)
+    return data
+
+
+def _comparison_origin(prompt, unique_id, dynprompt=None):
+    """Tag previews with their generating crop, so later adoption cannot hit another region."""
+    def get_node(node_id):
+        if dynprompt is not None:
+            try:
+                return dynprompt.get_node(str(node_id))
+            except KeyError:
+                pass
+        return prompt.get(str(node_id), prompt.get(node_id)) if isinstance(prompt, dict) else None
+    node = get_node(unique_id)
+    link = node.get("inputs", {}).get("original") if isinstance(node, dict) else None
+    queue = [str(link[0])] if isinstance(link, list) and len(link) == 2 else []
+    seen, sources = set(), {}
+    while queue and len(seen) < 256:
+        node_id = queue.pop()
+        if node_id in seen:
+            continue
+        seen.add(node_id)
+        current = get_node(node_id)
+        if not isinstance(current, dict):
+            continue
+        inputs = current.get("inputs", {})
+        if current.get("class_type") == "AdachiInlineMask":
+            mode = inputs.get("edit_mode", "mask")
+            state = _crop_state(inputs.get("crop_state", "{}")) if mode == "crop" else {}
+            sources[node_id] = {"source": node_id, "mode": mode, "job": state.get("job", "")}
+            continue
+        for value in inputs.values():
+            if isinstance(value, list) and len(value) == 2 and type(value[1]) is int:
+                queue.append(str(value[0]))
+    return next(iter(sources.values())) if len(sources) == 1 else None
+
+
 NODE_CLASS_MAPPINGS = {
     "AdachiInlineMask": AdachiInlineMask,
     "AdachiInpaintPrepare": AdachiInpaintPrepare,
@@ -863,13 +1038,15 @@ NODE_CLASS_MAPPINGS = {
     "AdachiWebUIInpaint": AdachiWebUIInpaint,
     "AdachiImageCompare": AdachiImageCompare,
     "AdachiExclusiveSwitch": AdachiExclusiveSwitch,
+    "AdachiLayerCommit": AdachiLayerCommit,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
     "AdachiInlineMask": "Inline Mask",
     "AdachiInpaintPrepare": "Inpaint Prepare",
     "AdachiInpaintComposite": "Inpaint Composite",
-    "AdachiWebUIInpaint": "WebUI Inpaint",
+    "AdachiWebUIInpaint": "Adachi Inpaint Workbench",
     "AdachiImageCompare": "Image Compare",
     "AdachiExclusiveSwitch": "Auto Exclusive Switch",
+    "AdachiLayerCommit": "Internal Layer Commit",
 }

examples/01-inline-inpaint.json

@@ -1,694 +0,0 @@
-{
-  "id": "00000000-0000-0000-0000-000000000000",
-  "revision": 0,
-  "last_node_id": 8,
-  "last_link_id": 11,
-  "nodes": [
-    {
-      "id": 1,
-      "type": "AdachiInlineMask",
-      "pos": [
-        40,
-        80
-      ],
-      "size": [
-        660,
-        700
-      ],
-      "flags": {},
-      "order": 0,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "image",
-          "type": "STRING",
-          "widget": {
-            "name": "image"
-          },
-          "link": null
-        },
-        {
-          "name": "mask_data",
-          "type": "STRING",
-          "widget": {
-            "name": "mask_data"
-          },
-          "link": null
-        },
-        {
-          "name": "source_data",
-          "type": "STRING",
-          "widget": {
-            "name": "source_data"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "links": [
-            7,
-            9
-          ]
-        },
-        {
-          "name": "mask",
-          "type": "MASK",
-          "links": [
-            8
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "AdachiInlineMask"
-      },
-      "widgets_values": [
-        "",
-        "",
-        ""
-      ]
-    },
-    {
-      "id": 2,
-      "type": "CheckpointLoaderSimple",
-      "pos": [
-        780,
-        80
-      ],
-      "size": [
-        340,
-        130.65625
-      ],
-      "flags": {},
-      "order": 1,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "ckpt_name",
-          "type": "COMBO",
-          "widget": {
-            "name": "ckpt_name"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "MODEL",
-          "type": "MODEL",
-          "links": [
-            3
-          ]
-        },
-        {
-          "name": "CLIP",
-          "type": "CLIP",
-          "links": [
-            1,
-            2
-          ]
-        },
-        {
-          "name": "VAE",
-          "type": "VAE",
-          "links": [
-            4
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "CheckpointLoaderSimple"
-      },
-      "widgets_values": [
-        "CHOOSE_YOUR_CHECKPOINT.safetensors"
-      ]
-    },
-    {
-      "id": 3,
-      "type": "CLIPTextEncode",
-      "pos": [
-        780,
-        250
-      ],
-      "size": [
-        350,
-        150
-      ],
-      "flags": {},
-      "order": 3,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "clip",
-          "type": "CLIP",
-          "link": 1
-        },
-        {
-          "name": "text",
-          "type": "STRING",
-          "widget": {
-            "name": "text"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "CONDITIONING",
-          "type": "CONDITIONING",
-          "links": [
-            5
-          ]
-        }
-      ],
-      "title": "Positive",
-      "properties": {
-        "Node name for S&R": "CLIPTextEncode"
-      },
-      "widgets_values": [
-        "A detailed natural photograph, coherent lighting, high quality"
-      ]
-    },
-    {
-      "id": 4,
-      "type": "CLIPTextEncode",
-      "pos": [
-        780,
-        450
-      ],
-      "size": [
-        350,
-        150
-      ],
-      "flags": {},
-      "order": 4,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "clip",
-          "type": "CLIP",
-          "link": 2
-        },
-        {
-          "name": "text",
-          "type": "STRING",
-          "widget": {
-            "name": "text"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "CONDITIONING",
-          "type": "CONDITIONING",
-          "links": [
-            6
-          ]
-        }
-      ],
-      "title": "Negative",
-      "properties": {
-        "Node name for S&R": "CLIPTextEncode"
-      },
-      "widgets_values": [
-        "blurry, distorted, artifacts"
-      ]
-    },
-    {
-      "id": 5,
-      "type": "AdachiWebUIInpaint",
-      "pos": [
-        1220,
-        80
-      ],
-      "size": [
-        430,
-        1020
-      ],
-      "flags": {},
-      "order": 5,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "model",
-          "type": "MODEL",
-          "link": 3
-        },
-        {
-          "name": "positive",
-          "type": "CONDITIONING",
-          "link": 5
-        },
-        {
-          "name": "negative",
-          "type": "CONDITIONING",
-          "link": 6
-        },
-        {
-          "name": "vae",
-          "type": "VAE",
-          "link": 4
-        },
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "link": 7
-        },
-        {
-          "name": "mask",
-          "type": "MASK",
-          "link": 8
-        },
-        {
-          "name": "refiner_model",
-          "shape": 7,
-          "type": "MODEL",
-          "link": null
-        },
-        {
-          "name": "refiner_positive",
-          "shape": 7,
-          "type": "CONDITIONING",
-          "link": null
-        },
-        {
-          "name": "refiner_negative",
-          "shape": 7,
-          "type": "CONDITIONING",
-          "link": null
-        },
-        {
-          "name": "area",
-          "type": "COMBO",
-          "widget": {
-            "name": "area"
-          },
-          "link": null
-        },
-        {
-          "name": "width",
-          "type": "INT",
-          "widget": {
-            "name": "width"
-          },
-          "link": null
-        },
-        {
-          "name": "height",
-          "type": "INT",
-          "widget": {
-            "name": "height"
-          },
-          "link": null
-        },
-        {
-          "name": "padding",
-          "type": "INT",
-          "widget": {
-            "name": "padding"
-          },
-          "link": null
-        },
-        {
-          "name": "mask_blur",
-          "type": "FLOAT",
-          "widget": {
-            "name": "mask_blur"
-          },
-          "link": null
-        },
-        {
-          "name": "mask_grow",
-          "type": "INT",
-          "widget": {
-            "name": "mask_grow"
-          },
-          "link": null
-        },
-        {
-          "name": "mask_mode",
-          "type": "COMBO",
-          "widget": {
-            "name": "mask_mode"
-          },
-          "link": null
-        },
-        {
-          "name": "fill",
-          "type": "COMBO",
-          "widget": {
-            "name": "fill"
-          },
-          "link": null
-        },
-        {
-          "name": "resize_mode",
-          "type": "COMBO",
-          "widget": {
-            "name": "resize_mode"
-          },
-          "link": null
-        },
-        {
-          "name": "seed",
-          "type": "INT",
-          "widget": {
-            "name": "seed"
-          },
-          "link": null
-        },
-        {
-          "name": "steps",
-          "type": "INT",
-          "widget": {
-            "name": "steps"
-          },
-          "link": null
-        },
-        {
-          "name": "cfg",
-          "type": "FLOAT",
-          "widget": {
-            "name": "cfg"
-          },
-          "link": null
-        },
-        {
-          "name": "sampler_name",
-          "type": "COMBO",
-          "widget": {
-            "name": "sampler_name"
-          },
-          "link": null
-        },
-        {
-          "name": "scheduler",
-          "type": "COMBO",
-          "widget": {
-            "name": "scheduler"
-          },
-          "link": null
-        },
-        {
-          "name": "denoise",
-          "type": "FLOAT",
-          "widget": {
-            "name": "denoise"
-          },
-          "link": null
-        },
-        {
-          "name": "batch_size",
-          "type": "INT",
-          "widget": {
-            "name": "batch_size"
-          },
-          "link": null
-        },
-        {
-          "name": "refiner_switch",
-          "type": "FLOAT",
-          "widget": {
-            "name": "refiner_switch"
-          },
-          "link": null
-        },
-        {
-          "name": "inpaint_model",
-          "shape": 7,
-          "type": "BOOLEAN",
-          "widget": {
-            "name": "inpaint_model"
-          },
-          "link": null
-        },
-        {
-          "name": "soft_edge",
-          "shape": 7,
-          "type": "BOOLEAN",
-          "widget": {
-            "name": "soft_edge"
-          },
-          "link": null
-        },
-        {
-          "name": "soft_strength",
-          "shape": 7,
-          "type": "FLOAT",
-          "widget": {
-            "name": "soft_strength"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "links": [
-            10
-          ]
-        },
-        {
-          "name": "crop",
-          "type": "IMAGE",
-          "links": null
-        },
-        {
-          "name": "mask",
-          "type": "MASK",
-          "links": null
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "AdachiWebUIInpaint"
-      },
-      "widgets_values": [
-        "only_masked",
-        1024,
-        1024,
-        32,
-        4,
-        0,
-        "inside",
-        "original",
-        "stretch",
-        42,
-        "fixed",
-        20,
-        7,
-        "dpmpp_2m",
-        "karras",
-        0.75,
-        1,
-        0.8,
-        false,
-        false,
-        0.5
-      ]
-    },
-    {
-      "id": 6,
-      "type": "AdachiImageCompare",
-      "pos": [
-        1770,
-        80
-      ],
-      "size": [
-        590,
-        620
-      ],
-      "flags": {},
-      "order": 6,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "original",
-          "type": "IMAGE",
-          "link": 9
-        },
-        {
-          "name": "result",
-          "type": "IMAGE",
-          "link": 10
-        }
-      ],
-      "outputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "links": [
-            11
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "AdachiImageCompare"
-      },
-      "widgets_values": []
-    },
-    {
-      "id": 7,
-      "type": "SaveImage",
-      "pos": [
-        2470,
-        80
-      ],
-      "size": [
-        430,
-        450
-      ],
-      "flags": {},
-      "order": 7,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "images",
-          "type": "IMAGE",
-          "link": 11
-        },
-        {
-          "name": "filename_prefix",
-          "type": "STRING",
-          "widget": {
-            "name": "filename_prefix"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [],
-      "properties": {},
-      "widgets_values": [
-        "Adachi/Inpaint"
-      ]
-    },
-    {
-      "id": 8,
-      "type": "Note",
-      "pos": [
-        780,
-        670
-      ],
-      "size": [
-        350,
-        220
-      ],
-      "flags": {},
-      "order": 2,
-      "mode": 0,
-      "inputs": [],
-      "outputs": [],
-      "properties": {},
-      "widgets_values": [
-        "Adachi-Inpaint-Workbench\n1. Pick your own checkpoint\n2. Load an image in the left node and paint directly on it\n3. Select the brush and adjust inpainting parameters in the WebUI Inpaint node\n4. width/height is the sampling resolution; the final image is composited back at the source size\nTick \"Embed source in workflow\" in the inpaint node before sharing"
-      ],
-      "color": "#432",
-      "bgcolor": "#653"
-    }
-  ],
-  "links": [
-    [
-      1,
-      2,
-      1,
-      3,
-      0,
-      "CLIP"
-    ],
-    [
-      2,
-      2,
-      1,
-      4,
-      0,
-      "CLIP"
-    ],
-    [
-      3,
-      2,
-      0,
-      5,
-      0,
-      "MODEL"
-    ],
-    [
-      4,
-      2,
-      2,
-      5,
-      3,
-      "VAE"
-    ],
-    [
-      5,
-      3,
-      0,
-      5,
-      1,
-      "CONDITIONING"
-    ],
-    [
-      6,
-      4,
-      0,
-      5,
-      2,
-      "CONDITIONING"
-    ],
-    [
-      7,
-      1,
-      0,
-      5,
-      4,
-      "IMAGE"
-    ],
-    [
-      8,
-      1,
-      1,
-      5,
-      5,
-      "MASK"
-    ],
-    [
-      9,
-      1,
-      0,
-      6,
-      0,
-      "IMAGE"
-    ],
-    [
-      10,
-      5,
-      0,
-      6,
-      1,
-      "IMAGE"
-    ],
-    [
-      11,
-      6,
-      0,
-      7,
-      0,
-      "IMAGE"
-    ]
-  ],
-  "groups": [],
-  "config": {},
-  "extra": {
-    "ds": {
-      "scale": 1,
-      "offset": [
-        0,
-        0
-      ]
-    }
-  },
-  "version": 0.4
-}

examples/03-group-presets.json

@@ -1,335 +0,0 @@
-{
-  "id": "00000000-0000-0000-0000-000000000000",
-  "revision": 0,
-  "last_node_id": 6,
-  "last_link_id": 3,
-  "nodes": [
-    {
-      "id": 1,
-      "type": "EmptyImage",
-      "pos": [
-        -350,
-        140
-      ],
-      "size": [
-        270,
-        160
-      ],
-      "flags": {},
-      "order": 0,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "width",
-          "type": "INT",
-          "widget": {
-            "name": "width"
-          },
-          "link": null
-        },
-        {
-          "name": "height",
-          "type": "INT",
-          "widget": {
-            "name": "height"
-          },
-          "link": null
-        },
-        {
-          "name": "batch_size",
-          "type": "INT",
-          "widget": {
-            "name": "batch_size"
-          },
-          "link": null
-        },
-        {
-          "name": "color",
-          "type": "INT",
-          "widget": {
-            "name": "color"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            1
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "EmptyImage"
-      },
-      "widgets_values": [
-        512,
-        512,
-        1,
-        8096934
-      ]
-    },
-    {
-      "id": 2,
-      "type": "ImageInvert",
-      "pos": [
-        120,
-        160
-      ],
-      "size": [
-        225,
-        48
-      ],
-      "flags": {},
-      "order": 3,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "link": 1
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            2
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "ImageInvert"
-      },
-      "widgets_values": []
-    },
-    {
-      "id": 3,
-      "type": "ImageInvert",
-      "pos": [
-        510,
-        160
-      ],
-      "size": [
-        225,
-        48
-      ],
-      "flags": {},
-      "order": 4,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "link": 2
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            3
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "ImageInvert"
-      },
-      "widgets_values": []
-    },
-    {
-      "id": 4,
-      "type": "SaveImage",
-      "pos": [
-        920,
-        140
-      ],
-      "size": [
-        270,
-        76
-      ],
-      "flags": {},
-      "order": 5,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "images",
-          "type": "IMAGE",
-          "link": 3
-        },
-        {
-          "name": "filename_prefix",
-          "type": "STRING",
-          "widget": {
-            "name": "filename_prefix"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [],
-      "properties": {},
-      "widgets_values": [
-        "Adachi/GroupPresetExample"
-      ]
-    },
-    {
-      "id": 5,
-      "type": "AdachiGroupPresets",
-      "pos": [
-        80,
-        440
-      ],
-      "size": [
-        620,
-        410
-      ],
-      "flags": {},
-      "order": 1,
-      "mode": 0,
-      "inputs": [],
-      "outputs": [],
-      "properties": {
-        "adachi_group_presets": {
-          "version": 1,
-          "activeId": "01dbf1d1-c15a-4db1-97d5-9d522781c482",
-          "presets": [
-            {
-              "id": "01dbf1d1-c15a-4db1-97d5-9d522781c482",
-              "name": "All enabled",
-              "groups": {
-                "group:1": {
-                  "title": "Base",
-                  "mode": 0
-                },
-                "group:2": {
-                  "title": "Detail",
-                  "mode": 0
-                }
-              }
-            },
-            {
-              "id": "fe43bc1f-1a1e-4f49-a37b-c6e6ca7cbc8a",
-              "name": "Skip detail",
-              "groups": {
-                "group:1": {
-                  "title": "Base",
-                  "mode": 0
-                },
-                "group:2": {
-                  "title": "Detail",
-                  "mode": 4
-                }
-              }
-            },
-            {
-              "id": "b881a666-4643-4af6-8feb-f1d02c47f113",
-              "name": "Bypass both",
-              "groups": {
-                "group:1": {
-                  "title": "Base",
-                  "mode": 4
-                },
-                "group:2": {
-                  "title": "Detail",
-                  "mode": 4
-                }
-              }
-            }
-          ]
-        }
-      }
-    },
-    {
-      "id": 6,
-      "type": "Note",
-      "pos": [
-        820,
-        550
-      ],
-      "size": [
-        350,
-        180
-      ],
-      "flags": {},
-      "order": 2,
-      "mode": 0,
-      "inputs": [],
-      "outputs": [],
-      "properties": {},
-      "widgets_values": [
-        "Adachi-Inpaint-Workbench - Group Bypass Presets\nSelect a preset on the left to switch a whole group between enabled and bypassed; no connections required\nAdjust each group's state on the right, or click \"Save current group states\" to store the current workflow\nRight-click the node -> Apply preset to switch quickly\nThis example only inverts an image, so no model is needed"
-      ],
-      "color": "#432",
-      "bgcolor": "#653"
-    }
-  ],
-  "links": [
-    [
-      1,
-      1,
-      0,
-      2,
-      0,
-      "IMAGE"
-    ],
-    [
-      2,
-      2,
-      0,
-      3,
-      0,
-      "IMAGE"
-    ],
-    [
-      3,
-      3,
-      0,
-      4,
-      0,
-      "IMAGE"
-    ]
-  ],
-  "groups": [
-    {
-      "id": 1,
-      "title": "Base",
-      "bounding": [
-        80,
-        100,
-        300,
-        230
-      ],
-      "color": "#3f789e",
-      "font_size": 24,
-      "flags": {}
-    },
-    {
-      "id": 2,
-      "title": "Detail",
-      "bounding": [
-        470,
-        100,
-        300,
-        230
-      ],
-      "color": "#3f789e",
-      "font_size": 24,
-      "flags": {}
-    }
-  ],
-  "config": {},
-  "extra": {
-    "ds": {
-      "scale": 1,
-      "offset": [
-        0,
-        0
-      ]
-    }
-  },
-  "version": 0.4
-}

examples/04-exclusive-switch.json

@@ -1,423 +0,0 @@
-{
-  "id": "00000000-0000-0000-0000-000000000000",
-  "revision": 0,
-  "last_node_id": 7,
-  "last_link_id": 5,
-  "nodes": [
-    {
-      "id": 1,
-      "type": "EmptyImage",
-      "pos": [
-        0,
-        120
-      ],
-      "size": [
-        270,
-        130
-      ],
-      "flags": {},
-      "order": 0,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "width",
-          "type": "INT",
-          "widget": {
-            "name": "width"
-          },
-          "link": null
-        },
-        {
-          "name": "height",
-          "type": "INT",
-          "widget": {
-            "name": "height"
-          },
-          "link": null
-        },
-        {
-          "name": "batch_size",
-          "type": "INT",
-          "widget": {
-            "name": "batch_size"
-          },
-          "link": null
-        },
-        {
-          "name": "color",
-          "type": "INT",
-          "widget": {
-            "name": "color"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            1
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "EmptyImage"
-      },
-      "widgets_values": [
-        64,
-        64,
-        1,
-        16711680
-      ]
-    },
-    {
-      "id": 2,
-      "type": "EmptyImage",
-      "pos": [
-        0,
-        450
-      ],
-      "size": [
-        270,
-        130
-      ],
-      "flags": {},
-      "order": 1,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "width",
-          "type": "INT",
-          "widget": {
-            "name": "width"
-          },
-          "link": null
-        },
-        {
-          "name": "height",
-          "type": "INT",
-          "widget": {
-            "name": "height"
-          },
-          "link": null
-        },
-        {
-          "name": "batch_size",
-          "type": "INT",
-          "widget": {
-            "name": "batch_size"
-          },
-          "link": null
-        },
-        {
-          "name": "color",
-          "type": "INT",
-          "widget": {
-            "name": "color"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            2
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "EmptyImage"
-      },
-      "widgets_values": [
-        64,
-        64,
-        1,
-        255
-      ]
-    },
-    {
-      "id": 3,
-      "type": "ImageInvert",
-      "pos": [
-        300,
-        120
-      ],
-      "size": [
-        140,
-        26
-      ],
-      "flags": {},
-      "order": 3,
-      "mode": 4,
-      "inputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "link": 1
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            3
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "ImageInvert"
-      }
-    },
-    {
-      "id": 4,
-      "type": "ImageInvert",
-      "pos": [
-        300,
-        450
-      ],
-      "size": [
-        140,
-        26
-      ],
-      "flags": {},
-      "order": 4,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "image",
-          "type": "IMAGE",
-          "link": 2
-        }
-      ],
-      "outputs": [
-        {
-          "name": "IMAGE",
-          "type": "IMAGE",
-          "links": [
-            4
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "ImageInvert"
-      }
-    },
-    {
-      "id": 5,
-      "type": "AdachiExclusiveSwitch",
-      "pos": [
-        620,
-        200
-      ],
-      "size": [
-        230,
-        80
-      ],
-      "flags": {},
-      "order": 5,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "input_a",
-          "shape": 7,
-          "type": "*",
-          "link": 3
-        },
-        {
-          "name": "input_b",
-          "shape": 7,
-          "type": "*",
-          "link": 4
-        }
-      ],
-      "outputs": [
-        {
-          "name": "output",
-          "type": "*",
-          "links": [
-            5
-          ]
-        }
-      ],
-      "properties": {
-        "Node name for S&R": "AdachiExclusiveSwitch"
-      },
-      "widgets_values": [
-        -1
-      ]
-    },
-    {
-      "id": 6,
-      "type": "SaveImage",
-      "pos": [
-        940,
-        200
-      ],
-      "size": [
-        270,
-        58
-      ],
-      "flags": {},
-      "order": 6,
-      "mode": 0,
-      "inputs": [
-        {
-          "name": "images",
-          "type": "IMAGE",
-          "link": 5
-        },
-        {
-          "name": "filename_prefix",
-          "type": "STRING",
-          "widget": {
-            "name": "filename_prefix"
-          },
-          "link": null
-        }
-      ],
-      "outputs": [],
-      "properties": {},
-      "widgets_values": [
-        "exclusive-test"
-      ]
-    },
-    {
-      "id": 7,
-      "type": "AdachiGroupPresets",
-      "pos": [
-        600,
-        430
-      ],
-      "size": [
-        620,
-        410
-      ],
-      "flags": {},
-      "order": 2,
-      "mode": 0,
-      "inputs": [],
-      "outputs": [],
-      "properties": {
-        "adachi_group_presets": {
-          "version": 1,
-          "activeId": "2908ca1a-7a3f-4eba-8203-524ab3f52d27",
-          "presets": [
-            {
-              "id": "dd436f10-d4a4-46bc-aecb-fe185710f6e0",
-              "name": "Use A only",
-              "groups": {
-                "group:1": {
-                  "title": "A",
-                  "mode": 0
-                },
-                "group:2": {
-                  "title": "B",
-                  "mode": 4
-                }
-              }
-            },
-            {
-              "id": "2908ca1a-7a3f-4eba-8203-524ab3f52d27",
-              "name": "Use B only",
-              "groups": {
-                "group:1": {
-                  "title": "A",
-                  "mode": 4
-                },
-                "group:2": {
-                  "title": "B",
-                  "mode": 0
-                }
-              }
-            }
-          ]
-        }
-      }
-    }
-  ],
-  "links": [
-    [
-      1,
-      1,
-      0,
-      3,
-      0,
-      "IMAGE"
-    ],
-    [
-      2,
-      2,
-      0,
-      4,
-      0,
-      "IMAGE"
-    ],
-    [
-      3,
-      3,
-      0,
-      5,
-      0,
-      "IMAGE"
-    ],
-    [
-      4,
-      4,
-      0,
-      5,
-      1,
-      "IMAGE"
-    ],
-    [
-      5,
-      5,
-      0,
-      6,
-      0,
-      "IMAGE"
-    ]
-  ],
-  "groups": [
-    {
-      "id": 1,
-      "title": "A",
-      "bounding": [
-        270,
-        70,
-        310,
-        210
-      ],
-      "color": "#3f789e",
-      "font_size": 24,
-      "flags": {}
-    },
-    {
-      "id": 2,
-      "title": "B",
-      "bounding": [
-        270,
-        400,
-        310,
-        210
-      ],
-      "color": "#3f789e",
-      "font_size": 24,
-      "flags": {}
-    }
-  ],
-  "config": {},
-  "extra": {
-    "ds": {
-      "scale": 0.8,
-      "offset": [
-        100,
-        20
-      ]
-    }
-  },
-  "version": 0.4
-}

install.ps1

@@ -54,5 +54,14 @@ Get-ChildItem -LiteralPath $packageRoot -Force | Where-Object {
 } | ForEach-Object {
     Copy-Item -LiteralPath $_.FullName -Destination $destination -Recurse -Force
 }
+# These tracked files were retired in 0.6.2. The previous installation ZIP above
+# retains them for rollback; remove them from an overwritten installation too.
+foreach ($retiredRelative in @('web\magic_wand.js', 'web\magic_wand_worker.js', 'web\wand_selection.js', 'tests\test_magic_wand.mjs')) {
+    $retiredPath = [IO.Path]::GetFullPath((Join-Path $destination $retiredRelative))
+    if (-not $retiredPath.StartsWith($destination.TrimEnd('\') + '\', [StringComparison]::OrdinalIgnoreCase)) {
+        throw 'Retired plugin file must stay within the install directory.'
+    }
+    if (Test-Path -LiteralPath $retiredPath -PathType Leaf) { Remove-Item -LiteralPath $retiredPath -Force }
+}
 Write-Host "Installed Adachi-Inpaint-Workbench: $destination"
 Write-Host 'Restart ComfyUI, then refresh the browser with Ctrl+F5.'

locales/en/nodeDefs.json

@@ -14,6 +14,18 @@
       "source_data": {
         "name": "source_data",
         "tooltip": "Embedded source image for portable workflows."
+      },
+      "edit_mode": {
+        "name": "edit_mode"
+      },
+      "crop_state": {
+        "name": "crop_state"
+      },
+      "blend_data": {
+        "name": "blend_data"
+      },
+      "layer_image": {
+        "name": "layer_image"
       }
     },
     "outputs": {
@@ -22,6 +34,9 @@
       },
       "1": {
         "name": "mask"
+      },
+      "2": {
+        "name": "full_image"
       }
     }
   },
@@ -114,7 +129,7 @@
     }
   },
   "AdachiWebUIInpaint": {
-    "display_name": "WebUI Inpaint",
+    "display_name": "Adachi Inpaint Workbench",
     "description": "WebUI-style inpainting, crop, latent fill, sampling and original-size compositing.",
     "inputs": {
       "image": {
@@ -185,6 +200,10 @@
         "name": "soft_edge",
         "tooltip": "Extra feathering, not A1111 Soft Inpainting."
       },
+      "soft_strength": {
+        "name": "soft_strength",
+        "tooltip": "Adds 0–16 pixels of Gaussian feathering."
+      },
       "refiner_model": {
         "name": "refiner_model",
         "tooltip": "Optional compatible refiner model, e.g. SDXL refiner."

locales/zh/nodeDefs.json

@@ -14,6 +14,18 @@
       "source_data": {
         "name": "嵌入原图",
         "tooltip": "嵌入原图，便于工作流在其他机器使用。"
+      },
+      "edit_mode": {
+        "name": "编辑模式"
+      },
+      "crop_state": {
+        "name": "裁切选区"
+      },
+      "blend_data": {
+        "name": "融合蒙版"
+      },
+      "layer_image": {
+        "name": "重绘图层"
       }
     },
     "outputs": {
@@ -22,6 +34,9 @@
       },
       "1": {
         "name": "蒙版"
+      },
+      "2": {
+        "name": "完整图像"
       }
     }
   },
@@ -114,7 +129,7 @@
     }
   },
   "AdachiWebUIInpaint": {
-    "display_name": "WebUI 局部重绘 / WebUI Inpaint",
+    "display_name": "安达重绘工作台",
     "description": "WebUI 式局部重绘：整图/蒙版区域、填充、采样与原尺寸回贴。",
     "inputs": {
       "image": {

preview_cleanup.py

@@ -0,0 +1,55 @@
+"""Expire only this extension's comparison previews in ComfyUI's temp folder."""
+
+import logging
+import os
+import re
+import threading
+import time
+
+PREVIEW_MAX_AGE_SECONDS = 24 * 60 * 60
+SWEEP_INTERVAL_SECONDS = 60 * 60
+_PREVIEW_NAME = re.compile(r"adachi_compare_(?:original|result)_temp_[a-z]{5}_[0-9]{5,}_\.png\Z")
+_started = False
+_start_lock = threading.Lock()
+
+
+def cleanup_comparison_previews(directory, *, now=None, max_age=PREVIEW_MAX_AGE_SECONDS):
+    """Remove expired, regular preview files, leaving all other files untouched."""
+    cutoff = (time.time() if now is None else now) - max_age
+    removed = 0
+    try:
+        with os.scandir(directory) as entries:
+            for entry in entries:
+                if not _PREVIEW_NAME.fullmatch(entry.name):
+                    continue
+                try:
+                    if not entry.is_file(follow_symlinks=False):
+                        continue
+                    if entry.stat(follow_symlinks=False).st_mtime >= cutoff:
+                        continue
+                    os.unlink(entry.path)
+                    removed += 1
+                except OSError as exc:
+                    logging.debug("Could not remove expired comparison preview %s: %s", entry.name, exc)
+    except FileNotFoundError:
+        pass
+    except OSError as exc:
+        logging.warning("Could not inspect comparison preview directory: %s", exc)
+    return removed
+
+
+def start_preview_cleanup():
+    """Sweep in the background so a long-running ComfyUI does not accumulate previews."""
+    global _started
+    with _start_lock:
+        if _started:
+            return
+        _started = True
+
+    def run():
+        import folder_paths
+        while True:
+            cleanup_comparison_previews(folder_paths.get_temp_directory())
+            time.sleep(SWEEP_INTERVAL_SECONDS)
+
+    threading.Thread(target=run, name="adachi-preview-cleanup", daemon=True).start()

preview_store.py

@@ -0,0 +1,85 @@
+"""Bounded in-memory comparison PNGs. No preview is written to disk."""
+
+from collections import OrderedDict
+import hashlib
+import os
+import secrets
+import threading
+
+
+def _limit():
+    try:
+        mib = int(os.environ.get("ADACHI_COMPARE_MEMORY_MIB", "512"))
+    except ValueError:
+        mib = 512
+    return max(64, min(4096, mib)) * 1024 * 1024
+
+
+class PreviewStore:
+    def __init__(self, limit=None, max_batches=512):
+        self.limit = _limit() if limit is None else limit
+        self.max_batches = max(1, int(max_batches))
+        self.nonce = secrets.token_hex(12)
+        self.epoch = 0
+        self.size = 0
+        self.groups = OrderedDict()
+        self.keys = {}
+        self.latest = {}
+        self.lock = threading.RLock()
+
+    def fingerprint(self):
+        with self.lock:
+            return f"{self.nonce}:{self.epoch}"
+
+    def put_batch(self, images, owner):
+        """Atomically retain a whole comparison batch, or explicitly reject it."""
+        if not images or any(not image for image in images):
+            raise ValueError("The comparison batch must contain nonempty images.")
+        total = sum(len(image) for image in images)
+        if total > self.limit:
+            raise ValueError("The comparison batch exceeds the in-memory preview limit. Reduce batch size or increase ADACHI_COMPARE_MEMORY_MIB (64–4096 MiB).")
+        digest = hashlib.sha256()
+        for image in images:
+            digest.update(len(image).to_bytes(8, "big")); digest.update(image)
+        signature = digest.digest()
+        owner = str(owner)
+        with self.lock:
+            before = self.epoch
+            last = self.latest.get(owner)
+            # Cache fingerprinting happens before execution. If eviction of an
+            # unrelated or older batch caused a rerun but pixels are unchanged,
+            # keep the existing descriptors and let the fingerprint stabilise.
+            if last and last[1] != before and last[0] in self.groups:
+                group = self.groups[last[0]]
+                if group["signature"] == signature:
+                    self.groups.move_to_end(last[0])
+                    self.latest[owner] = (last[0], before)
+                    return list(group["descriptors"])
+            while self.groups and (self.size + total > self.limit or len(self.groups) >= self.max_batches):
+                group_id, group = self.groups.popitem(last=False)
+                self.size -= group["size"]
+                for key in group["keys"]:
+                    self.keys.pop(key, None)
+                if self.latest.get(group["owner"], (None,))[0] == group_id:
+                    self.latest.pop(group["owner"], None)
+                self.epoch += 1
+            group_id = secrets.token_urlsafe(24)
+            keys = [secrets.token_urlsafe(24) for _ in images]
+            descriptors = [{"type": "adachi_memory", "subfolder": "", "filename": key + ".png"} for key in keys]
+            self.groups[group_id] = {"owner": owner, "size": total, "keys": keys, "signature": signature, "descriptors": descriptors}
+            for key, image in zip(keys, images):
+                self.keys[key] = (group_id, image)
+            self.size += total
+            self.latest[owner] = (group_id, before)
+            return list(descriptors)
+
+    def get(self, key):
+        with self.lock:
+            found = self.keys.get(key)
+            if not found:
+                return None
+            self.groups.move_to_end(found[0])
+            return found[1]
+
+
+PREVIEWS = PreviewStore()

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "adachi-inpaint-workbench"
-version = "0.5.13"
-description = "Inline mask painting and WebUI-style inpainting nodes for ComfyUI"
+version = "2.0.1"
+description = "Inline mask painting, crop blending and a standalone inpainting workbench for ComfyUI"
 requires-python = ">=3.10"
 license = { file = "LICENSE" }
 readme = "README.md"

registry/README.md

@@ -1,22 +1,25 @@
 # Manager / Registry publishing
 
-A public GitHub repository does not by itself mean the plugin is listed in the Manager search index or in the Comfy Registry.
-The Manager listing submission for this plugin is [PR #3378](https://github.com/Comfy-Org/ComfyUI-Manager/pull/3378), currently waiting on the upstream maintainer.
+The node is registered under publisher `adachi0130` in [Comfy Registry](https://registry.comfy.org/nodes/adachi-inpaint-workbench). Version visibility and installation can depend on Registry validation, so check the current status there.
 
 ## Manager listing
 
-[manager-entry.json](manager-entry.json) is the entry submitted to [ComfyUI-Manager](https://github.com/Comfy-Org/ComfyUI-Manager); it is added to the `custom-node-list.json` at the repository root in a pull request, and the maintainer merges it and refreshes the index.
-Follow the [official Manager instructions](https://github.com/Comfy-Org/ComfyUI-Manager#how-to-register-your-custom-node-into-comfyui-manager).
-Until it is listed you can still install through the Manager's Git URL option, using `https://github.com/adachi0130/Adachi-Inpaint-Workbench`.
+[manager-entry.json](manager-entry.json) records the legacy list entry submitted to [ComfyUI-Manager](https://github.com/Comfy-Org/ComfyUI-Manager).
+For current registration requirements, follow the [official Manager instructions](https://github.com/Comfy-Org/ComfyUI-Manager#how-to-register-your-custom-node-into-comfyui-manager) and the Registry publishing guide below.
+You can also install through the Manager's Git URL option, using `https://github.com/adachi0130/Adachi-Inpaint-Workbench`.
 
 ## Comfy Registry
 
-Per the [official publishing guide](https://docs.comfy.org/registry/publishing), the first publish requires you to create your own Publisher and API key.
+Per the [official publishing guide](https://docs.comfy.org/registry/publishing), publishing requires the registered Publisher and its API key.
 
-1. Sign in at [Comfy Registry](https://registry.comfy.org/) and create a Publisher, then note the actual Publisher ID.
-2. Replace the empty `[tool.comfy].PublisherId` in `pyproject.toml` with that ID and push it to GitHub.
+1. Sign in at [Comfy Registry](https://registry.comfy.org/) as the owner of publisher `adachi0130`.
+2. Check `[tool.comfy].PublisherId` in `pyproject.toml` (currently `adachi0130`) and confirm it matches the Publisher you own before publishing.
 3. In the repository Settings → Secrets and variables → Actions, add `REGISTRY_ACCESS_TOKEN` with the Registry publishing API key.
 4. Run **Publish to Comfy Registry** manually from the Actions tab and check the publish and platform validation results.
 
-The workflow is manual-only today and will not publish while no Publisher ID or API key is configured.
+The publishing workflow is manual-only. Check its result and the Registry version status after publishing.
 A Registry version cannot be overwritten once published, so later releases have to bump the version in `pyproject.toml`.
+
+## 2.0.1 verification
+
+After publishing 2.0.1, check both the Actions result and the new version's status in the [public Registry API](https://api.comfy.org/nodes/adachi-inpaint-workbench/versions?include_status_reason=true). A successful publishing job confirms upload, not Registry approval or Manager availability. Previous versions received automated scan flags; [scan review notes](SECURITY-REVIEW.md) explain the public findings and the checks required for this release. Do not report the version as approved while its status is `Pending` or `Flagged`.

registry/SECURITY-REVIEW.md

@@ -0,0 +1,23 @@
+# Registry scan review notes
+
+Publication and Registry approval are separate steps. Check the [public version API, including scan reasons](https://api.comfy.org/nodes/adachi-inpaint-workbench/versions?include_status_reason=true) after publishing **2.0.0**; a successful upload does not establish an `Active` version or Manager availability. The publisher cannot change the Registry review status through this release workflow.
+
+## Previous scan findings
+
+On 2026-11-10, **2.0.0** is also `NodeVersionStatusFlagged`, with extraction `pending`. Its six `info` records match the configuration read, `Function.bind()` in `web/compare_node.js`, `web/inline_crop.js`, `web/result_recovery.js` and `web/workspace_ui.js`, and the upload send in `web/upload_transport.js`. The same code-purpose explanations below apply; no official approval is claimed.
+
+On 2026-11-09, versions **1.0.0** and **1.0.1** were `NodeVersionStatusFlagged`, with `comfy_node_extract_status: pending`. Their public scan reasons contained the following three `info` findings. Line numbers refer to those releases.
+
+| File / line | Reported rule | Purpose and assessment |
+| --- | --- | --- |
+| `preview_store.py:12` | `python_environment_manipulation` | Reads `ADACHI_COMPARE_MEMORY_MIB` to bound the in-memory image cache. It reads one documented setting and does not modify environment variables. We infer that the rule is flagging a benign configuration read. |
+| `web/inline_crop.js:47` (also 55, 92 and 93) | `python_network_operations` | JavaScript `Function.bind()` retains the editor object as `this` when wrapping upload, flush and pointer handlers. This does not open a network socket; we infer a cross-language pattern false positive. |
+| `web/upload_transport.js:15` | `python_network_operations` | `XMLHttpRequest.send()` uploads a user-selected image to ComfyUI's own `/upload/image` route through its API URL, allowing real upload progress. Network activity here is intentional; we infer that the finding reflects required functionality rather than data exfiltration. |
+
+These assessments describe the matched code, not an official security clearance or a comprehensive audit. Required functionality remains readable; no code is obfuscated to avoid scanning. Development tests and review documents are excluded from the Registry archive by `.comfyignore`.
+
+## Release checks
+
+- Inspect the new version's actual `status` and any `status_reason`; do not assume the previous findings describe a new release.
+- If flagged, retain the version and artifact identifiers and provide the relevant code explanation for manual review. Registry approval remains an external decision.
+- Follow the [official Registry quality and security standards](https://docs.comfy.org/registry/standards), including the prohibitions on code obfuscation, `eval` / `exec` and runtime package installation.

tests/test_backend.py

@@ -61,18 +61,17 @@ class BackendTests(unittest.TestCase):
     def test_comparison_previews_and_result_passthrough(self):
         image, _ = source()
         result = image.repeat_interleave(3, 0)
-        calls = []
-        class Preview:
-            def save_images(self, images, filename_prefix):
-                calls.append((images.clone(), filename_prefix))
-                return {"ui": {"images": [{"filename": f"{filename_prefix}_{i}.png", "subfolder": "", "type": "temp"} for i in range(len(images))]}}
-        with patch.dict(sys.modules, {"nodes": types.SimpleNamespace(PreviewImage=Preview)}):
-            output = backend.AdachiImageCompare().compare(image, result)
+        output = backend.AdachiImageCompare().compare(image, result, unique_id="test")
         self.assertTrue(torch.equal(output["result"][0], result))
         self.assertEqual(len(output["ui"]["adachi_originals"]), 1)
         self.assertEqual(len(output["ui"]["adachi_results"]), 3)
-        self.assertTrue(torch.equal(calls[0][0], image))
-        self.assertTrue(torch.equal(calls[1][0], result))
+        for preview in output["ui"]["adachi_originals"] + output["ui"]["adachi_results"]:
+            self.assertEqual(preview["type"], "adachi_memory")
+            raw = backend.PREVIEWS.get(preview["filename"][:-4])
+            with Image.open(io.BytesIO(raw)) as png:
+                actual = np.asarray(png)
+            expected = np.clip(255 * image[0].numpy(), 0, 255).astype(np.uint8)
+            np.testing.assert_array_equal(actual, expected)
 
     def test_comparison_rejects_crop_and_incompatible_batch(self):
         image, _ = source(batch=2)
@@ -177,13 +176,13 @@ class BackendTests(unittest.TestCase):
     def test_embedded_source_and_mask_convention(self):
         rgba = Image.new("RGBA", (8, 4), (10, 20, 30, 255))
         rgba.putpixel((0, 0), (10, 20, 30, 0))
-        pixels, alpha_mask = backend.AdachiInlineMask().load(source_data=data_url(rgba))
+        pixels, alpha_mask, _ = backend.AdachiInlineMask().load(source_data=data_url(rgba))
         self.assertEqual(tuple(pixels.shape), (1, 4, 8, 3))
         self.assertEqual(float(alpha_mask[0, 0, 0]), 1)
         self.assertEqual(float(alpha_mask[0, 1, 1]), 0)
         blackwhite = Image.new("L", (8, 4), 0)
         blackwhite.putpixel((2, 1), 255)
-        _, explicit_mask = backend.AdachiInlineMask().load(source_data=data_url(rgba), mask_data=data_url(blackwhite))
+        _, explicit_mask, _ = backend.AdachiInlineMask().load(source_data=data_url(rgba), mask_data=data_url(blackwhite))
         self.assertEqual(float(explicit_mask[0, 1, 2]), 1)
         self.assertEqual(float(explicit_mask[0, 0, 0]), 0)
 

tests/test_crop_dimensions.mjs

@@ -0,0 +1,23 @@
+import assert from 'node:assert/strict';
+import {syncCropDimensions,syncCurrentDimensions} from '../web/parameter_bridge.js';
+const nodes=new Map(),links={},graph={links,getNodeById:id=>nodes.get(id),_nodes:[]};
+const source={id:1,type:'AdachiInlineMask',graph};nodes.set(1,source);graph._nodes.push(source);
+const make=(id,type,linkedWidth=false)=>{
+ const n={id,type,graph,inputs:[{name:'image',link:id},...(linkedWidth?[{name:'width',link:999}]:[])],widgets:[{name:'width',value:640},{name:'height',value:640}]};
+ nodes.set(id,n);graph._nodes.push(n);links[id]={origin_id:1};return n;
+};
+const sampler=make(2,'AdachiWebUIInpaint'),prepare=make(3,'AdachiInpaintPrepare',true);
+syncCropDimensions(source,1024,1536);
+assert.deepEqual(sampler.widgets.map(w=>w.value),[1024,1536]);assert.deepEqual(prepare.widgets.map(w=>w.value),[640,1536]);
+syncCropDimensions(source,9000,17);assert.deepEqual(sampler.widgets.map(w=>w.value),[8192,64]);
+source.adachiEditor={ready:true,image:{width:1001,height:1551}};
+source.adachiCrop={mode:'mask',region:{locked:false}};
+syncCurrentDimensions(source,sampler);
+assert.deepEqual(sampler.widgets.map(w=>w.value),[1000,1552]);
+assert.deepEqual(prepare.widgets.map(w=>w.value),[640,64]);
+source.adachiCrop={mode:'crop',region:{locked:false,width:512,height:768}};
+syncCurrentDimensions(source);assert.deepEqual(sampler.widgets.map(w=>w.value),[1000,1552]);
+source.adachiCrop.region.locked=true;
+syncCurrentDimensions(source);assert.deepEqual(sampler.widgets.map(w=>w.value),[512,768]);
+assert.deepEqual(prepare.widgets.map(w=>w.value),[640,768]);
+console.log('Resolution sync: source/crop modes, connection target, linked-input protection and alignment passed');

tests/test_crop_geometry.mjs

@@ -0,0 +1,29 @@
+import assert from "node:assert/strict";
+import { CROP_PRESETS, cropAt } from "../web/crop_geometry.js";
+assert.ok(CROP_PRESETS.length <= 6);
+assert.ok(CROP_PRESETS.some(([w,h]) => w === 512 && h === 512));
+assert.ok(CROP_PRESETS.some(([w,h]) => w === 2048 && h === 2048));
+assert.ok(CROP_PRESETS.every(([w, h]) => w % 64 === 0 && h % 64 === 0 && Math.min(w,h) >= 512 && Math.max(w,h) <= 2048));
+assert.deepEqual(cropAt(2048, 3072, 1024, 1536, 400, 700), {
+  x: 400,
+  y: 700,
+  width: 1024,
+  height: 1536,
+});
+assert.deepEqual(cropAt(2048, 3072, 1024, 1536, 2040, 3070), {
+  x: 1024,
+  y: 1536,
+  width: 1024,
+  height: 1536,
+});
+assert.equal(cropAt(512, 512, 1024, 1536, 0, 0), null);
+assert.equal(cropAt(2048, 3072, 1025, 1536, 0, 0), null);
+assert.equal(cropAt(2048, 3072, 1024, 1536, NaN, 0), null);
+for (const x of [-100, 0, 12.9, 1000, 3000])
+  for (const y of [-1, 0, 2200, 4000]) {
+    const r = cropAt(2048, 3072, 1024, 1536, x, y);
+    assert.ok(
+      r.x >= 0 && r.y >= 0 && r.x + r.width <= 2048 && r.y + r.height <= 3072,
+    );
+  }
+console.log("Crop presets, source coordinates and edge clamping passed");

tests/test_crop_workbench.py

@@ -0,0 +1,84 @@
+import base64
+import importlib.util
+import io
+import json
+from pathlib import Path
+import tempfile
+import types
+import unittest
+from unittest.mock import patch
+import numpy as np
+from PIL import Image
+import torch
+
+spec=importlib.util.spec_from_file_location('inline_crop_tests',Path(__file__).resolve().parents[1]/'backend.py')
+backend=importlib.util.module_from_spec(spec);spec.loader.exec_module(backend)
+def url(image):
+    b=io.BytesIO();image.save(b,format='PNG');return 'data:image/png;base64,'+base64.b64encode(b.getvalue()).decode()
+
+class InlineCropTests(unittest.TestCase):
+    def setUp(self):
+        pixels=np.arange(96*128*3,dtype=np.uint8).reshape(96,128,3)
+        self.source=url(Image.fromarray(pixels));self.original=torch.from_numpy(pixels.astype(np.float32)/255)[None]
+        self.region=dict(x=32,y=16,width=64,height=64,locked=True,job='region-a')
+        self.state=json.dumps(self.region)
+    def test_legacy_mask_mode_ignores_result_and_keeps_original_ports(self):
+        image,mask,full=backend.AdachiInlineMask().load(source_data=self.source,result=torch.ones(1,64,64,3))
+        self.assertTrue(torch.equal(image,self.original));self.assertTrue(torch.equal(image,full));self.assertEqual(tuple(mask.shape),(1,96,128))
+    def test_crop_source_and_generation_mask_are_sliced_at_exact_coordinates(self):
+        pixels=np.zeros((96,128),dtype=np.uint8);pixels[20:24,35:40]=255
+        image,mask,full=backend.AdachiInlineMask().load(source_data=self.source,mask_data=url(Image.fromarray(pixels)),edit_mode='crop',crop_state=self.state)
+        self.assertTrue(torch.equal(image,self.original[:,16:80,32:96]));self.assertTrue(torch.equal(full,self.original))
+        self.assertEqual(mask.sum().item(),20);self.assertEqual(mask[0,4,3].item(),1)
+    def test_layer_blend_preserves_outside_and_supports_batch_and_soft_edges(self):
+        alpha=np.zeros((64,64),dtype=np.uint8);alpha[:,:32]=255;alpha[:,32:48]=128
+        result=torch.stack([torch.zeros(64,64,3),torch.ones(64,64,3)])
+        blended=backend._blend_layer(self.original,result,url(Image.fromarray(alpha)),self.region)
+        self.assertEqual(tuple(blended.shape),(2,96,128,3))
+        outside=torch.ones(96,128,dtype=torch.bool);outside[16:80,32:96]=False
+        for i in range(2):
+            self.assertTrue(torch.equal(blended[i][outside],self.original[0][outside]))
+            self.assertTrue(torch.equal(blended[i,16:80,32:64],result[i,:,:32]))
+            self.assertTrue(torch.equal(blended[i,16:80,80:96],self.original[0,16:80,80:96]))
+            self.assertTrue(torch.allclose(blended[i,16:80,64:80],self.original[0,16:80,64:80]*(127/255)+result[i,:,32:48]*(128/255)))
+    def test_result_sink_persists_only_one_content_addressed_layer(self):
+        with tempfile.TemporaryDirectory() as folder,patch.dict('sys.modules',{'folder_paths':types.SimpleNamespace(get_input_directory=lambda:folder)}):
+            args=dict(result=torch.ones(2,64,64,3),source_data=self.source,crop_state=self.state,target='42')
+            a=backend.AdachiLayerCommit().compose(**args);b=backend.AdachiLayerCommit().compose(**args)
+            self.assertEqual(a['ui'],b['ui']);self.assertEqual(a['ui']['adachi_layer'][0]['target'],'42')
+            self.assertEqual(a['ui']['adachi_layer'][0]['job'],'region-a')
+            self.assertEqual(len(list(Path(folder).rglob('*.png'))),1)
+            descriptor=a['ui']['adachi_layer'][0]['image'];name=descriptor['subfolder']+'/'+descriptor['filename']
+            image,_,full=backend.AdachiInlineMask().load(source_data=self.source,edit_mode='crop',crop_state=self.state,layer_image=name)
+            self.assertTrue(torch.equal(image,torch.ones(1,64,64,3)));self.assertTrue(torch.equal(full,a['result'][0][:1]))
+    def test_invalid_regions_or_result_sizes_fail_before_writing(self):
+        for change in ({'x':-1},{'y':80},{'width':65},{'height':0},{'x':True}):
+            with self.subTest(change=change),self.assertRaises(ValueError):
+                backend._blend_layer(self.original,torch.ones(1,64,64,3),'',{**self.region,**change})
+        with patch.object(backend,'_store_layer') as store:
+            with self.assertRaisesRegex(ValueError,'must match'):
+                backend.AdachiLayerCommit().compose(torch.ones(1,32,64,3),source_data=self.source,crop_state=self.state)
+            store.assert_not_called()
+    def test_mask_mode_removes_feedback_even_when_connected(self):
+        data={'prompt':{'1':{'class_type':'AdachiInlineMask','inputs':{'result':['2',0]}},'2':{'class_type':'ImageInvert','inputs':{'image':['1',0]}}}}
+        backend.normalize_exclusive_prompt(data)
+        self.assertNotIn('result',data['prompt']['1']['inputs']);self.assertEqual(len(data['prompt']),2)
+    def test_crop_mode_has_no_result_socket_or_automatic_sink(self):
+        self.assertNotIn('result', backend.AdachiInlineMask.INPUT_TYPES()['optional'])
+        data={'prompt':{'1':{'class_type':'AdachiInlineMask','inputs':{'result':['2',0],'edit_mode':'crop','crop_state':self.state}},'2':{'class_type':'ImageInvert','inputs':{'image':['1',0]}},'3':{'class_type':'SaveImage','inputs':{'images':['1',2]}}}}
+        backend.normalize_exclusive_prompt(data)
+        self.assertEqual(len(data['prompt']),3)
+        self.assertNotIn('result',data['prompt']['1']['inputs'])
+        self.assertEqual(data['prompt']['3']['inputs']['images'],['1',2])
+    def test_comparison_origin_tracks_exact_generating_region(self):
+        prompt={'7':{'class_type':'AdachiInlineMask','inputs':{'edit_mode':'crop','crop_state':self.state}},'8':{'class_type':'Reroute','inputs':{'image':['7',0]}},'9':{'class_type':'AdachiImageCompare','inputs':{'original':['8',0]}}}
+        self.assertEqual(backend._comparison_origin(prompt,'9'),{'source':'7','mode':'crop','job':'region-a'})
+        prompt['7']['inputs']['crop_state']=json.dumps({**self.region,'job':'next-region'})
+        self.assertEqual(backend._comparison_origin(prompt,'9')['job'],'next-region')
+        self.assertIsNone(backend._comparison_origin({},'9'))
+    def test_fingerprint_tracks_mode_blend_and_layer(self):
+        first=backend.AdachiInlineMask.IS_CHANGED(source_data=self.source,edit_mode='mask')
+        for extra in ({'edit_mode':'crop'},{'blend_data':'new'},{'layer_image':'changed.png'}):
+            self.assertNotEqual(first,backend.AdachiInlineMask.IS_CHANGED(source_data=self.source,**extra))
+
+if __name__=='__main__':unittest.main()

tests/test_draft_guard.mjs

@@ -0,0 +1,118 @@
+import assert from "node:assert/strict";
+import { canStoreDraft, prepareInlineDraft, syncInlineWidgets, inlineWidgetNames } from "../web/draft_guard.js";
+class Storage {
+  constructor(limit = 10000) {
+    this.limit = limit;
+    this.data = new Map();
+    this.removed = [];
+  }
+  getItem(k) {
+    return this.data.get(k) ?? null;
+  }
+  setItem(k, v) {
+    const used = [...this.data].reduce(
+      (n, [key, value]) => n + (key === k ? 0 : value.length),
+      0,
+    );
+    if (used + v.length > this.limit)
+      throw Object.assign(new Error("full"), { name: "QuotaExceededError" });
+    this.data.set(k, v);
+  }
+  removeItem(k) {
+    this.removed.push(k);
+    this.data.delete(k);
+  }
+}
+const storage = new Storage();
+storage.setItem("unrelated", "keep");
+assert.equal(canStoreDraft({ nodes: [] }, { storage }), true);
+assert.equal(canStoreDraft({ data: "x".repeat(20000) }, { storage }), false);
+assert.equal(storage.getItem("unrelated"), "keep");
+assert.ok(storage.removed.every((key) => key.startsWith("Adachi.DraftProbe:")));
+assert.equal(storage.data.size, 1);
+// Replacing a draft must reuse its existing space, rather than testing two full copies.
+const path = "workflows/Test.json";
+let hash = 2166136261;
+for (const c of path) hash = Math.imul(hash ^ c.charCodeAt(0), 16777619);
+const payloadKey = `Comfy.Workflow.Draft.v2:personal:${(hash >>> 0).toString(16).padStart(8, "0")}`;
+const oldGraph = { image: "x".repeat(6000) },
+  oldRaw = JSON.stringify({
+    data: JSON.stringify(oldGraph),
+    updatedAt: Date.now(),
+  });
+storage.setItem(
+  "Comfy.Workflow.DraftIndex.v2:personal",
+  JSON.stringify({ entries: { one: { path } } }),
+);
+storage.setItem(payloadKey, oldRaw);
+assert.equal(
+  canStoreDraft({ ...oldGraph, mode: "crop" }, { path, storage }),
+  true,
+);
+assert.equal(storage.getItem(payloadKey), oldRaw);
+assert.equal(
+  canStoreDraft({ image: "x".repeat(12000) }, { path, storage }),
+  false,
+);
+assert.equal(storage.getItem(payloadKey), oldRaw);
+assert.equal(
+  canStoreDraft(
+    {},
+    {
+      storage: {
+        getItem() {
+          throw Error("blocked");
+        },
+        setItem() {
+          throw Object.assign(Error("blocked"), { name: "SecurityError" });
+        },
+        removeItem() {},
+      },
+    },
+  ),
+  null,
+);
+console.log(
+  "Draft guard: quota detection, replacement credit, probe cleanup and preservation of existing storage passed",
+);
+
+const large = "data:image/png;base64," + "a".repeat(300000);
+const node = () => ({ type: "AdachiInlineMask", widgets_values: ["old.png", "mask", large, "crop", "region", "blend", "layer.png"], widgets_values_named: { image: "stale.png", mask_data: "stale-mask", source_data: large, edit_mode: "mask" } });
+const first = node(), nested = node();
+const graph = { nodes: [first, {type:"Other", widgets_values:[large]}], definitions: {subgraphs:[{nodes:[nested]}]} };
+const original = structuredClone(graph);
+let uploads = 0;
+assert.equal(await prepareInlineDraft(graph, { fits: g => g.nodes[0].widgets_values[2].length < 100 && g.definitions.subgraphs[0].nodes[0].widgets_values[2].length < 100, upload: async bytes => { assert.equal(bytes,large); return `source-${++uploads}.png`; } }),2);
+assert.equal(uploads,2);
+for (const n of [first,nested]) {
+  assert.equal(n.widgets_values[2],"");
+  assert.equal(n.widgets_values_named.source_data,"");
+  assert.equal(n.widgets_values_named.image,n.widgets_values[0]);
+  assert.equal(n.widgets_values_named.mask_data,"mask");
+  assert.deepEqual(n.widgets_values.slice(3),["crop","region","blend","layer.png"]);
+}
+assert.deepEqual(graph.nodes[1],original.nodes[1]);
+for (const fits of [() => true, () => null]) {
+  const g = {nodes:[node()]};
+  assert.equal(await prepareInlineDraft(g,{fits,upload:()=>{throw Error("unexpected upload")}}),0);
+  assert.equal(g.nodes[0].widgets_values[2],large);
+}
+const failure = {nodes:[node()]};
+await assert.rejects(prepareInlineDraft(failure,{fits:()=>false,upload:async()=>{throw Error("upload failed")}}),/upload failed/);
+assert.equal(failure.nodes[0].widgets_values[2],large);
+assert.equal(failure.nodes[0].widgets_values[0],"old.png");
+const namedOnly = {nodes:[{type:"AdachiInlineMask", widgets_values_named:{source_data:large,mask_data:"keep"}}]};
+await prepareInlineDraft(namedOnly,{fits:()=>false,upload:async()=>"named.png"});
+assert.equal(namedOnly.nodes[0].widgets_values_named.source_data,"");
+assert.equal(namedOnly.nodes[0].widgets_values_named.mask_data,"keep");
+const widgets=inlineWidgetNames.map((name,index)=>({name,value:`live-${index}`}));
+widgets.splice(1,0,{name:"unrelated",value:"untouched"});
+const snapshot={widgets_values:Array(8).fill("old"),widgets_values_named:{source_data:large,unrelated:"keep"}};
+syncInlineWidgets(snapshot,widgets);
+assert.equal(snapshot.widgets_values[3],"live-2");
+assert.equal(snapshot.widgets_values[1],"old");
+assert.equal(snapshot.widgets_values_named.source_data,"live-2");
+assert.equal(snapshot.widgets_values_named.unrelated,"keep");
+const arrayOnly={widgets_values:[]}; syncInlineWidgets(arrayOnly,widgets);
+assert.equal(arrayOnly.widgets_values_named,undefined);
+console.log("Draft preflight: root/subgraph sources, dual snapshots, unchanged masks/layers, fitting sources and failed uploads passed");

tests/test_i18n.mjs

@@ -0,0 +1,55 @@
+import assert from 'node:assert/strict';
+import {T, localeDict} from '../web/i18n.js';
+import {WorkspaceProgress} from '../web/workspace_progress.js';
+
+const saved = Object.fromEntries(['comfyAPI','app','navigator'].map(key => [key,Object.getOwnPropertyDescriptor(globalThis,key)]));
+const settings = getLocale => ({ui:{settings:{getSettingValue(key) {
+    assert.equal(key,'Comfy.Locale');
+    return getLocale();
+}}}});
+try {
+    delete globalThis.comfyAPI; delete globalThis.app;
+    Object.defineProperty(globalThis,'navigator',{configurable:true,value:{language:'zh-CN',languages:['zh-CN']}});
+    assert.equal(T('Adopt'),'Adopt','English remains the default even for a Chinese-language browser');
+    assert.equal(localeDict(),null);
+    assert.equal(T('Missing {count} model files',{count:0}),'Missing 0 model files');
+
+    let locale='en-US';
+    globalThis.comfyAPI={app:{app:settings(()=>locale)}};
+    globalThis.app=settings(()=>'zh');
+    assert.equal(T('Open Adachi Workbench'),'Open Adachi Workbench','the current ComfyUI API takes priority over legacy globals');
+    locale='zh-CN';
+    assert.equal(T('Open Adachi Workbench'),'打开安达工作台');
+    assert.equal(T('Missing {count} model files',{count:3}),'缺少 3 份模型文件');
+    assert.equal(T('Settings panel {width} pixels wide',{width:360}),'右侧配置 360 像素');
+    assert.equal(T('Fill','enum'),'填充','existing enum contexts remain compatible');
+    assert.equal(T('New message {count}',{count:false}),'New message false','untranslated strings still interpolate');
+    assert.equal(T('Unknown {value} / {missing}',{value:'<img src=x>'}),'Unknown <img src=x> / {missing}','values remain literal text and missing placeholders remain visible');
+    assert.equal(T('toString'),'toString','dictionary prototype properties are not translations');
+
+    locale='zh_TW';
+    assert.equal(T('Workflow connections'),'工作流连接','regional locale separators both resolve to the Chinese dictionary');
+    const progress=new WorkspaceProgress();
+    progress.begin({'7':{class_type:'AdachiWebUIInpaint'}},'task');
+    progress.handle('progress',{prompt_id:'task',node:'7',value:2,max:10});
+    assert.equal(progress.label,'重绘采样（K 采样器） · 2 / 10');
+    locale='en';
+    progress.handle('progress',{prompt_id:'task',node:'7',value:3,max:10});
+    assert.equal(progress.label,'Inpaint sampling (KSampler) · 3 / 10','execution messages use the current setting rather than an import-time locale');
+    assert.equal(T('Adopt'),'Adopt','locale changes do not require module reload');
+    locale='fr';
+    assert.equal(T('Adopt'),'Adopt','unsupported locales fall back to English');
+
+    globalThis.comfyAPI={app:settings(()=>'zh')}; delete globalThis.app;
+    assert.equal(T('Adopt'),'采用','older ComfyUI API shape is supported');
+    globalThis.comfyAPI={app:{app:settings(()=>{throw Error('not initialized');})}};
+    globalThis.app=settings(()=>'zh');
+    assert.equal(T('Adopt'),'采用','a not-yet-ready modern setting can fall back to the legacy app');
+    delete globalThis.comfyAPI; delete globalThis.app;
+    assert.equal(T('Adopt'),'Adopt','an unavailable app does not break custom UI initialization');
+} finally {
+    for(const [key,descriptor] of Object.entries(saved)) {
+        if(descriptor)Object.defineProperty(globalThis,key,descriptor);else delete globalThis[key];
+    }
+}
+console.log('Localization: ComfyUI settings precedence, English fallback, live locale changes, placeholders, regional locales and enum compatibility pass');

tests/test_legacy_feedback.mjs

@@ -0,0 +1,11 @@
+import assert from 'node:assert/strict';
+import {removeLegacyFeedback} from '../web/legacy_feedback.js';
+const make=()=>({nodes:[{id:1,type:'AdachiInlineMask',inputs:[{name:'result',link:10},{name:'image',link:11}],outputs:[]},{id:2,type:'Source',inputs:[],outputs:[{links:[10,11]}]}],links:[[10,2,0,1,0,'IMAGE'],[11,2,0,1,1,'STRING']]});
+const graph=make();graph.definitions={subgraphs:[make()]};removeLegacyFeedback(graph);
+for(const g of [graph,...graph.definitions.subgraphs]){
+ assert.deepEqual(g.nodes[0].inputs,[{name:'image',link:11}]);assert.deepEqual(g.links,[[11,2,0,1,0,'STRING']]);assert.deepEqual(g.nodes[1].outputs[0].links,[11]);
+}
+removeLegacyFeedback(graph);assert.equal(graph.links.length,1);
+const modern=make();modern.links=modern.links.map(([id,origin_id,origin_slot,target_id,target_slot,type])=>({id,origin_id,origin_slot,target_id,target_slot,type}));removeLegacyFeedback(modern);assert.equal(modern.links.length,1);assert.equal(modern.links[0].target_slot,0);
+const orphan=make();orphan.nodes[0].inputs=[{name:'image',link:null}];removeLegacyFeedback(orphan);assert.equal(orphan.links.length,0);
+console.log('Legacy feedback migration: tuple/object links, slot reindexing, nested graphs, orphan cleanup and idempotence passed');

tests/test_mask_boundaries.py

@@ -0,0 +1,34 @@
+"""Image-edge and dense-mask equivalence for the bounded-memory preparation path."""
+import unittest
+import torch
+from test_backend import backend
+
+
+class MaskBoundaryTests(unittest.TestCase):
+    def test_dense_sparse_empty_and_skinny_mask_bounds(self):
+        for height, width in [(1, 1), (1, 79), (81, 1), (192, 256)]:
+            for kind in ('empty', 'full', 'corners', 'middle'):
+                mask = torch.zeros(height, width)
+                if kind == 'full': mask.fill_(1)
+                elif kind == 'corners': mask[0, 0] = mask[-1, -1] = .01
+                elif kind == 'middle': mask[height//2, width//2] = .7
+                for padding in (0, 7, 2048):
+                    box = backend._bbox(mask, padding, width/height)
+                    x0, y0, x1, y1 = box
+                    self.assertTrue(0 <= x0 < x1 <= width)
+                    self.assertTrue(0 <= y0 < y1 <= height)
+                    self.assertEqual(float(mask.sum()), float(mask[y0:y1, x0:x1].sum()))
+                    if kind in ('full', 'corners', 'empty'):
+                        self.assertEqual(box, (0, 0, width, height))
+
+    def test_sanitizing_never_mutates_upstream_tensors(self):
+        image = torch.tensor([float('nan'), float('inf'), -1.0, 2.0]*3).reshape(1, 2, 2, 3)
+        original = image.clone()
+        output = backend._image_tensor(image)
+        self.assertTrue(torch.equal(torch.isnan(original), torch.isnan(image)))
+        self.assertTrue(torch.equal(original[~torch.isnan(original)], image[~torch.isnan(image)]))
+        self.assertTrue(torch.isfinite(output).all())
+        self.assertTrue(((output >= 0) & (output <= 1)).all())
+        mask = torch.full((1, 2, 2), 2.0)
+        self.assertEqual(float(backend._mask_tensor(mask, output).max()), 1)
+        self.assertEqual(float(mask.min()), 2)

tests/test_official_workflow.py

@@ -0,0 +1,131 @@
+"""The public template stays portable and routes each preset independently."""
+import json
+from pathlib import Path
+import re
+import unittest
+
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+class OfficialWorkflowTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.path = ROOT / "examples" / "Adachi-Workbench.json"
+        cls.raw = cls.path.read_text(encoding="utf-8")
+        cls.workflow = json.loads(cls.raw)
+        cls.nodes = {node["id"]: node for node in cls.workflow["nodes"]}
+        cls.links = {link[0]: link for link in cls.workflow["links"]}
+        cls.groups = {group["id"]: group for group in cls.workflow["groups"]}
+
+    def typed(self, name):
+        return [node for node in self.nodes.values() if node["type"] == name]
+
+    def group_members(self, group):
+        x, y, width, height = group["bounding"]
+        return [node for node in self.nodes.values()
+                if x <= node["pos"][0] and y <= node["pos"][1]
+                and node["pos"][0] + node["size"][0] <= x + width
+                and node["pos"][1] + node["size"][1] <= y + height]
+
+    def test_one_complete_official_template(self):
+        self.assertEqual([self.path.name], sorted(p.name for p in (ROOT / "examples").glob("*.json")))
+        for name in ("CheckpointLoaderSimple", "EmptyLatentImage", "KSampler", "VAEDecode",
+                     "SaveImage", "AdachiInlineMask", "AdachiWebUIInpaint", "AdachiImageCompare",
+                     "AdachiGroupPresets"):
+            self.assertEqual(len(self.typed(name)), 1, name)
+        self.assertEqual(len(self.typed("AdachiExclusiveSwitch")), 3)
+        self.assertEqual(len(self.typed("CLIPTextEncode")), 4)
+        allowed = {"CheckpointLoaderSimple", "EmptyLatentImage", "KSampler", "VAEDecode", "SaveImage",
+                   "CLIPTextEncode", "Note", "AdachiInlineMask", "AdachiWebUIInpaint",
+                   "AdachiImageCompare", "AdachiGroupPresets", "AdachiExclusiveSwitch"}
+        self.assertTrue({node["type"] for node in self.nodes.values()} <= allowed)
+
+    def test_ids_and_reciprocal_link_types(self):
+        self.assertEqual(len(self.nodes), len(self.workflow["nodes"]))
+        self.assertEqual(len(self.links), len(self.workflow["links"]))
+        self.assertEqual(self.workflow["last_node_id"], max(self.nodes))
+        self.assertEqual(self.workflow["last_link_id"], max(self.links))
+        for key, source, out, target, inp, kind in self.links.values():
+            origin = self.nodes[source]["outputs"][out]
+            destination = self.nodes[target]["inputs"][inp]
+            self.assertIn(key, origin["links"])
+            self.assertEqual(destination["link"], key)
+            for socket in (origin, destination):
+                self.assertTrue(kind == socket["type"] or "*" in (kind, socket["type"]),
+                                (key, kind, socket))
+        for node in self.nodes.values():
+            for index, socket in enumerate(node.get("inputs", [])):
+                if socket.get("link") is not None:
+                    self.assertEqual(self.links[socket["link"]][3:5], [node["id"], index])
+            for index, socket in enumerate(node.get("outputs", [])):
+                for key in socket.get("links") or []:
+                    self.assertEqual(self.links[key][1:3], [node["id"], index])
+
+    def test_presets_enable_one_branch_and_keep_common_nodes_enabled(self):
+        controller = self.typed("AdachiGroupPresets")[0]
+        state = controller["properties"]["adachi_group_presets"]
+        self.assertEqual(state["activeId"], "text-to-image")
+        self.assertEqual({preset["id"] for preset in state["presets"]}, {"text-to-image", "inpaint"})
+        membership = {}
+        for key, group in self.groups.items():
+            for node in self.group_members(group):
+                self.assertNotIn(node["id"], membership, "Mode groups must not overlap")
+                membership[node["id"]] = key
+        for node in self.nodes.values():
+            if node["type"] in ("AdachiExclusiveSwitch", "CheckpointLoaderSimple", "SaveImage", "AdachiGroupPresets"):
+                self.assertNotIn(node["id"], membership, node["type"])
+        for preset in state["presets"]:
+            self.assertEqual({value["mode"] for value in preset["groups"].values()}, {0, 4})
+            modes = {node["id"]: preset["groups"][f"group:{membership[node['id']]}"]["mode"]
+                     if node["id"] in membership else node["mode"] for node in self.nodes.values()}
+            expected = 1 if preset["id"] == "text-to-image" else 2
+            for switch in self.typed("AdachiExclusiveSwitch"):
+                active = sum((1 << index) for index, socket in enumerate(switch["inputs"])
+                             if modes[self.links[socket["link"]][1]] == 0)
+                self.assertEqual(active, expected, (preset["id"], switch["title"]))
+            if preset["id"] == state["activeId"]:
+                for node in self.nodes.values():
+                    self.assertEqual(node["mode"], modes[node["id"]], node["title"])
+
+    def test_no_private_source_or_locale_is_bundled(self):
+        self.assertEqual(self.typed("CheckpointLoaderSimple")[0]["widgets_values"], ["YOUR_CHECKPOINT.safetensors"])
+        self.assertIsNone(re.search(r"[\u3400-\u9fff]", self.raw), "Template text must be English")
+        self.assertIsNone(re.search(r"[A-Za-z]:[\\/]|data:image|https?://|/Users/|/root/", self.raw))
+        inline = self.typed("AdachiInlineMask")[0]
+        self.assertEqual(inline["widgets_values"][:7], ["", "", "", "mask", "{}", "", ""])
+        self.assertEqual(set(self.workflow["extra"]), {"ds"})
+
+    def test_saving_text_to_image_does_not_depend_on_an_uploaded_image(self):
+        controller = self.typed("AdachiGroupPresets")[0]
+        presets = controller["properties"]["adachi_group_presets"]["presets"]
+        members = {node["id"]: group["id"] for group in self.groups.values()
+                   for node in self.group_members(group)}
+        for preset in presets:
+            enabled = {key for key, node in self.nodes.items()
+                       if (preset["groups"][f"group:{members[key]}"]["mode"]
+                           if key in members else node["mode"]) == 0}
+            seen, todo = set(), [self.typed("SaveImage")[0]["id"]]
+            while todo:
+                key = todo.pop()
+                if key in seen:
+                    continue
+                self.assertIn(key, enabled, "Active output must not use a disabled branch")
+                seen.add(key)
+                node = self.nodes[key]
+                sockets = node.get("inputs", [])
+                if node["type"] == "AdachiExclusiveSwitch":
+                    sockets = [socket for socket in sockets if self.links[socket["link"]][1] in enabled]
+                    self.assertEqual(len(sockets), 1)
+                todo.extend(self.links[socket["link"]][1] for socket in sockets if socket.get("link") is not None)
+            types = {self.nodes[key]["type"] for key in seen}
+            if preset["id"] == "text-to-image":
+                self.assertTrue({"KSampler", "VAEDecode", "EmptyLatentImage"} <= types)
+                self.assertTrue({"AdachiInlineMask", "AdachiWebUIInpaint", "AdachiImageCompare"}.isdisjoint(types))
+            else:
+                self.assertTrue({"AdachiInlineMask", "AdachiWebUIInpaint", "AdachiImageCompare"} <= types)
+                self.assertTrue({"KSampler", "VAEDecode", "EmptyLatentImage"}.isdisjoint(types))
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_preview_cleanup.py

@@ -0,0 +1,48 @@
+"""Regression checks for comparison preview cleanup boundaries."""
+
+import importlib.util
+import os
+from pathlib import Path
+import tempfile
+import unittest
+
+SPEC = importlib.util.spec_from_file_location("preview_cleanup_under_test", Path(__file__).resolve().parents[1] / "preview_cleanup.py")
+cleanup = importlib.util.module_from_spec(SPEC)
+SPEC.loader.exec_module(cleanup)
+
+
+class PreviewCleanupTests(unittest.TestCase):
+    def test_only_expired_comparison_previews_in_temp_are_removed(self):
+        now = 2_000_000
+        with tempfile.TemporaryDirectory() as root:
+            temp = Path(root) / "temp"
+            inputs = Path(root) / "input"
+            temp.mkdir()
+            inputs.mkdir()
+            old_original = temp / "adachi_compare_original_temp_abcde_00001_.png"
+            old_result = temp / "adachi_compare_result_temp_abcde_00002_.png"
+            recent = temp / "adachi_compare_result_temp_abcde_00003_.png"
+            unrelated = temp / "ComfyUI_temp_abcde_00001_.png"
+            wrong_name = temp / "adachi_compare_result_mine.png"
+            adopted = inputs / old_original.name
+            nested = temp / "nested"
+            nested.mkdir()
+            nested_preview = nested / old_original.name
+            for path in (old_original, old_result, recent, unrelated, wrong_name, adopted, nested_preview):
+                path.write_bytes(b"test")
+                os.utime(path, (now - 100_000, now - 100_000))
+            os.utime(recent, (now - 60, now - 60))
+            removed = cleanup.cleanup_comparison_previews(temp, now=now)
+            self.assertEqual(removed, 2)
+            for path in (old_original, old_result):
+                self.assertFalse(path.exists())
+            for path in (recent, unrelated, wrong_name, adopted, nested_preview):
+                self.assertTrue(path.exists())
+
+    def test_missing_directory_is_harmless(self):
+        with tempfile.TemporaryDirectory() as root:
+            self.assertEqual(cleanup.cleanup_comparison_previews(Path(root) / "missing"), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_preview_images.mjs

@@ -0,0 +1,36 @@
+import assert from 'node:assert/strict';
+import {decodePreview,loadPreviewPair} from '../web/preview_images.js';
+const deferred=()=>{let resolve,reject;const promise=new Promise((a,b)=>{resolve=a;reject=b;});return{promise,resolve,reject};};
+const blob=new Blob(['PNG']);
+let allocated=0;const revoked=[];
+const urls={createObjectURL(){return 'blob:'+ ++allocated;},revokeObjectURL:url=>revoked.push(url)};
+class ImageStub{width=512;height=512;decode(){return Promise.resolve();}}
+const canceled=new AbortController();canceled.abort();
+await assert.rejects(decodePreview(blob,canceled.signal,{urls,ImageClass:ImageStub}),{name:'AbortError'});
+assert.equal(allocated,0,'already-canceled decoding never allocates a URL');
+class Stalled extends ImageStub{decode(){return new Promise(()=>{});}}
+await assert.rejects(decodePreview(blob,new AbortController().signal,{urls,ImageClass:Stalled,timeoutMs:5}),/timed out/);
+assert.deepEqual(revoked,['blob:1'],'a stalled decode settles and releases the URL');
+class Huge extends ImageStub{width=20000;}
+await assert.rejects(decodePreview(blob,new AbortController().signal,{urls,ImageClass:Huge}),/dimensions/);
+class Borderline extends ImageStub{width=5500;height=6000;}
+await assert.rejects(decodePreview(blob,new AbortController().signal,{urls,ImageClass:Borderline}),/dimensions/);
+const abort=new AbortController(),decode=decodePreview(blob,abort.signal,{urls,ImageClass:Stalled});abort.abort();
+await assert.rejects(decode,{name:'AbortError'});
+assert.equal(revoked.length,4);
+
+// A sibling may finish after the first failure, even if an integration ignores
+// cancellation. Wait for it and release its late URL before returning.
+const later=deferred(),lost=[];let decodes=0;
+const pending=loadPreviewPair(['a','b'],{cached:[null,null],signal:new AbortController().signal,
+ loadPreview:async()=>blob,decode:async()=>{if(++decodes===1)throw Error('corrupt PNG');return later.promise;},revoke:url=>lost.push(url)});
+await new Promise(r=>setTimeout(r,0));later.resolve({url:'late-url'});
+await assert.rejects(pending,/corrupt PNG/);assert.deepEqual(lost,['late-url']);
+
+const bad=new Blob(['bad']),good=new Blob(['good']),cache=[bad,good];let reads=0;
+const options={cached:cache,signal:new AbortController().signal,loadPreview:async()=>{reads++;return good;},
+ decode:async b=>{if(b===bad)throw Error('invalid pixels');return{url:'good-url'};},revoke:()=>{}};
+await assert.rejects(loadPreviewPair(['a','b'],options),/invalid pixels/);
+assert.equal(cache[0],null,'invalid downloaded bytes cannot poison every future retry');
+await loadPreviewPair(['a','b'],options);assert.equal(reads,1,'retry fetches the corrupt half again and keeps the valid half');
+console.log('Preview decoding: pre-cancel, decode timeout, dimension limits, late sibling URL cleanup and corrupt-cache recovery pass');

tests/test_preview_store.py

@@ -0,0 +1,74 @@
+import unittest
+from preview_store import PreviewStore
+
+
+def key(descriptor):
+    return descriptor["filename"][:-4]
+
+
+class PreviewStoreTests(unittest.TestCase):
+    def test_tiny_batches_have_a_metadata_budget_and_empty_batches_are_atomic(self):
+        store = PreviewStore(1024 * 1024, max_batches=2)
+        first = store.put_batch([b"a"], "first")
+        second = store.put_batch([b"b"], "second")
+        store.get(key(first[0]))
+        generation = store.fingerprint()
+        with self.assertRaises(ValueError):
+            store.put_batch([], "empty")
+        with self.assertRaises(ValueError):
+            store.put_batch([b""], "empty")
+        self.assertEqual(store.fingerprint(), generation)
+        store.put_batch([b"c"], "third")
+        self.assertEqual(store.get(key(first[0])), b"a")
+        self.assertIsNone(store.get(key(second[0])))
+        self.assertEqual(len(store.groups), 2)
+        self.assertEqual(store.size, 2)
+
+    def test_whole_batch_lru_and_cache_generation(self):
+        store = PreviewStore(10)
+        first = store.put_batch([b"aaa", b"bbb"], "1")
+        before = store.fingerprint()
+        second = store.put_batch([b"ccc", b"ddd"], "2")
+        self.assertNotEqual(store.fingerprint(), before)
+        self.assertIsNone(store.get(key(first[0])))
+        self.assertIsNone(store.get(key(first[1])))
+        self.assertEqual(store.get(key(second[0])), b"ccc")
+        self.assertEqual(store.get(key(second[1])), b"ddd")
+        self.assertNotEqual(store.fingerprint(), PreviewStore(10).fingerprint())
+
+    def test_eviction_rerun_reuses_current_batch_without_continuous_eviction(self):
+        store = PreviewStore(10)
+        store.put_batch([b"aaa", b"bbb"], "node")
+        current = store.put_batch([b"ccc", b"ddd"], "node")
+        after_eviction = store.fingerprint()
+        reused = store.put_batch([b"ccc", b"ddd"], "node")
+        self.assertEqual(reused, current)
+        self.assertEqual(store.fingerprint(), after_eviction)
+        # A true new execution with no intervening eviction has fresh identity.
+        regenerated = store.put_batch([b"ccc", b"ddd"], "node")
+        self.assertNotEqual(regenerated, current)
+        self.assertNotEqual(store.fingerprint(), after_eviction)
+
+    def test_oversized_batch_never_discards_prior_previews(self):
+        store = PreviewStore(6)
+        existing = store.put_batch([b"aa", b"bb"], "x")
+        generation = store.fingerprint()
+        with self.assertRaisesRegex(ValueError, "preview limit"):
+            store.put_batch([b"aaa", b"bbbb"], "x")
+        self.assertEqual(store.get(key(existing[0])), b"aa")
+        self.assertEqual(store.get(key(existing[1])), b"bb")
+        self.assertEqual(store.fingerprint(), generation)
+
+    def test_alternating_inputs_invalidates_the_old_cached_descriptors(self):
+        store = PreviewStore(6)
+        first = store.put_batch([b"aaa", b"bbb"], "node")
+        token = store.fingerprint()
+        store.put_batch([b"ccc", b"ddd"], "node")
+        self.assertNotEqual(token, store.fingerprint())
+        self.assertIsNone(store.get(key(first[0])))
+        restored = store.put_batch([b"aaa", b"bbb"], "node")
+        self.assertEqual(store.get(key(restored[0])), b"aaa")
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_preview_transport.mjs

@@ -0,0 +1,23 @@
+import assert from 'node:assert/strict';
+import {readPreview} from '../web/preview_transport.js';
+const bytes=new TextEncoder().encode('image-bytes');
+const events=[];
+const blob=await readPreview(async()=>new Response(new ReadableStream({start(c){c.enqueue(bytes.slice(0,3));c.enqueue(bytes.slice(3));c.close();}}),{headers:{'content-length':String(bytes.length),'content-type':'image/png'}}),'/preview',{onProgress:e=>events.push(e)});
+assert.equal(await blob.text(),'image-bytes');assert.equal(events[1].loaded,3);assert.equal(events[1].total,bytes.length);assert.equal(events.at(-1).complete,true);
+const unknown=[];await readPreview(async()=>new Response(bytes),'/preview',{onProgress:e=>unknown.push(e)});assert.equal(unknown[0].total,0);assert.equal(unknown.at(-1).loaded,bytes.length);
+await assert.rejects(readPreview(async()=>new Response(bytes,{headers:{'content-length':'99'}}),'/truncated'),/interrupted/);
+await assert.rejects(readPreview(async()=>new Response(bytes,{status:404}),'/expired'),e=>e.status===404);
+await assert.rejects(readPreview(async()=>new Response(bytes),'/too-large',{maxBytes:3}),/size limit/);
+const cancel=new AbortController();
+const stalled=(_,options)=>Promise.resolve(new Response(new ReadableStream({start(c){options.signal.addEventListener('abort',()=>c.error(new DOMException('Aborted','AbortError')));}})));
+await assert.rejects(readPreview(stalled,'/stalled',{idleMs:10}),/timed out/);
+const pending=readPreview(stalled,'/cancel',{signal:cancel.signal});cancel.abort();await assert.rejects(pending,e=>e.name==='AbortError');
+// A long transfer is allowed while actual data continues to arrive.
+let ticks=0;await readPreview(async(_,options)=>new Response(new ReadableStream({start(c){const t=setInterval(()=>{if(++ticks===5){clearInterval(t);c.close();}else c.enqueue(bytes);},5);options.signal.addEventListener('abort',()=>{clearInterval(t);c.error(Error('timed out'));});}})),'/slow',{idleMs:20});assert.equal(ticks,5);
+console.log('Preview transport: streamed byte progress, unknown length, truncation, expiry, size limit, inactivity timeout, cancellation and slow advancing transfer pass');
+let requests=0;
+await assert.rejects(readPreview(async()=>{requests++;return new Response(bytes);},'/never-start',{signal:AbortSignal.abort()}),{name:'AbortError'});
+assert.equal(requests,0);
+let stopped=false;
+await assert.rejects(readPreview(async()=>new Response(new ReadableStream({cancel(){stopped=true;}})),'/uncooperative-wrapper',{idleMs:5}),/timed out/);
+assert.equal(stopped,true,'cancel the response reader even when a custom fetch wrapper ignores its signal');

tests/test_result_recovery.mjs

@@ -0,0 +1,32 @@
+import assert from 'node:assert/strict';
+import {ResultRecovery} from '../web/result_recovery.js';
+const node={id:3,type:'AdachiImageCompare'},graph={getNodeById:id=>String(id)==='3'?node:null};
+let active=graph,calls=0,fail=true,output=[],terminal=[];
+const entry={status:{completed:true,status_str:'success'},outputs:{'3':{adachi_originals:[{filename:'a'}],adachi_results:[{filename:'b'}]}}};
+const r=new ResultRecovery({getGraph:()=>active,interval:60000,fetchApi:async path=>{calls++;assert.equal(path,'/history/task');if(fail)throw Error('offline');return {ok:true,json:async()=>({task:entry})};},onOutput:(n,o,id)=>output.push({n,o,id}),onTerminal:(...x)=>terminal.push(x)});
+try {
+ r.track(graph,'task','3');await r.recover();assert.equal(output.length,0);assert.equal(r.pending.size,1,'network failure retains the task even without any WebSocket events');
+ fail=false;await r.recover();assert.equal(output.length,1);assert.equal(output[0].n,node);assert.equal(r.pending.size,0);
+ r.track(graph,'task','3');await r.recover();assert.equal(output.length,1,'success cannot replay after adoption');
+ r.delivered.clear();r.track(graph,'task','3');active={getNodeById:()=>node};await r.recover();assert.equal(output.length,1,'another graph must not receive a result with the same node ID');
+ active=graph;r.track(graph,'task','3');graph.getNodeById=()=>({...node});await r.recover();assert.equal(output.length,1,'reconfigured nodes are not the queued node');graph.getNodeById=id=>String(id)==='3'?node:null;
+ r.track(graph,'task','3');entry.outputs={};entry.status.status_str='error';await r.recover();assert.equal(terminal.length,1);assert.equal(r.pending.size,0);
+ assert.ok(calls>=3);
+} finally {r.destroy();}
+console.log('Result recovery: lost notifications, offline retries, duplicate delivery, workflow/node replacement and terminal errors pass');
+
+const nested={id:8,type:'AdachiImageCompare'},child={getNodeById:id=>String(id)==='8'?nested:null},container={id:5,subgraph:child},root={getNodeById:id=>String(id)==='5'?container:null};
+let nestedResult;
+const sub=new ResultRecovery({getGraph:()=>root,fetchApi:async()=>({ok:true,json:async()=>({nestedTask:{status:{completed:true,status_str:'success'},outputs:{'5:8':{adachi_originals:[{}],adachi_results:[{}]}}}})}),onOutput:n=>{nestedResult=n;}});
+try{sub.track(root,'nestedTask','5:8');await sub.recover();assert.equal(nestedResult,nested);}finally{sub.destroy();}
+console.log('Result recovery: native subgraph execution IDs resolve to the queued live comparison');
+{
+ let resolve,signal,discarded=0,delivered=0;
+ const pending=new ResultRecovery({getGraph:()=>graph,onOutput:()=>delivered++,onDiscard:()=>discarded++,
+  fetchApi:async(_path,options)=>{signal=options.signal;return new Promise(r=>resolve=r);}});
+ pending.track(graph,'late','3');const task=pending.recover();pending.destroy();
+ assert.equal(signal.aborted,true);assert.equal(discarded,1);
+ resolve({ok:true,json:async()=>({late:{status:{completed:true},outputs:{'3':{adachi_originals:[{}],adachi_results:[{}]}}}})});
+ await task;assert.equal(delivered,0,'destroyed recovery cannot deliver a late response');
+ pending.track(graph,'new','3');assert.equal(pending.pending.size,0);assert.equal(pending.controllers.size,0);
+}

tests/test_upload_transport.mjs

@@ -0,0 +1,68 @@
+import assert from "node:assert/strict";
+import {uploadImageWithProgress} from "../web/upload_transport.js";
+
+let current, status = 200;
+globalThis.XMLHttpRequest = class {
+  constructor() { current = this; this.upload = {}; this.headers = {}; }
+  open(method, url) { this.method = method; this.url = url; }
+  setRequestHeader(name, value) { this.headers[name] = value; }
+  send(body) {
+    this.body = body;
+    this.upload.onprogress({loaded: 40, total: 100, lengthComputable: true});
+    this.upload.onload();
+    this.status = status;
+    this.responseText = '{"name":"test.png"}';
+    this.onload();
+  }
+};
+
+const body = new FormData();
+let fallback = 0, phases = [];
+const api = {
+  user: "alice",
+  apiURL: path => `/cloud/proxy/api${path}`,
+  async fetchApi(path, options) {
+    fallback++;
+    assert.equal(path, "/upload/image"); assert.equal(options.body, body);
+    return {ok:true, json:async()=>({name:"fallback.png"})};
+  },
+};
+assert.deepEqual(await uploadImageWithProgress(api, body, event => phases.push(event)), {name:"test.png"});
+assert.equal(current.method, "POST");
+assert.equal(current.url, "/cloud/proxy/api/upload/image");
+assert.equal(current.headers["Comfy-User"], "alice");
+assert.deepEqual(phases, [{phase:"upload", loaded:40, total:100}, {phase:"processing"}]);
+assert.equal(fallback, 0);
+
+status = 401; phases = [];
+assert.deepEqual(await uploadImageWithProgress(api, body, event => phases.push(event)), {name:"fallback.png"});
+assert.equal(fallback, 1);
+assert.equal(phases.at(-2).phase, "authenticated");
+assert.equal(phases.at(-1).phase, "processing");
+status = 500;
+await assert.rejects(uploadImageWithProgress(api, body), /500/);
+console.log("Upload transport: real byte progress, proxy URL, multi-user header, auth fallback and server error passed");
+let aborted=0;
+globalThis.XMLHttpRequest=class {
+ constructor(){this.upload={};}open(){}setRequestHeader(){}send(){}abort(){aborted++;this.onabort?.();}
+};
+await assert.rejects(uploadImageWithProgress(api,body,null,{idleMs:5}),/timed out/);
+assert.equal(aborted,1);
+const cancel=new AbortController(),pending=uploadImageWithProgress(api,body,null,{signal:cancel.signal});cancel.abort();
+await assert.rejects(pending,{name:'AbortError'});assert.equal(aborted,2);
+// Overall duration may exceed the inactivity limit as long as real bytes move.
+globalThis.XMLHttpRequest=class {
+ constructor(){this.upload={};}open(){}setRequestHeader(){}
+ send(){let n=0;this.timer=setInterval(()=>{this.upload.onprogress?.({loaded:++n,total:6,lengthComputable:true});if(n===6){clearInterval(this.timer);this.status=200;this.responseText='{"name":"slow.png"}';this.onload();}},3);}
+ abort(){clearInterval(this.timer);this.onabort?.();}
+};
+assert.equal((await uploadImageWithProgress(api,body,null,{idleMs:12})).name,'slow.png');
+console.log('Upload transport: inactivity timeout, cancellation and slow advancing uploads pass');
+globalThis.XMLHttpRequest=class {
+ constructor(){this.upload={};}open(){}setRequestHeader(){}
+ send(){this.status=403;this.onload();}abort(){this.onabort?.();}
+};
+let fallbackSignal;
+const stalledAuth={...api,fetchApi:async(_path,options)=>{fallbackSignal=options.signal;return new Promise((_,reject)=>options.signal.addEventListener('abort',()=>reject(new DOMException('Aborted','AbortError')),{once:true}));}};
+await assert.rejects(uploadImageWithProgress(stalledAuth,body,null,{idleMs:5}),/timed out/);
+assert.equal(fallbackSignal.aborted,true,'the authenticated fallback also has a cancellation deadline');

tests/test_workbench_analysis.mjs

@@ -0,0 +1,17 @@
+import assert from 'node:assert/strict';
+import {inspectWorkbench,selectWorkbenchPipeline} from '../web/workbench_analysis.js';
+const graph={_nodes:[{type:'AdachiInlineMask'},{type:'AdachiWebUIInpaint'},
+ {type:'AdachiImageCompare'},{type:'AdachiGroupPresets'},
+ {type:'CheckpointLoaderSimple'},{type:'Power Lora Loader (rgthree)'}]};
+const result=inspectWorkbench(graph);
+assert.deepEqual(result.missing,[]);
+assert.equal(result.checkpoints.length,1);
+assert.equal(result.loras.length,1);
+assert.deepEqual(inspectWorkbench(null).missing,['inline_mask','inpaint','compare']);
+console.log('Workbench discovery checks passed');
+
+const first={id:10,type:'AdachiInlineMask',mode:0,inputs:[]},second={id:20,type:'AdachiInlineMask',mode:0,inputs:[]},a={id:30,type:'AdachiWebUIInpaint',mode:0,inputs:[{name:'image',link:1}]},b={id:40,type:'AdachiWebUIInpaint',mode:0,inputs:[{name:'image',link:2}]},outA={id:50,type:'AdachiImageCompare',mode:0,inputs:[{name:'result',link:3}]},outB={id:60,type:'AdachiImageCompare',mode:0,inputs:[{name:'result',link:4}]};
+const multi={_nodes:[first,second,a,b,outA,outB],links:{1:{origin_id:10},2:{origin_id:20},3:{origin_id:30},4:{origin_id:40}},getNodeById(id){return this._nodes.find(n=>n.id==id);}};multi._nodes.forEach(n=>n.graph=multi);
+const chosen=selectWorkbenchPipeline(multi,b);assert.equal(chosen.editor,second);assert.equal(chosen.compare,outB);assert.equal(chosen.ambiguous,false);
+b.inputs=[];outB.inputs=[];assert.equal(selectWorkbenchPipeline(multi,b).ambiguous,true,'unconnected ambiguous canvases need an explicit choice');
+console.log('Workbench selection: multiple independent pipelines follow actual connections; ambiguity is explicit');

tests/test_workspace_activation.mjs

@@ -0,0 +1,13 @@
+import assert from 'node:assert/strict';
+import {workspaceRoutingNodes} from '../web/workspace_activation.js';
+const model={id:11,type:'CLIPTextEncode',mode:4,inputs:[]},setter={id:12,type:'SetNode',mode:4,inputs:[{link:1}]},getter={id:13,type:'GetNode',mode:4,inputs:[],findSetter:()=>setter},gate={id:14,type:'AdachiExclusiveSwitch',mode:4,inputs:[{link:2}]},root={id:15,type:'AdachiWebUIInpaint',mode:4,inputs:[{link:3}]};
+const nodes=[model,setter,getter,gate,root],graph={_nodes:nodes,links:{1:{origin_id:11},2:{origin_id:13},3:{origin_id:14}},getNodeById:id=>nodes.find(n=>n.id==id)};nodes.forEach(n=>n.graph=graph);
+graph._groups=[{size:[500,500],_nodes:[getter,gate,root]}];
+assert.deepEqual([...workspaceRoutingNodes([root])].map(n=>n.id).sort(),[13,14]);
+assert.equal(setter.mode,4);assert.equal(model.mode,4,'model processing is never activated by routing discovery');
+root.mode=0;assert.equal(workspaceRoutingNodes([root]).size,0,'manually ignored routes of an enabled workbench stay ignored');
+root.mode=4;gate.mode=0;assert.equal(workspaceRoutingNodes([root]).size,0,'a mixed group is not guessed to be entirely bypassed');
+const controller={active:{id:'preset'},planStates:()=>({plan:new Map([[root,4],[getter,4],[gate,4]])})};
+assert.deepEqual([...workspaceRoutingNodes([root],controller)].map(n=>n.id),[13],'an explicit bypass preset allows only its necessary routing');
+controller.planStates=()=>({plan:new Map([[root,0]])});assert.equal(workspaceRoutingNodes([root],controller).size,0,'an individually ignored sampler cannot revive other intentionally disabled nodes');
+console.log('Workspace activation: bypassed group routing only, preserved external setters and processors, explicit preset state and mixed-group safety pass');

tests/test_workspace_bindings.mjs

@@ -0,0 +1,37 @@
+import assert from 'node:assert/strict';
+import {WorkspaceBindings} from '../web/workspace_bindings.js';
+const nodes=[];
+const add=(id,type,values={},inputs=[],outputs=[])=>{const n={id,type,mode:0,properties:{},widgets:Object.entries(values).map(([name,value])=>({name,value})),inputs,outputs:outputs.map(type=>({type,name:type}))};nodes.push(n);return n;};
+const unrelated=add('old-unrelated','CheckpointLoaderSimple',{ckpt_name:'do-not-pick.safetensors'});
+const a=add('base-a','CheckpointLoaderSimple',{ckpt_name:'custom-A.safetensors'});
+const b=add('base-b','UNETLoader',{unet_name:'anything-B.safetensors'});
+const clip=add('separate-encoder','CLIPLoader',{clip_name:'language.safetensors'});
+const vae=add('separate-vae','VAELoader',{vae_name:'image.safetensors'});
+const posA=add('encode-a','CLIPTextEncode',{text:'original A'},[{name:'clip',type:'CLIP'}],['CONDITIONING']);
+const negA=add('negative-a','CLIPTextEncode',{text:'negative A'},[{name:'clip',type:'CLIP'}],['CONDITIONING']);
+const posB=add('encode-b','CLIPTextEncode',{text:'original B'},[{name:'clip',type:'CLIP'}],['CONDITIONING']);
+const negB=add('negative-b','CLIPTextEncode',{text:'negative B'},[{name:'clip',type:'CLIP'}],['CONDITIONING']);
+add('model-patch','ModelSamplingSD3',{},[{name:'model',type:'MODEL'}],['MODEL']);
+const sampler=add('workbench','AdachiWebUIInpaint',{refiner_switch:.8},[{name:'model',type:'MODEL'},{name:'positive',type:'CONDITIONING'},{name:'negative',type:'CONDITIONING'},{name:'vae',type:'VAE'}]);
+const controller=add('preset-owner','AdachiGroupPresets');const applied=[];
+controller.adachiGroupPresets={model:{node:controller,data:{activeId:'one',presets:[{id:'one',name:'Portrait'},{id:'two',name:'Studio'}]},apply(id){this.data.activeId=id;applied.push(id);}}};
+const graph={_nodes:nodes,getNodeById:id=>nodes.find(n=>String(n.id)===String(id)),links:{}};
+const p=(kind)=>{
+ const root=kind==='one'?'base-a':'base-b',positive=kind==='one'?'encode-a':'encode-b',negative=kind==='one'?'negative-a':'negative-b';
+ return {[root]:{class_type:graph.getNodeById(root).type,inputs:{}},'separate-encoder':{class_type:'CLIPLoader',inputs:{}},'separate-vae':{class_type:'VAELoader',inputs:{}},
+  'model-patch':{class_type:'ModelSamplingSD3',inputs:{model:[root,0]}},
+  [positive]:{class_type:'CLIPTextEncode',inputs:{clip:kind==='one'?[root,1]:['separate-encoder',0],text:'base text'}},
+  [negative]:{class_type:'CLIPTextEncode',inputs:{clip:kind==='one'?[root,1]:['separate-encoder',0],text:'negative text'}},
+  workbench:{class_type:'AdachiWebUIInpaint',inputs:{model:['model-patch',0],positive:[positive,0],negative:[negative,0],vae:kind==='one'?[root,2]:['separate-vae',0]}}};
+};
+const x=new WorkspaceBindings({graph,sampler});assert.equal(x.profileId,'one');assert.equal(x.model,null,'discovery cannot guess from filenames or active loader order');
+x.updateExecutionGraph(p('one'),'workbench');assert.equal(x.model,a);assert.notEqual(x.model,unrelated);assert.equal(x.positive,posA);assert.equal(x.getExecutionBinding().loraMode,'inherit');
+assert.deepEqual(x.getExecutionBinding().samplerInputs,{},'final model-processing references are preserved');
+x.settings.positive='custom A';assert.deepEqual(x.getExecutionBinding().inputOverrides['encode-a'],{text:'custom A'});
+x.settings.loraMode='override';x.settings.loras=[{name:'style.safetensors',strength:.7}];assert.equal(x.getExecutionBinding().loras[0].name,'style.safetensors');
+x.selectProfile('two');x.updateExecutionGraph(p('two'),'workbench');assert.equal(x.model,b);assert.equal(x.clip,clip);assert.equal(x.vae,vae);assert.equal(x.positive,posB);assert.equal(x.getExecutionBinding().loraMode,'inherit');assert.equal(x.getExecutionBinding().inputOverrides['encode-b'],undefined);
+x.selectProfile('one');x.updateExecutionGraph(p('one'),'workbench');assert.equal(x.settings.positive,'custom A');assert.deepEqual(applied,['two','one']);
+controller.mode=4;const ignored=new WorkspaceBindings({graph,sampler});assert.equal(ignored.controller,undefined);assert.deepEqual(ignored.profiles.map(p=>p.id),['current']);
+controller.mode=0;controller.adachiGroupPresets.model.data.presets.push({id:'new',name:'Any other title'});ignored.refreshProfiles();assert.equal(ignored.profiles.length,3);
+controller.adachiGroupPresets.model.data.activeId='two';ignored.refreshProfiles();assert.equal(ignored.profileId,'two');
+console.log('Workspace bindings: arbitrary IDs and titles, actual loader/CLIP/VAE routes, preserved model patches, inherited/custom LoRAs, per-profile edits and dynamic controllers pass');

tests/test_workspace_execution.py

@@ -0,0 +1,84 @@
+"""Workspace validation must not execute, save or modify the submitted graph."""
+import importlib.util
+from pathlib import Path
+import types
+import unittest
+
+SPEC = importlib.util.spec_from_file_location("adachi_workspace_execution", Path(__file__).resolve().parents[1] / "workspace_execution.py")
+workspace = importlib.util.module_from_spec(SPEC)
+SPEC.loader.exec_module(workspace)
+
+
+class WorkspaceTests(unittest.IsolatedAsyncioTestCase):
+    def prompt(self):
+        return {
+            "1": {"class_type": "ImageSource", "inputs": {}},
+            "2": {"class_type": "ImageInvert", "inputs": {"image": ["1", 0]}},
+            "3": {"class_type": "AdachiImageCompare", "inputs": {"original": ["1", 0], "result": ["2", 0]}},
+            "4": {"class_type": "SaveImage", "inputs": {"images": ["2", 0]}},
+            "5": {"class_type": "NotInstalled", "inputs": {}},
+        }
+
+    async def test_only_comparison_branch_is_validated(self):
+        original = self.prompt()
+        calls = []
+
+        async def validate(prompt_id, prompt, targets):
+            calls.append((prompt_id, prompt, targets))
+            return True, None, ["3"], {}
+
+        server = types.SimpleNamespace(trigger_on_prompt=lambda data: data)
+        data, _, valid = await workspace.prepare_workspace(server, types.SimpleNamespace(validate_prompt=validate), {"prompt": original, "compare_id": "3"})
+        self.assertTrue(valid[0])
+        self.assertEqual(set(data["prompt"]), {"1", "2", "3"})
+        self.assertEqual(calls[0][2], ["3"])
+        self.assertEqual(set(original), {"1", "2", "3", "4", "5"})
+
+    async def test_hooks_restore_links_before_pruning(self):
+        original = self.prompt()
+        original["2"]["inputs"].clear()
+
+        def hook(data):
+            data["prompt"]["2"]["inputs"]["image"] = ["1", 0]
+            return data
+
+        def validate(prompt):
+            self.assertIn("1", prompt)
+            return True, None, ["3"], {}
+
+        server = types.SimpleNamespace(trigger_on_prompt=hook)
+        await workspace.prepare_workspace(server, types.SimpleNamespace(validate_prompt=validate), {"prompt": original, "compare_id": "3"})
+        self.assertEqual(original["2"]["inputs"], {})
+
+    def test_missing_dependency_and_non_comparison_output_rejected(self):
+        with self.assertRaises(ValueError):
+            workspace.dependency_prompt(self.prompt(), "4")
+        prompt = self.prompt()
+        del prompt["2"]
+        with self.assertRaisesRegex(ValueError, "missing node"):
+            workspace.dependency_prompt(prompt, "3")
+
+    def test_malformed_node_inputs_are_a_validation_error(self):
+        for value in ([], None, "invalid", 42):
+            prompt = self.prompt()
+            prompt["2"]["inputs"] = value
+            with self.assertRaisesRegex(ValueError, "Invalid input mapping"):
+                workspace.dependency_prompt(prompt, "3")
+
+    async def test_bad_inputs_never_reach_core_replacement_hooks(self):
+        def unexpected(_prompt):
+            self.fail("Malformed inputs reached a core hook")
+        server = types.SimpleNamespace(trigger_on_prompt=lambda data: data,
+            node_replace_manager=types.SimpleNamespace(apply_replacements=unexpected))
+        prompt = self.prompt()
+        prompt["2"]["inputs"] = None
+        with self.assertRaisesRegex(ValueError, "Invalid input mapping"):
+            await workspace.prepare_workspace(server, types.SimpleNamespace(), {"prompt":prompt, "compare_id":"3"})
+
+    async def test_validation_error_is_returned_without_queueing(self):
+        def validate(prompt):
+            return False, {"message": "missing checkpoint"}, [], {"1": {"errors": []}}
+        server = types.SimpleNamespace(trigger_on_prompt=lambda data: data)
+        _, _, valid = await workspace.prepare_workspace(server, types.SimpleNamespace(validate_prompt=validate), {"prompt": self.prompt(), "compare_id": "3"})
+        self.assertFalse(valid[0])
+        self.assertEqual(valid[1]["message"], "missing checkpoint")

tests/test_workspace_flow.mjs

@@ -0,0 +1,206 @@
+import assert from 'node:assert/strict';
+import {inspectExecutionFlow, applyExecutionBinding} from '../web/workspace_flow.js';
+
+// Execution IDs, filenames and titles deliberately have no relationship to the
+// author's personal workflow. Colon IDs represent nodes inside a subgraph.
+function fixture(prefix='graph-a') {
+    const ids=Object.fromEntries(['checkpoint','clip','vae','modelLora','negativeLora','modelPatch','clipPatch','positive','positive2','negative','combine','control','controlLoader','image','unusedCheckpoint','unusedEncoder','unusedLora','gate','sampler','compare'].map((k,i)=>[k,`${prefix}-${113+i*17}`]));
+    ids.positive2=`${prefix}-instance:937`;
+    const prompt={},live=new Map();
+    function add(key,type,inputs,outputs=[],fields=[]) {
+        const id=ids[key];prompt[id]={class_type:type,inputs};
+        live.set(id,{id:id.includes(':')?id.split(':').at(-1):id,type,mode:0,
+            outputs:outputs.map((o,i)=>typeof o==='string'?{name:o,type:o}:{...o,slot_index:i}),
+            inputs:Object.entries(inputs).map(([name,value])=>({name,type:Array.isArray(value)?'CONDITIONING':'STRING'})),
+            widgets:fields.map(name=>({name,value:inputs[name]}))});
+    }
+    add('checkpoint','CheckpointLoaderSimple',{ckpt_name:'arbitrary-custom-model.safetensors'},['MODEL','CLIP','VAE']);
+    add('clip','CLIPLoader',{clip_name:'independent-language-encoder.safetensors'},['CLIP']);
+    add('vae','VAELoader',{vae_name:'independent-image-vae.safetensors'},['VAE']);
+    add('modelLora','Power Lora Loader (rgthree)',{model:[ids.checkpoint,0],clip:[ids.checkpoint,1],
+        lora_1:{on:true,lora:'original-model-style.safetensors',strength:.8,strengthTwo:.4},
+        lora_9:{on:false,lora:'original-disabled.safetensors',strength:1}},['MODEL','CLIP']);
+    add('negativeLora','LoraLoader',{model:[ids.checkpoint,0],clip:[ids.clip,0],lora_name:'original-negative-style.safetensors',strength_model:.6,strength_clip:.3},['MODEL','CLIP']);
+    add('modelPatch','ModelSamplingSD3',{model:[ids.modelLora,0],shift:2.7},['MODEL']);
+    add('clipPatch','CLIPSetLastLayer',{clip:[ids.modelLora,1],stop_at_clip_layer:-2},['CLIP']);
+    add('positive','CLIPTextEncode',{clip:[ids.clipPatch,0],text:'a quiet city'},['CONDITIONING'],['text']);
+    add('positive2','CLIPTextEncodeSDXL',{clip:[ids.clipPatch,0],text_g:'city details',text_l:'architecture',width:1024,height:1024},['CONDITIONING'],['text_g','text_l']);
+    add('negative','CLIPTextEncode',{clip:[ids.negativeLora,1],text:'blurry'},['CONDITIONING'],['text']);
+    add('combine','ConditioningCombine',{conditioning_1:[ids.positive,0],conditioning_2:[ids.positive2,0]},['CONDITIONING']);
+    add('controlLoader','ControlNetLoader',{control_net_name:'edge-control.safetensors'},['CONTROL_NET']);
+    add('image','LoadImage',{image:'fixture.png'},['IMAGE','MASK']);
+    add('control','ControlNetApplyAdvanced',{positive:[ids.combine,0],negative:[ids.negative,0],control_net:[ids.controlLoader,0],image:[ids.image,0],strength:.7},
+        [{name:'positive',type:'CONDITIONING'},{name:'negative',type:'CONDITIONING'}]);
+    add('unusedCheckpoint','CheckpointLoaderSimple',{ckpt_name:'a-different-family.safetensors'},['MODEL','CLIP','VAE']);
+    add('unusedEncoder','CLIPTextEncode',{clip:[ids.unusedCheckpoint,1],text:'unused prompt'},['CONDITIONING'],['text']);
+    add('unusedLora','LoraLoader',{model:[ids.unusedCheckpoint,0],clip:[ids.unusedCheckpoint,1],lora_name:'unrelated.safetensors',strength_model:1,strength_clip:1},['MODEL','CLIP']);
+    add('gate','AdachiExclusiveSwitch',{active_inputs:1,input_a:[ids.control,0],input_b:[ids.unusedEncoder,0]},['CONDITIONING']);
+    add('sampler','AdachiWebUIInpaint',{model:[ids.modelPatch,0],positive:[ids.gate,0],negative:[ids.control,1],vae:[ids.vae,0],image:[ids.image,0],mask:[ids.image,1],steps:23,cfg:5.5},['IMAGE','IMAGE','MASK']);
+    add('compare','AdachiImageCompare',{original:[ids.image,0],result:[ids.sampler,0]},['IMAGE']);
+    for(const [id,node] of live)node.inputs=Object.entries(prompt[id].inputs).map(([name,value])=>({name,
+        type:Array.isArray(value)?live.get(value[0])?.outputs?.[value[1]]?.type||'*':typeof value==='number'?'FLOAT':'STRING'}));
+    return {ids,prompt,lookup:id=>live.get(String(id))||null,live};
+}
+const sorted=values=>[...values].sort();
+const entryIds=entries=>sorted(entries.map(e=>String(e.id)));
+const effective=(prompt,ref)=>{
+    const seen=new Set();
+    while(ref && prompt[ref[0]]?.class_type==='AdachiExclusiveSwitch' && !seen.has(ref[0])) {
+        seen.add(ref[0]);const n=prompt[ref[0]],flag=n.inputs.active_inputs;
+        if(flag!==1 && flag!==2)break;
+        ref=n.inputs[flag===1?'input_a':'input_b'];
+    }
+    return ref;
+};
+function loraChain(prompt,initial,field) {
+    const rows=[],seen=new Set();let ref=initial;
+    while(ref && prompt[ref[0]]?.class_type==='LoraLoader') {
+        assert.ok(!seen.has(ref[0]),'temporary LoRA chain cannot contain a cycle');seen.add(ref[0]);
+        assert.equal(ref[1],field==='model'?0:1,'model and CLIP must use the correct LoRA output slot');
+        const n=prompt[ref[0]];rows.unshift(n.inputs);ref=n.inputs[field];
+    }
+    return {rows,base:ref};
+}
+const inherit={loraMode:'inherit',loras:[]};
+
+// Inspection preserves the actual producer and discovers editable components
+// behind model/CLIP processing; it does not guess from a preset/model name.
+for(const prefix of ['renumbered-12001','completely-different-workflow']) {
+    const {ids,prompt,lookup}=fixture(prefix),before=structuredClone(prompt);
+    const flow=inspectExecutionFlow(prompt,ids.sampler,lookup);
+    assert.deepEqual(flow.refs.model,[ids.modelPatch,0]);
+    assert.deepEqual(effective(prompt,flow.refs.positive),[ids.control,0]);
+    assert.deepEqual(flow.refs.negative,[ids.control,1]);
+    assert.deepEqual(flow.refs.vae,[ids.vae,0]);
+    assert.deepEqual(entryIds(flow.modelNodes),[ids.checkpoint]);
+    assert.deepEqual(entryIds(flow.clipNodes),sorted([ids.checkpoint,ids.clip]));
+    assert.deepEqual(entryIds(flow.vaeNodes),[ids.vae]);
+    assert.deepEqual(entryIds(flow.loras),sorted([ids.modelLora,ids.negativeLora]));
+    assert.deepEqual(entryIds(flow.encoders.positive),sorted([ids.positive,ids.positive2]));
+    assert.deepEqual(entryIds(flow.encoders.negative),[ids.negative]);
+    const second=flow.encoders.positive.find(e=>e.id===ids.positive2);
+    assert.equal(second.node,lookup(ids.positive2),'execution IDs map back to the actual live subgraph node');
+    assert.deepEqual(sorted(second.fields),['text_g','text_l']);
+    assert.equal(flow.modelNodes[0].api,prompt[ids.checkpoint]);
+    assert.deepEqual(prompt,before,'inspection is read-only');
+}
+
+// A checkpoint's unused CLIP/VAE outputs must not displace separately wired
+// components merely because that checkpoint supplies the MODEL input.
+{
+    const {ids,prompt,lookup}=fixture();
+    prompt[ids.modelLora].inputs.clip=[ids.clip,0];
+    const flow=inspectExecutionFlow(prompt,ids.sampler,lookup);
+    assert.deepEqual(entryIds(flow.clipNodes),[ids.clip]);
+    assert.deepEqual(entryIds(flow.vaeNodes),[ids.vae]);
+}
+
+// Inherited LoRA rows and processing nodes survive switch simplification.
+{
+    const {ids,prompt}=fixture(),before=structuredClone(prompt);
+    const result=applyExecutionBinding(prompt,ids.sampler,inherit);
+    assert.notEqual(result,prompt);assert.notEqual(result[ids.sampler].inputs,prompt[ids.sampler].inputs);
+    assert.deepEqual(result[ids.sampler].inputs.positive,[ids.control,0]);
+    assert.deepEqual(result[ids.sampler].inputs.model,[ids.modelPatch,0]);
+    for(const key of ['modelLora','negativeLora','modelPatch','clipPatch','control','combine','unusedCheckpoint','unusedEncoder','unusedLora'])assert.deepEqual(result[ids[key]],before[ids[key]],key+' must keep its original fields');
+    assert.deepEqual(prompt,before);
+}
+
+// Explicitly disabling LoRAs bypasses both output types inside this sampler's
+// dependency chain while retaining model-sampling and conditioning processors.
+{
+    const {ids,prompt}=fixture(),before=structuredClone(prompt);
+    const result=applyExecutionBinding(prompt,ids.sampler,{loraMode:'override',loras:[]});
+    assert.deepEqual(result[ids.sampler].inputs.model,[ids.modelPatch,0]);
+    assert.deepEqual(result[ids.modelPatch].inputs.model,[ids.checkpoint,0]);
+    assert.deepEqual(result[ids.clipPatch].inputs.clip,[ids.checkpoint,1]);
+    assert.deepEqual(result[ids.negative].inputs.clip,[ids.clip,0]);
+    assert.deepEqual(result[ids.positive].inputs.clip,[ids.clipPatch,0]);
+    assert.deepEqual(result[ids.sampler].inputs.positive,[ids.control,0]);
+    assert.deepEqual(result[ids.sampler].inputs.negative,[ids.control,1]);
+    assert.deepEqual(result[ids.control],before[ids.control]);
+    assert.deepEqual(result[ids.unusedLora],before[ids.unusedLora]);
+    assert.deepEqual(prompt,before,'LoRA overrides only affect the returned execution copy');
+}
+
+// Added LoRAs affect MODEL and each explicitly selected CLIP boundary. Distinct
+// positive/negative CLIP sources stay distinct and downstream conditions stay put.
+{
+    const {ids,prompt}=fixture(),before=structuredClone(prompt);
+    const chosen=[{name:'custom-one.safetensors',strength:.7,clipStrength:0,on:true},
+        {name:'custom-two.safetensors',strength:1.1,clipStrength:.25,on:true},
+        {name:'disabled-row.safetensors',strength:9,on:false}];
+    const result=applyExecutionBinding(prompt,ids.sampler,{loraMode:'override',loras:chosen,
+        clipConsumers:[{id:ids.clipPatch,input:'clip'},{id:ids.negative,input:'clip'}]});
+    const model=loraChain(result,result[ids.sampler].inputs.model,'model');
+    assert.deepEqual(model.base,[ids.modelPatch,0]);
+    assert.deepEqual(model.rows.map(r=>r.lora_name),chosen.slice(0,2).map(r=>r.name));
+    assert.deepEqual(model.rows.map(r=>r.strength_model),[.7,1.1]);
+    const positive=loraChain(result,result[ids.clipPatch].inputs.clip,'clip');
+    const negative=loraChain(result,result[ids.negative].inputs.clip,'clip');
+    for(const chain of [positive,negative]) {
+        assert.deepEqual(chain.rows.map(r=>r.lora_name),chosen.slice(0,2).map(r=>r.name));
+        assert.deepEqual(chain.rows.map(r=>r.strength_clip),[0,.25],'zero CLIP strength must not fall back to model strength');
+    }
+    assert.deepEqual(positive.base,[ids.checkpoint,1]);assert.deepEqual(negative.base,[ids.clip,0]);
+    assert.deepEqual(result[ids.positive].inputs.clip,[ids.clipPatch,0]);
+    assert.deepEqual(result[ids.sampler].inputs.positive,[ids.control,0]);assert.deepEqual(result[ids.sampler].inputs.negative,[ids.control,1]);
+    assert.deepEqual(result[ids.control],before[ids.control]);assert.deepEqual(result[ids.unusedLora],before[ids.unusedLora]);
+    assert.ok(Object.keys(result).some(id=>!Object.hasOwn(prompt,id)),'custom LoRAs are execution-only nodes');
+    assert.deepEqual(prompt,before);
+}
+
+// Primitive edits can target an encoder inside a subgraph without replacing
+// its CLIP or the downstream ControlNet/conditioning route.
+{
+    const {ids,prompt}=fixture(),before=structuredClone(prompt);
+    prompt[ids.sampler].inputs.positive=[ids.control,0];
+    prompt[ids.sampler].inputs.refiner_model=[ids.checkpoint,0];
+    const result=applyExecutionBinding(prompt,ids.sampler,{...inherit,
+        inputOverrides:{[ids.positive2]:{text_g:'edited global prompt'}},
+        removeInputs:{[ids.sampler]:['refiner_model']}});
+    assert.equal(result[ids.positive2].inputs.text_g,'edited global prompt');
+    assert.equal(result[ids.positive2].inputs.text_l,before[ids.positive2].inputs.text_l);
+    assert.deepEqual(result[ids.positive2].inputs.clip,before[ids.positive2].inputs.clip);
+    assert.deepEqual(result[ids.control],before[ids.control]);
+    assert.deepEqual(result[ids.gate],before[ids.gate],'unreachable routers must remain untouched');
+    assert.equal(Object.hasOwn(result[ids.sampler].inputs,'refiner_model'),false);
+    assert.equal(prompt[ids.positive2].inputs.text_g,before[ids.positive2].inputs.text_g);
+    assert.ok(Object.hasOwn(prompt[ids.sampler].inputs,'refiner_model'));
+}
+
+// Selecting B is deterministic; unresolved/invalid selectors may be rejected
+// explicitly or left intact for validation, but must never silently select A.
+{
+    const {ids,prompt,lookup}=fixture();prompt[ids.gate].inputs.active_inputs=2;
+    const flow=inspectExecutionFlow(prompt,ids.sampler,lookup);
+    assert.deepEqual(entryIds(flow.encoders.positive),[ids.unusedEncoder]);
+    assert.deepEqual(applyExecutionBinding(prompt,ids.sampler,inherit)[ids.sampler].inputs.positive,[ids.unusedEncoder,0]);
+    for(const flags of [0,3,-1]) {
+        prompt[ids.gate].inputs.active_inputs=flags;
+        try {
+            const result=applyExecutionBinding(prompt,ids.sampler,inherit);
+            assert.deepEqual(result[ids.sampler].inputs.positive,[ids.gate,0]);
+            assert.deepEqual(result[ids.gate].inputs,prompt[ids.gate].inputs);
+        } catch(error) {
+            assert.notEqual(error.name,'TypeError');
+            assert.match(error.message,/唯一|歧义|互斥|启用|exclusive|ambiguous|enabled|branch|input/i);
+        }
+    }
+}
+
+console.log('Workspace execution flow: dynamic IDs, processing chains, split CLIP sources, multi-encoder conditioning, inherited/overridden LoRAs, switch safety and immutable unrelated branches pass');
+
+// Real Vue-backed widget rows are proxies; ComfyUI transmits their JSON data.
+{
+ const {ids,prompt}=fixture();prompt[ids.modelLora].inputs.lora_1=new Proxy({on:true,lora:'vue-row.safetensors',strength:.4},{});
+ const copy=applyExecutionBinding(prompt,ids.sampler,inherit);assert.equal(copy[ids.modelLora].inputs.lora_1.lora,'vue-row.safetensors');
+}
+
+{
+ const {ids,prompt,lookup}=fixture();prompt[ids.gate]={class_type:'ComfySwitchNode',inputs:{switch:false,on_true:[ids.control,0],on_false:[ids.unusedEncoder,0]}};
+ assert.deepEqual(entryIds(inspectExecutionFlow(prompt,ids.sampler,lookup).encoders.positive),[ids.unusedEncoder]);
+ assert.deepEqual(applyExecutionBinding(prompt,ids.sampler,inherit)[ids.sampler].inputs.positive,[ids.gate,0],'native switch execution is preserved');
+ prompt[ids.gate]={class_type:'ComfySoftSwitchNode',inputs:{switch:true,on_false:[ids.unusedEncoder,0]}};
+ assert.deepEqual(entryIds(inspectExecutionFlow(prompt,ids.sampler,lookup).encoders.positive),[ids.unusedEncoder]);
+}

tests/test_workspace_history.mjs

@@ -0,0 +1,70 @@
+import assert from "node:assert/strict";
+import {WorkspaceHistory,captureSourceState,restoreSourceState} from "../web/workspace_history.js";
+
+class Raster {
+    constructor(pixels) { this.pixels = pixels; this.width = 4; this.height = 3; }
+    toDataURL() { return JSON.stringify({pixels:this.pixels}); }
+}
+const decode = async data => new Raster(JSON.parse(data).pixels);
+const source = name => ({src:name, width:4,height:3});
+const context = prefix => ({mask:new Raster(prefix+"-mask"), baseMask:new Raster(prefix+"-base"),history:[{type:"brush",points:[{x:1,y:2}]}],future:[{type:"erase",points:[{x:2,y:1}]}],lastMask:prefix+"-serialized",dirty:false});
+const widgetNames=["image","mask_data","source_data","edit_mode","crop_state","blend_data","layer_image"];
+function node() {
+    const n={properties:{},widgets:widgetNames.map(name=>({name,value:name})),graph:{setDirtyCanvas(){}}};
+    const e=n.adachiEditor={...context("repaint"),image:source("original"),ready:true,loadVersion:1,maskRevision:0,pan:{x:1,y:2},zoom:2,keepFit:true,view:"overlay",filename:{textContent:"original.png"},sizeLabel:{},embed:{checked:false},flush(){},schedule(){},updateButtons(){}};
+    n.adachiCrop={contexts:{blend:context("blend")},target:"mask",layer:source("layer"),cropWidth:512,cropHeight:768,revision:0,acceptVersion:0,sync(){},syncPreset(){},stopOriginalPreview(){}};
+    return n;
+}
+
+{
+ const n=node(),e=n.adachiEditor,c=n.adachiCrop;
+ n.widgets.find(w=>w.name==="edit_mode").value="crop";
+ n.widgets.find(w=>w.name==="crop_state").value=JSON.stringify({locked:true,x:1,y:2,width:512,height:768,job:"job-a"});
+ const before=captureSourceState(n);
+ e.mask.pixels="edited";e.history[0].points[0].x=99;e.image=source("adopted");c.layer=source("new-layer");c.target="blend";
+ n.widgets.find(w=>w.name==="crop_state").value="{}";
+ await restoreSourceState(n,before,{decode});
+ assert.equal(e.image.src,"original");assert.equal(c.layer.src,"layer");
+ assert.equal(e.mask.pixels,"repaint-mask");assert.equal(e.baseMask.pixels,"repaint-base");
+ assert.equal(c.contexts.blend.mask.pixels,"blend-mask");assert.equal(c.target,"mask");
+ assert.equal(e.history[0].points[0].x,1);assert.equal(e.future[0].type,"erase");
+ assert.equal(JSON.parse(n.widgets.find(w=>w.name==="crop_state").value).job,"job-a");
+ assert.deepEqual(e.pan,{x:1,y:2});assert.equal(e.zoom,2);assert.equal(e.keepFit,true);
+ assert.notEqual(e.mask,before.contexts.mask.mask);
+ e.mask.pixels="changed-again";await restoreSourceState(n,before,{decode});assert.equal(e.mask.pixels,"repaint-mask");
+}
+{
+ const n=node(),c=n.adachiCrop,e=n.adachiEditor;
+ c.contexts.mask=context("saved-repaint");Object.assign(e,context("active-blend"));c.target="blend";
+ const before=captureSourceState(n);e.mask.pixels="lost";await restoreSourceState(n,before,{decode});
+ assert.equal(e.mask.pixels,"active-blend-mask");assert.equal(c.contexts.mask.mask.pixels,"saved-repaint-mask");assert.equal(c.target,"blend");
+}
+{
+ const n=node(),before=captureSourceState(n),old=n.adachiEditor.image;
+ await assert.rejects(restoreSourceState(n,before,{decode:async()=>{throw Error("broken PNG")}}),/broken PNG/);
+ assert.equal(n.adachiEditor.image,old);
+ await assert.rejects(restoreSourceState(n,before,{decode:async data=>{n.adachiEditor.loadVersion++;return decode(data)}}),/image changed/);
+ assert.equal(n.adachiEditor.image,old);
+}
+{
+ const n=node(),h=new WorkspaceHistory(n,{maxEntries:3});h.checkpoint("original");
+ await h.transact("adopted",async()=>{n.adachiEditor.image=source("adopted");n.widgets[0].value="adopted.png"});
+ assert.equal(h.entries.length,2);assert.equal(h.current.label,"adopted");assert.equal(h.entries[0].state.image.src,"original");
+ const count=h.entries.length;await assert.rejects(h.transact("failed",async()=>{throw Error("upload failed")}),/upload failed/);assert.equal(h.entries.length,count);assert.equal(h.busy,false);
+ for(let i=0;i<6;i++){n.adachiEditor.image=source("next"+i);n.widgets[0].value="next"+i;h.checkpoint("next"+i);}
+ assert.equal(h.entries.length,3);assert.equal(h.entries[0].label,"original");assert.equal(h.current.label,"next5");
+ n.adachiEditor.stroke={};assert.throws(()=>h.checkpoint("stroke"),/Finish the current edit/);
+}
+console.log("Workspace source history: repaint/blend pixels, crop state, layer, commands, failure atomicity and retention passed");
+{
+ const n=node(),before=captureSourceState(n),created=[];
+ let count=0;
+ await assert.rejects(restoreSourceState(n,before,{decode:async data=>{if(++count===2)throw Error('bad base');const c=await decode(data);created.push(c);return c;}}),/bad base/);
+ assert.equal(created[0].width,0,'a partial decode failure frees already allocated masks');
+ assert.equal(n.adachiEditor.image,before.image);
+ for(const change of [()=>n.adachiEditor.maskRevision++,()=>n.adachiCrop.revision++]){
+  await assert.rejects(restoreSourceState(n,before,{decode:async data=>{change();return decode(data);}}),/image changed/);
+  assert.equal(n.adachiEditor.image,before.image,'a completed stroke or crop change during decoding is preserved');
+ }
+ await assert.rejects(restoreSourceState(n,before,{decode:async data=>{n.graph=null;return decode(data);}}),/image changed/);
+}

tests/test_workspace_models.mjs

@@ -0,0 +1,23 @@
+import assert from 'node:assert/strict';
+import {WorkspaceBindings} from '../web/workspace_bindings.js';
+const n=(id,type,values={})=>({id,type,mode:0,properties:{},widgets:Object.entries(values).map(([name,value])=>({name,value})),inputs:[]});
+const sampler=n(501,'AdachiWebUIInpaint');
+const nodes=[sampler,n(359,'CheckpointLoaderSimple',{ckpt_name:'anima_baseV10.safetensors'}),n(357,'CLIPLoader',{clip_name:'qwen_3_06b_base.safetensors'}),n(358,'VAELoader',{vae_name:'qwen_image_vae.safetensors'}),n(547,'CLIPTextEncode',{text:''}),n(16,'CLIPTextEncode',{text:''}),n(518,'CheckpointLoaderSimple',{ckpt_name:'wai.safetensors'}),n(521,'CLIPTextEncode',{text:''}),n(520,'CLIPTextEncode',{text:''})];
+const controller=n(524,'AdachiGroupPresets');nodes.push(controller);
+controller.adachiGroupPresets={model:{node:controller,data:{activeId:'anima',presets:[{id:'anima',name:'anima修图'},{id:'wai',name:'wai生图'}]},apply(id){this.data.activeId=id;}}};
+const graph={_nodes:nodes,links:{},getNodeById:id=>nodes.find(n=>String(n.id)===String(id))};
+const compiled=(profile)=>profile==='anima'?{
+ '359':{class_type:'CheckpointLoaderSimple',inputs:{}},'357':{class_type:'CLIPLoader',inputs:{}},'358':{class_type:'VAELoader',inputs:{}},
+ '547':{class_type:'CLIPTextEncode',inputs:{clip:['357',0],text:''}},'16':{class_type:'CLIPTextEncode',inputs:{clip:['357',0],text:''}},
+ '501':{class_type:'AdachiWebUIInpaint',inputs:{model:['359',0],positive:['547',0],negative:['16',0],vae:['358',0]}}
+}:{'518':{class_type:'CheckpointLoaderSimple',inputs:{}},'521':{class_type:'CLIPTextEncode',inputs:{clip:['518',1],text:''}},'520':{class_type:'CLIPTextEncode',inputs:{clip:['518',1],text:''}},'501':{class_type:'AdachiWebUIInpaint',inputs:{model:['518',0],positive:['521',0],negative:['520',0],vae:['518',2]}}};
+const b=new WorkspaceBindings({graph,sampler});b.updateExecutionGraph(compiled('anima'),'501');assert.equal(b.missingDependencies().length,0,'not-yet-loaded lists cannot prove missing weights');
+const info=(type,name,values)=>({[type]:{input:{required:{[name]:[values]}}}});
+const sources={'/object_info/CheckpointLoaderSimple':info('CheckpointLoaderSimple','ckpt_name',['wai.safetensors']),'/object_info/UNETLoader':info('UNETLoader','unet_name',[]),'/object_info/CLIPLoader':info('CLIPLoader','clip_name',[]),'/object_info/VAELoader':info('VAELoader','vae_name',['pixel_space']),'/models/loras':[]};
+await b.loadCatalog(async path=>({ok:true,json:async()=>sources[path]}));
+assert.deepEqual(b.missingDependencies().map(d=>d.widget.value),['anima_baseV10.safetensors','qwen_3_06b_base.safetensors','qwen_image_vae.safetensors']);
+assert.deepEqual(b.missingDependencies().map(d=>d.folder),['checkpoints','text_encoders','vae']);
+b.selectProfile('wai');b.updateExecutionGraph(compiled('wai'),'501');assert.equal(b.missingDependencies().length,0,'an available checkpoint supplies its own CLIP and VAE');assert.equal(b.modelDependencies().length,1);
+b.selectProfile('anima');b.updateExecutionGraph(compiled('anima'),'501');b.catalog.checkpoints.push('anima_baseV10.safetensors');b.catalog.clips.push('qwen_3_06b_base.safetensors');b.catalog.vaes.push('qwen_image_vae.safetensors');assert.equal(b.missingDependencies().length,0);
+const failed=new WorkspaceBindings({graph,sampler});failed.updateExecutionGraph(compiled('anima'),'501');await failed.loadCatalog(async()=>{throw Error('network offline');});assert.equal(failed.missingDependencies().length,0,'a network failure must not be reported as missing models');assert.ok(failed.catalogError);
+console.log('Workspace model dependencies: separate model/CLIP/VAE, absent weights, built-in checkpoint components and unavailable catalogs pass');

tests/test_workspace_progress.mjs

@@ -0,0 +1,22 @@
+import assert from 'node:assert/strict';
+import {WorkspaceProgress} from '../web/workspace_progress.js';
+const nodes={'1':{class_type:'CheckpointLoaderSimple'},'2':{class_type:'AdachiWebUIInpaint'},'3':{class_type:'AdachiImageCompare'}};
+const p=new WorkspaceProgress();p.begin(nodes);
+p.handle('executing',null);
+p.handle('execution_start',{prompt_id:'mine'});p.handle('progress',{prompt_id:'other',node:'2',value:9,max:10});
+p.attach('mine');assert.equal(p.state,'running');assert.equal(p.value,null,'a different prompt must not set the percentage');
+p.handle('executing',1);assert.equal(p.label,'Loading main model','legacy scalar node IDs remain supported');
+p.handle('executing',{prompt_id:'mine',node:'1'});assert.equal(p.label,'Loading main model');assert.equal(p.max,null);
+p.handle('progress',{prompt_id:'mine',node:'2',value:3,max:20});assert.equal(p.value,3);assert.equal(p.max,20);assert.match(p.label,/KSampler/);
+p.handle('executing',{prompt_id:'mine',node:null});assert.equal(p.value,3,'terminal executing packets cannot reset the sampler progress');
+assert.equal(p.handle('progress',{prompt_id:'other',node:'2',value:19,max:20}),false);assert.equal(p.value,3);
+p.handle('progress_state',{prompt_id:'mine',nodes:{'1':{state:'finished',value:1,max:1},'2':{state:'running',value:4,max:20}}});assert.equal(p.value,4);
+p.handle('executing',{prompt_id:'mine',node:'3'});assert.equal(p.value,null,'a new node cannot inherit the sampler percentage');assert.equal(p.label,'Preparing comparison images');
+p.connection(false);assert.equal(p.max,null);assert.match(p.label,/Connection lost/);
+p.handle('execution_success',{prompt_id:'mine'});assert.equal(p.value/p.max,1);assert.equal(p.state,'complete');
+p.handle('progress',{prompt_id:'mine',node:'2',value:1,max:20});assert.equal(p.state,'complete','late packets must not regress completion');
+p.begin(nodes,'error');p.handle('execution_error',{prompt_id:'error',exception_message:'missing model'});assert.equal(p.state,'error');assert.equal(p.label,'missing model');assert.equal(p.value,null);
+p.begin(nodes,'cache');p.handle('execution_cached',{prompt_id:'cache',nodes:['1','2']});assert.match(p.label,/2 nodes/);assert.equal(p.value,null);
+p.handle('progress_state',{prompt_id:'cache',nodes:{'2':{state:'running',value:0,max:1}}});assert.equal(p.max,null,'node-start placeholder 0/1 must remain indeterminate');
+p.begin(nodes);p.handle('execution_success',{prompt_id:'fast'});p.handle('executing',null);p.attach('fast');assert.equal(p.state,'complete','legacy completion before HTTP acknowledgment must not throw');
+console.log('Workspace progress: actual stage/steps, task filtering, early events, both progress protocols, cache, reconnect and late-packet safety pass');

tests/test_workspace_routes.mjs

@@ -0,0 +1,10 @@
+import assert from 'node:assert/strict';import {conditioningSource} from '../web/workspace_routes.js';
+const nodes=[];const add=(id,type,inputs=[],value)=>{const n={id,type,mode:0,inputs,widgets:[{value}]};nodes.push(n);return n};
+const a=add(1,'CLIPTextEncode'),b=add(2,'CLIPTextEncode');const sa=add(3,'SetNode',[{name:'x',link:1}],'p');const ga=add(4,'GetNode',[],'p');const sb=add(5,'SetNode',[{name:'x',link:2}],'p2');const gb=add(6,'GetNode',[],'p2');const sw=add(7,'AdachiExclusiveSwitch',[{name:'input_a',link:3},{name:'input_b',link:4}]);const sampler=add(8,'AdachiWebUIInpaint',[{name:'positive',link:5},{name:'negative',link:6}]);
+const links={1:{origin_id:1,origin_slot:0},2:{origin_id:2,origin_slot:0},3:{origin_id:4,origin_slot:0},4:{origin_id:6,origin_slot:0},5:{origin_id:7,origin_slot:0},6:{origin_id:2,origin_slot:0}};
+const g={_nodes:nodes,links,getNodeById:id=>nodes.find(n=>n.id==id)};nodes.forEach(n=>n.graph=g);
+assert.equal(conditioningSource(g,sampler,'positive'),null,'ambiguous branches must not be guessed');
+assert.equal(conditioningSource(g,sampler,'positive',new Map([[sb,4]])),a);
+assert.equal(conditioningSource(g,sampler,'negative'),b);
+sa.inputs[0].link=3;assert.equal(conditioningSource(g,sampler,'positive',new Map([[sb,4]])),null,'cycle must stop');
+console.log('Conditioning discovery: actual sampler route, Get/Set, preset state, ambiguity and cycles pass');

tests/test_workspace_transport.mjs

@@ -0,0 +1,15 @@
+import assert from 'node:assert/strict';
+import {workspaceRequest} from '../web/workspace_transport.js';
+let posts=0;const reply=(status,data)=>({status,ok:status<300,json:async()=>data});
+await assert.rejects(workspaceRequest({fetchApi:async(p,o)=>{if(o.method)posts++;return reply(405,{});}},'submit',{}),/backend is not loaded/);assert.equal(posts,0);
+const seen=[];
+const api={fetchApi:async(p,o)=>{seen.push(p);if(p==='/adachi-inpaint/workspace/capabilities')return reply(404,{});if(p.endsWith('/capabilities'))return reply(200,{api_version:1,can_validate:true});return reply(200,{valid:true});}};
+assert.equal((await workspaceRequest(api,'validate',{})).valid,true);assert.equal(seen.at(-1),'/api/adachi-inpaint/workspace/validate');
+await assert.rejects(workspaceRequest({fetchApi:async(p)=>p.endsWith('/capabilities')?reply(200,{api_version:1,can_validate:true}):reply(400,{valid:false,error:{message:'bad model'}})},'submit',{}),/bad model/);
+console.log('Workspace transport: missing backend blocks submission, /api proxy fallback and validation errors pass');
+const missing={fetchApi:async p=>p.endsWith('/capabilities')?reply(200,{api_version:1,can_validate:true}):reply(200,{valid:false,node_errors:{'357':{errors:[{type:'value_not_in_list',message:'Value not in list',details:'long technical inventory',extra_info:{input_name:'clip_name',received_value:'qwen_3_06b_base.safetensors'}}]}}})};
+await assert.rejects(workspaceRequest(missing,'validate',{}),e=>e.message==='Text encoder not found: qwen_3_06b_base.safetensors' && !!e.nodeErrors['357']);
+globalThis.comfyAPI={app:{app:{ui:{settings:{getSettingValue:()=> 'zh'}}}}};
+try {
+    await assert.rejects(workspaceRequest(missing,'validate',{}),e=>e.message==='未找到文本编码器：qwen_3_06b_base.safetensors' && !!e.nodeErrors['357']);
+} finally {delete globalThis.comfyAPI;}

web/compare.js

@@ -1,17 +1,12 @@
 /* Original hover-comparison UI, MIT licensed. */
 import { T } from "./i18n.js";
+import {loadPreviewPair} from './preview_images.js';
 const element = (tag, className, text) => {
     const node = document.createElement(tag);
     if (className) node.className = className;
     if (text) node.textContent = text;
     return node;
 };
-const imageFrom = url => new Promise((resolve, reject) => {
-    const image = new Image();
-    image.onload = () => resolve(image);
-    image.onerror = () => reject(new Error(T("Preview unavailable; run the workflow again")));
-    image.src = url;
-});
 
 export class ImageCompareView {
     constructor(options) {
@@ -27,6 +22,15 @@ export class ImageCompareView {
         this.resultSelect.onchange = () => { this.index = Number(this.resultSelect.value); this.options.onSelect?.(this.index); this.loadSelected(); };
         toolbar.append(this.resultSelect, this.dimensionsLabel);
         const spacer = element("span", "dd-spacer"); toolbar.append(spacer);
+        this.reloadButton = element("button", "", "↻"); this.reloadButton.type = "button";
+        this.reloadButton.title = T("Reload comparison images"); this.reloadButton.setAttribute("aria-label", this.reloadButton.title);
+        this.reloadButton.onclick = async event => { event.preventDefault(); event.stopPropagation(); await this.options.recover?.(); this.loadSelected(); };
+        toolbar.append(this.reloadButton);
+        this.downloadButton = element("button", "", "↓"); this.downloadButton.type = "button";
+        this.downloadButton.title = T("Download the latest generated image");
+        this.downloadButton.setAttribute("aria-label", this.downloadButton.title);
+        this.downloadButton.onclick = event => { event.preventDefault(); event.stopPropagation(); this.downloadLatest(); };
+        toolbar.append(this.downloadButton);
         const fit = element("button", "", T("Fit")); fit.type = "button";
         fit.onclick = event => { event.preventDefault(); event.stopPropagation(); this.schedule(); }; toolbar.append(fit);
         this.stage = element("div", "dd-stage dd-compare-stage");
@@ -39,6 +43,13 @@ export class ImageCompareView {
         Object.assign(this.divider.style, { position: "absolute", background: "#ffffffb3", pointerEvents: "none" });
         this.divider.hidden = true;
         this.stage.append(this.canvas, this.divider);
+        this.progressBox=element("div","dd-preview-progress");this.progressBox.hidden=true;
+        this.progressBox.setAttribute("role","status");
+        this.progressText=element("span");this.progressBar=document.createElement("progress");this.progressBar.max=1;
+        this.progressBar.setAttribute("aria-label",T("Comparison image loading progress"));
+        this.progressBox.append(this.progressText,this.progressBar);this.stage.append(this.progressBox);
+        this.retryOnOnline=()=>{if(!this.ready && !this.loadingActive && this.results.length)this.loadSelected();};
+        window.addEventListener("online",this.retryOnOnline);
         this.empty = element("div", "dd-empty", T("Connect an original image and a result, then run the workflow")); this.stage.append(this.empty);
         const adoption = element("div", "dd-row dd-adoption");
         adoption.append(element("span", "dd-compare-label", T("Adopt into")));
@@ -83,7 +94,30 @@ export class ImageCompareView {
         this.updateButton();
     }
 
-    updateButton() { this.adoptButton.disabled = !this.ready || !this.targetSelect.value || !!this.busy; this.adoptButton.textContent = this.busy ? T("Adopting…") : T("Adopt this result"); }
+    updateButton() {
+        this.adoptButton.disabled = !this.ready || !this.targetSelect.value || !!this.busy || this.pendingOutputs>0;
+        this.adoptButton.textContent = this.busy ? T("Adopting…") : T("Adopt this result");
+        this.downloadButton.disabled = !this.latestResult || !!this.downloading;
+    }
+
+    setLatestResult(result) { this.latestResult = result || null; this.updateButton(); }
+
+    async downloadLatest() {
+        if (!this.latestResult || this.downloading) return;
+        const result = this.latestResult;
+        this.downloading = true; this.updateButton();
+        const controller=this.downloadController=new AbortController();
+        try {
+            const blob = await this.options.fetchResult(result,{signal:controller.signal});
+            if (this.destroyed) return;
+            const url = URL.createObjectURL(blob), link = document.createElement("a");
+            link.href = url; link.download = "adachi-inpaint-result.png";
+            document.body.append(link); link.click(); link.remove();
+            setTimeout(() => URL.revokeObjectURL(url), 60000);
+            this.status("");
+        } catch (error) { if (!this.destroyed) this.status(error.message, true); }
+        finally {if(this.downloadController===controller)this.downloadController=null;this.downloading = false; if (!this.destroyed) this.updateButton();}
+    }
 
     async setData(data, index = this.index) {
         this.originals = Array.isArray(data?.originals) ? data.originals : [];
@@ -99,33 +133,75 @@ export class ImageCompareView {
         await this.loadSelected();
     }
 
-    async loadSelected() {
-        const version = ++this.version;
-        this.ready = false; this.updateButton();
-        if (!this.originals.length || !this.results.length) return;
-        const repetitions = this.results.length / this.originals.length;
-        const original = this.originals[Math.min(this.originals.length - 1, Math.floor(this.index / repetitions))];
+    loading(text, states) {
+        this.progressBox.hidden=false;this.progressText.textContent=text;
+        if(states) {
+            const loaded=states.reduce((n,s)=>n+s.loaded,0),total=states.every(s=>s.total>0)?states.reduce((n,s)=>n+s.total,0):0;
+            const size=n=>n>=1048576?`${(n/1048576).toFixed(1)} MiB`:`${Math.round(n/1024)} KiB`;
+            this.progressText.textContent+=total?` · ${Math.min(100,Math.round(loaded/total*100))}% · ${size(loaded)} / ${size(total)}`:` · ${size(loaded)}`;
+            if(total)this.progressBar.value=Math.min(1,loaded/total);else this.progressBar.removeAttribute("value");
+        } else this.progressBar.removeAttribute("value");
+    }
+
+    async loadSelected({background=false}={}) {
+        if(this.destroyed)return;
+        clearTimeout(this.retryTimer);this.loadController?.abort();
+        const version=++this.version,controller=this.loadController=new AbortController();
+        this.reloadButton.disabled=false;this.loadingActive=true;
+        this.ready=false;this.updateButton();this.schedule();
+        if(!this.originals.length || !this.results.length){this.loadingActive=false;this.progressBox.hidden=true;return;}
+        const repetitions=this.results.length/this.originals.length;
+        const original=this.originals[Math.min(this.originals.length-1,Math.floor(this.index/repetitions))];
+        const current=()=>version===this.version && !this.destroyed && !controller.signal.aborted;
+        const blobKey=JSON.stringify([original,this.results[this.index]]);
+        if(this.blobKey!==blobKey){this.blobKey=blobKey;this.pendingBlobs=[null,null];}
+        const blobs=this.pendingBlobs||[null,null];
         try {
-            const images = await Promise.all([imageFrom(this.options.resolveUrl(original)), imageFrom(this.options.resolveUrl(this.results[this.index]))]);
-            if (version !== this.version || this.destroyed) return;
-            this.before = images[0]; this.after = images[1];
-            this.ready = true; this.empty.hidden = true;
-            this.dimensionsLabel.textContent = `${this.before.width} × ${this.before.height}`;
-            this.status("");
-            this.updateButton(); this.schedule();
-        } catch (error) { if (version === this.version && !this.destroyed) this.status(error.message, true); }
+            let images;
+            for(let attempt=0;attempt<3;attempt++) {
+                const states=blobs.map(blob=>({loaded:blob?.size||0,total:blob?.size||0,complete:!!blob}));
+                this.loading(attempt?T("Retrying comparison images")+` (${attempt+1}/3)`:T("Loading comparison images"),states);
+                try {
+                    images=await loadPreviewPair([original,this.results[this.index]],{loadPreview:(...args)=>this.options.loadPreview(...args),cached:blobs,signal:controller.signal,
+                        onProgress:(i,state)=>{states[i]=state;if(current())this.loading(T("Loading comparison images"),states);},
+                        onDecoding:()=>{if(current() && states.every(s=>s.complete))this.loading(T("Decoding comparison images…"));}});
+                    break;
+                } catch(error) {
+                    if(!current())return;
+                    if(attempt===2 || [404,410].includes(error.status))throw error;
+                    await new Promise(resolve=>{const done=()=>{clearTimeout(timer);controller.signal.removeEventListener('abort',done);resolve();};const timer=setTimeout(done,1000*(attempt+1));controller.signal.addEventListener('abort',done,{once:true});});
+                }
+            }
+            if(!current()){for(const item of images||[])URL.revokeObjectURL(item.url);return;}
+            for(const url of this.previewUrls||[])URL.revokeObjectURL(url);
+            this.previewUrls=images.map(item=>item.url);this.before=images[0].image;this.after=images[1].image;
+            this.ready=true;this.pendingBlobs=null;this.retryCount=0;this.empty.hidden=true;this.progressBox.hidden=true;
+            this.dimensionsLabel.textContent=`${this.before.width} × ${this.before.height}`;
+            this.status("");this.updateButton();this.schedule();
+        } catch(error) {
+            if(!current())return;
+            this.progressBox.title=error.message;
+            if([404,410].includes(error.status)){this.progressBox.hidden=true;this.pendingBlobs=null;this.status(T("Result preview expired; run the workflow again"),true);return;}
+            this.status(T("Image loading failed; retrying automatically. You can also click reload."),true);
+            const seconds=Math.min(60,5*2**Math.min(this.retryCount||0,4));this.retryCount=(this.retryCount||0)+1;
+            this.loading(T("Waiting to retry comparison images")+` · ${seconds}s`);
+            this.retryTimer=setTimeout(()=>this.loadSelected({background:true}),seconds*1000);
+        } finally {if(version===this.version)this.loadingActive=false;}
     }
 
     async adopt() {
         this.refreshTargets();
-        if (!this.ready || !this.targetSelect.value || this.busy) return;
+        if (this.busy) return;
+        if (!this.ready || this.pendingOutputs>0) { this.status(T("The result preview is still loading"), true); return; }
+        if (!this.targetSelect.value) { this.status(T("Select an inline mask to adopt into"), true); return; }
         const target = this.targetSelect.value, result = this.results[this.index];
         this.busy = true; this.updateButton(); this.status(T("Setting the result as the new source image…"));
+        const controller=this.adoptController=new AbortController();
         try {
-            await this.options.onAdopt(target, result);
+            await this.options.onAdopt(target, result,{signal:controller.signal});
             if (!this.destroyed) this.status(T("Adopted: back in the inline mask, keep painting to inpaint again"));
         } catch (error) { if (!this.destroyed) this.status(error.message, true); }
-        finally { this.busy = false; if (!this.destroyed) this.updateButton(); }
+        finally {if(this.adoptController===controller)this.adoptController=null;this.busy = false; if (!this.destroyed) this.updateButton();}
     }
 
     transform() {
@@ -193,7 +269,10 @@ export class ImageCompareView {
     }
 
     destroy() {
-        this.destroyed = true; ++this.version; cancelAnimationFrame(this.frame); this.observer.disconnect();
-        this.before = this.after = null; this.canvas.width = this.canvas.height = 0; this.root.remove();
+        this.destroyed = true; ++this.version; clearTimeout(this.retryTimer);this.loadController?.abort();
+        this.downloadController?.abort();this.adoptController?.abort();
+        window.removeEventListener("online",this.retryOnOnline);for(const url of this.previewUrls||[])URL.revokeObjectURL(url);
+        cancelAnimationFrame(this.frame); this.observer.disconnect();
+        this.pendingBlobs=null;this.before = this.after = null; this.canvas.width = this.canvas.height = 0; this.root.remove();
     }
 }

web/compare_node.js

@@ -1,54 +1,60 @@
 import { app } from "../../scripts/app.js";
 import { api } from "../../scripts/api.js";
-import { ImageCompareView } from "./compare.js?v=0.5.9";
+import { ImageCompareView } from "./compare.js?v=stability1";
+import { readPreview } from "./preview_transport.js";
+import { installComparisonRecovery } from "./result_recovery.js";
+import {executionId,executionNode,graphNodeEntries} from "./workspace_flow.js";
+import {upstreamNodes} from "./workbench_analysis.js";
 import { appendComparison } from "./compare_history.js?v=0.5.8";
 import { routeWheelOutsideImage } from "./wheel_routing.js";
 import { T } from "./i18n.js";
 import { bindWidgetToNodeWidth } from "./widget_layout.js?v=0.5.5";
 
-function viewUrl(image) {
+function viewPath(image) {
+    if (image?.type === "adachi_memory") return `/adachi-inpaint/compare/${encodeURIComponent(image.filename || "")}`;
     const type = ["input", "output", "temp"].includes(image?.type) ? image.type : "temp";
-    return api.apiURL(`/view?${new URLSearchParams({ filename: image?.filename || "", subfolder: image?.subfolder || "", type })}`);
+    return `/view?${new URLSearchParams({ filename: image?.filename || "", subfolder: image?.subfolder || "", type })}`;
 }
+const viewUrl = image => api.apiURL(viewPath(image));
 
 function targets(node) {
-    const graph = node.graph;
-    const editors = (graph?._nodes || []).filter(candidate => candidate.type === "AdachiInlineMask" && candidate.adachiUseResult)
-        .map(candidate => ({ id: candidate.id, label: `${candidate.title || T("Inline Mask")} #${candidate.id}` }));
-    const sourceLink = graph?.links?.[node.inputs?.find(input => input.name === "original")?.link];
-    const queue = sourceLink ? [sourceLink.origin_id] : [], seen = new Set(), upstream = new Set();
-    while (queue.length && seen.size < 256) {
-        const id = queue.shift(); if (seen.has(id)) continue; seen.add(id);
-        const current = graph.getNodeById(id); if (!current) continue;
-        if (current.type === "AdachiInlineMask") { upstream.add(String(id)); continue; }
-        for (const input of current.inputs || []) {
-            if (!["IMAGE", "*"].includes(input.type)) continue;
-            const link = graph.links?.[input.link]; if (link) queue.push(link.origin_id);
-        }
-    }
+    const graph=app.rootGraph||app.graph;
+    const editors=graphNodeEntries(graph).filter(({node})=>node.type==='AdachiInlineMask' && node.adachiUseResult)
+        .map(({node:editor,id})=>({id,label:`${editor.title||T('Inline Mask')} #${id}`}));
+    const upstream=new Set([...upstreamNodes(node,'original')].filter(n=>n.type==='AdachiInlineMask').map(n=>executionId(graph,n)));
     return { editors, preferred: upstream.size === 1 ? [...upstream][0] : "" };
 }
 
 app.registerExtension({
     name: "Adachi.ImageCompare",
+    setup() { installComparisonRecovery(api,()=>app.rootGraph||app.graph); },
     async beforeRegisterNodeDef(nodeType, nodeData) {
         if (nodeData.name !== "AdachiImageCompare") return;
         const created = nodeType.prototype.onNodeCreated;
         nodeType.prototype.onNodeCreated = function () {
             const value = created?.apply(this, arguments), node = this;
+            const recovery=installComparisonRecovery(api,()=>app.rootGraph||app.graph);
+            node.adachiTrackPrompt=(id,graph=app.rootGraph||app.graph)=>recovery.track(graph,id,executionId(graph,node));
             const compare = new ImageCompareView({
                 resolveUrl: viewUrl,
+                loadPreview: (image,options)=>readPreview(api.fetchApi.bind(api),viewPath(image),options),
+                recover: ()=>recovery.recover(),
+                waitingText: T("Waiting for the generated image…"),
+                failedText: T("Generation failed; check the ComfyUI task history"),
                 onGraphWheel: event => app.canvas?.processMouseWheel(event),
                 getTargets: () => targets(node),
                 onSelect: index => { node.properties.adachi_result_index = index; },
-                async onAdopt(id, image) {
-                    const target = node.graph?.getNodeById(id);
+                async fetchResult(image, options) {
+                    return readPreview(api.fetchApi.bind(api),viewPath(image),options);
+                },
+                async onAdopt(id, image, options={}) {
+                    const graph=app.rootGraph||app.graph,target = executionNode(graph,id);
                     if (!target?.adachiUseResult) throw new Error(T("No inline mask node is available to receive the result"));
-                    const response = await api.fetchApi(`/view?${new URLSearchParams({ filename: image.filename, subfolder: image.subfolder || "", type: image.type || "temp" })}`);
-                    if (!response.ok) throw new Error(T("Result preview expired; run the workflow again"));
-                    const blob = await response.blob();
+                    const blob = await readPreview(api.fetchApi.bind(api),viewPath(image),{...options,maxBytes:64*1024*1024});
+                    options.signal?.throwIfAborted();
+                    if(graph!==(app.rootGraph||app.graph)||target!==executionNode(graph,id))throw Error(T("The workflow changed; check connections again"));
                     if (blob.size > 64 * 1024 * 1024) throw new Error(T("Result image exceeds 64 MiB; downscale it first"));
-                    await target.adachiUseResult(new File([blob], `adachi-result-${Date.now()}.png`, { type: "image/png" }));
+                    await target.adachiUseResult(new File([blob], `adachi-result-${Date.now()}.png`, { type: "image/png" }), image.adachi_origin, options);
                     // Adoption starts a new round. Keep the accepted image visible
                     // as 1/1, then replace it with the next round's first outputs.
                     const data = { originals: [image], results: [image] };
@@ -73,9 +79,13 @@ app.registerExtension({
                 const previous = node.properties.adachi_compare_new_round
                     ? appendComparison() : appendComparison(node.properties.adachi_compare);
                 const data = appendComparison(previous, {
-                    originals: output.adachi_originals, results: output.adachi_results,
+                    originals: output.adachi_originals,
+                    results: output.adachi_results.map(image => output.adachi_origin?.[0]
+                        ? {...image, adachi_origin:output.adachi_origin[0]} : image),
                 });
                 node.properties.adachi_compare = data;
+                node.properties.adachi_latest_result = data.results[data.results.length - 1];
+                compare.setLatestResult(node.properties.adachi_latest_result);
                 node.properties.adachi_compare_new_round = false;
                 node.properties.adachi_result_index = data.results.length > previous.results.length
                     ? data.results.length - 1 : Math.max(0, Math.min(data.results.length - 1, node.properties.adachi_result_index || 0));
@@ -87,6 +97,7 @@ app.registerExtension({
                 const result = configured?.apply(this, arguments);
                 if (node.properties.adachi_compare) {
                     node.properties.adachi_compare = appendComparison(node.properties.adachi_compare);
+                    compare.setLatestResult(node.properties.adachi_latest_result || node.properties.adachi_compare.results.at(-1));
                     node.properties.adachi_result_index = Math.max(0, Math.min(node.properties.adachi_compare.results.length - 1, node.properties.adachi_result_index || 0));
                     compare.setData(node.properties.adachi_compare, node.properties.adachi_result_index);
                 }

web/crop_geometry.js

@@ -0,0 +1,27 @@
+export const CROP_PRESETS = [
+  [1024, 1536],
+  [512, 512],
+  [768, 1024],
+  [1024, 1024],
+  [1536, 2048],
+  [2048, 2048],
+];
+
+export function cropAt(width, height, cropWidth, cropHeight, x, y) {
+  if (
+    ![width, height, cropWidth, cropHeight, x, y].every(Number.isFinite) ||
+    cropWidth < 32 ||
+    cropHeight < 32 ||
+    cropWidth % 32 ||
+    cropHeight % 32 ||
+    cropWidth > width ||
+    cropHeight > height
+  )
+    return null;
+  return {
+    x: Math.max(0, Math.min(width - cropWidth, Math.floor(x))),
+    y: Math.max(0, Math.min(height - cropHeight, Math.floor(y))),
+    width: cropWidth,
+    height: cropHeight,
+  };
+}

web/draft_guard.js

@@ -0,0 +1,110 @@
+// Probe only our own temporary key. Existing ComfyUI drafts are never edited or deleted.
+export const inlineWidgetNames = [
+  "image", "mask_data", "source_data", "edit_mode", "crop_state", "blend_data", "layer_image",
+];
+
+// Some extensions also serialize a named snapshot. Keep it in step with the
+// actual widgets, otherwise a removed embedding can survive in that snapshot.
+export function syncInlineWidgets(data, widgets) {
+  for (const name of inlineWidgetNames) {
+    const index = widgets.findIndex((widget) => widget.name === name);
+    if (index < 0) continue;
+    if (Array.isArray(data.widgets_values)) data.widgets_values[index] = widgets[index].value;
+    if (data.widgets_values_named && typeof data.widgets_values_named === "object")
+      data.widgets_values_named[name] = widgets[index].value;
+  }
+}
+
+function inlineNodes(graph, seen = new Set()) {
+  if (!graph || seen.has(graph)) return [];
+  seen.add(graph);
+  return [
+    ...(graph.nodes || []).filter((node) => node.type === "AdachiInlineMask"),
+    ...(graph.definitions?.subgraphs || []).flatMap((subgraph) => inlineNodes(subgraph, seen)),
+  ];
+}
+
+// Called before graph configuration: native autosave must never see the large
+// source first, since a quota failure disables its storage for the entire tab.
+export async function prepareInlineDraft(graph, { fits, upload }) {
+  const nodes = inlineNodes(graph);
+  for (const node of nodes) {
+    if (!Array.isArray(node.widgets_values)) continue;
+    for (const [index, name] of inlineWidgetNames.entries()) {
+      if (index < node.widgets_values.length && node.widgets_values_named)
+        node.widgets_values_named[name] = node.widgets_values[index];
+    }
+  }
+  const candidates = nodes.filter((node) => {
+    const embedded = node.widgets_values?.[2] ?? node.widgets_values_named?.source_data;
+    return typeof embedded === "string" && embedded.startsWith("data:") && embedded.length >= 256 * 1024;
+  }).sort((a, b) => (b.widgets_values?.[2] ?? b.widgets_values_named.source_data).length - (a.widgets_values?.[2] ?? a.widgets_values_named.source_data).length);
+  let converted = 0;
+  for (const node of candidates) {
+    if (fits(graph) !== false) break;
+    const embedded = node.widgets_values?.[2] ?? node.widgets_values_named.source_data;
+    const filename = await upload(embedded);
+    if (typeof filename !== "string" || !filename) throw new Error("Source upload returned no filename");
+    // Commit only after a successful exact upload. All masks and layers stay intact.
+    if (Array.isArray(node.widgets_values)) {
+      node.widgets_values[0] = filename;
+      node.widgets_values[2] = "";
+    }
+    if (node.widgets_values_named) {
+      node.widgets_values_named.image = filename;
+      node.widgets_values_named.source_data = "";
+    }
+    converted++;
+  }
+  return converted;
+}
+
+export function canStoreDraft(
+  graph,
+  { path = "", workspace = "personal", storage } = {},
+) {
+  try {
+    storage ??= globalThis.localStorage;
+  } catch {
+    return null;
+  }
+  const serialized = JSON.stringify({
+    data: JSON.stringify(graph),
+    updatedAt: Date.now(),
+  });
+  let credit = 0;
+  try {
+    if (path) {
+      const index = JSON.parse(
+        storage.getItem(`Comfy.Workflow.DraftIndex.v2:${workspace}`) || "null",
+      );
+      if (
+        Object.values(index?.entries || {}).some((entry) => entry.path === path)
+      ) {
+        // Standard FNV-1a path hash used by the current ComfyUI draft format.
+        let hash = 2166136261;
+        for (let i = 0; i < path.length; i++)
+          hash = Math.imul(hash ^ path.charCodeAt(i), 16777619);
+        const key = `Comfy.Workflow.Draft.v2:${workspace}:${(hash >>> 0).toString(16).padStart(8, "0")}`;
+        credit = storage.getItem(key)?.length || 0;
+      }
+    }
+  } catch {
+    /* Unknown metadata: a conservative probe still preserves all existing drafts. */
+  }
+  const needed = Math.max(0, serialized.length - credit);
+  if (!needed) return true;
+  const key = `Adachi.DraftProbe:${crypto.randomUUID()}`;
+  try {
+    storage.setItem(key, " ".repeat(needed));
+    return true;
+  } catch (error) {
+    return error?.name === "QuotaExceededError" ? false : null;
+  } finally {
+    try {
+      storage.removeItem(key);
+    } catch {
+      /* Storage may be unavailable. */
+    }
+  }
+}

web/editor.css

@@ -41,6 +41,9 @@
 .dd-inpaint .dd-embed { display: flex; align-items: center; gap: 5px; padding: 4px 10px 0; font-size: 11px; color: var(--dd-muted); cursor: pointer; }
 .dd-inpaint .dd-embed input { accent-color: #e9ab68; margin: 0; }
 .dd-inpaint .dd-status { padding: 5px 11px 8px; color: var(--dd-muted); font-size: 10px; min-height: 27px; overflow-wrap: anywhere; }
+.dd-inpaint .dd-upload-progress { display: flex; flex-direction: column; gap: 4px; padding: 7px 9px; color: var(--dd-text); background: #20242bee; border-radius: 4px; font-size: 11px; pointer-events: none; }
+.dd-inpaint .dd-upload-bar { width: 100%; height: 6px; accent-color: #f3b06a; }
+.dd-inline-minimal > .dd-upload-progress { position: absolute; z-index: 3; top: 34px; left: 5px; width: min(240px, calc(100% - 10px)); }
 .dd-inpaint .dd-error { color: #ff9c92; }
 .dd-inpaint [hidden] { display: none !important; }
 .dd-inpaint .dd-parameters { flex: 0 0 auto; padding: 10px 12px 8px; border-top: 1px solid var(--dd-line); }
@@ -82,6 +85,7 @@
 .dd-inline-minimal > .dd-resize-handle { position: absolute; z-index: 2; bottom: 0; width: 100%; height: 5px; min-height: 5px; border: 0; color: transparent; background: transparent; }
 .dd-inline-minimal > .dd-resize-handle:hover { background: #ffc07966; }
 .dd-inline-minimal > .dd-status:not([hidden]) { position: absolute; bottom: 6px; left: 0; right: 0; background: #20242bee; }
+.dd-inline-minimal.dd-crop-mode > .dd-status:not([hidden]) { bottom: 40px; }
 .dd-editor-controls { height: auto; flex: 0 0 auto; border: 0; border-radius: 0; background: transparent; overflow: visible; }
 .dd-editor-controls .dd-source-info { justify-content: space-between; }
 .dd-editor-controls .dd-filename { padding-left: 0; flex: 1; }
@@ -99,10 +103,12 @@
 .dd-group-presets { border: 0; border-radius: 4px; container-type: inline-size; contain: size layout; }
 .dd-gp-toolbar, .dd-gp-form { display: flex; align-items: center; gap: 7px; padding: 9px 10px; flex: 0 0 auto; }
 .dd-gp-toolbar { justify-content: space-between; border-bottom: 1px solid var(--dd-line); }
+.dd-gp-toolbar-actions { display: flex; align-items: center; gap: 5px; }
 .dd-gp-toolbar strong { font-size: 12px; font-weight: 500; }
 .dd-group-presets button { min-height: 26px; padding: 4px 7px; font-size: 11px; }
 .dd-gp-form input { flex: 1; min-width: 0; color: var(--dd-text); background: var(--dd-panel); border: 1px solid var(--dd-line); padding: 5px 6px; font: inherit; border-radius: 3px; }
 .dd-gp-body { display: grid; grid-template-columns: minmax(130px, 40%) minmax(0, 1fr); flex: 1; min-height: 0; }
+.dd-gp-body.dd-gp-collapsed { grid-template-columns: minmax(0, 1fr); }
 .dd-gp-presets { border-right: 1px solid var(--dd-line); padding: 7px; display: flex; flex-direction: column; gap: 5px; }
 .dd-gp-presets button { text-align: left; flex: 0 0 auto; white-space: normal; overflow-wrap: anywhere; }
 .dd-gp-detail { display: flex; flex-direction: column; min-height: 0; min-width: 0; }
@@ -118,3 +124,30 @@
 .dd-gp-footer { display: flex; flex-wrap: wrap; gap: 5px; padding: 9px 8px; }
 .dd-gp-status { flex: 0 0 auto; padding: 5px 10px 7px; color: var(--dd-muted); font-size: 10px; border-top: 1px solid var(--dd-line); }
 .dd-gp-empty { color: var(--dd-muted); padding: 10px 4px; font-size: 11px; }
+
+/* Compact controls shared by the mask and crop/blend modes. */
+.dd-inline-minimal .dd-mode-select { position:absolute; z-index:3; top:5px; right:5px; max-width:42%; color:var(--dd-text); background:#252b34e8; border:1px solid var(--dd-line); border-radius:4px; height:23px; font:11px "Segoe UI","Microsoft YaHei",sans-serif; padding:1px 3px; pointer-events:auto; }
+.dd-inline-minimal .dd-crop-bar { position:absolute; z-index:3; bottom:7px; left:5px; right:5px; display:flex; gap:3px; flex-wrap:wrap; align-items:center; width:fit-content; max-width:calc(100% - 10px); background:#20242be8; border-radius:4px; padding:3px; pointer-events:auto; }
+.dd-inline-minimal .dd-crop-bar button { min-height:23px; height:23px; padding:1px 6px; min-width:24px; }
+.dd-inline-minimal .dd-crop-bar select,.dd-inline-minimal .dd-crop-custom input { color:var(--dd-text); background:var(--dd-panel); border:1px solid var(--dd-line); border-radius:3px; height:23px; font:11px "Segoe UI","Microsoft YaHei",sans-serif; max-width:112px; }
+.dd-inline-minimal .dd-crop-custom { display:flex; gap:3px; }
+.dd-inline-minimal .dd-crop-custom input { width:66px; }
+.dd-inline-minimal > .dd-files { right:auto; }
+
+
+.dd-workbench .dd-workbench-fold {flex:0 0 auto;border-bottom:1px solid var(--dd-line);}
+.dd-workbench .dd-workbench-fold:not([open]) > :not(summary) {display:none !important;}
+.dd-workbench .dd-workbench-fold>summary {padding:6px 10px;color:var(--dd-muted);font-size:11px;cursor:pointer;list-style:none;display:flex;gap:6px;align-items:center;}
+.dd-workbench .dd-workbench-fold>summary::before {content:"▸";font-size:10px;color:var(--dd-accent);}
+.dd-workbench .dd-workbench-fold[open]>summary::before {content:"▾";}
+.dd-workbench .dd-workbench-fold>summary:hover {color:var(--dd-text);background:#ffffff05;}
+.dd-workbench .dd-workbench-fold>summary:focus-visible {outline:1px solid var(--dd-accent);outline-offset:-2px;}
+.dd-workbench .dd-param-mask-settings {display:flex;flex-direction:column;gap:8px;padding:7px 12px 9px;}
+.dd-workbench .dd-workbench-actions {flex:0 0 auto;}
+.dd-workbench .dd-workbench-actions:empty {display:none;}
+.dd-workbench .dd-workbench-actions .dd-actions {padding:6px 10px;gap:4px;}
+.dd-workbench .dd-workbench-actions .dd-actions button {min-height:24px;padding:3px 7px;}
+
+/* Network progress belongs to the image surface, never to the graph canvas. */
+.dd-comparer .dd-preview-progress {position:absolute;left:10px;right:10px;bottom:12px;z-index:4;max-width:380px;padding:9px 12px;border:1px solid var(--dd-line);border-radius:5px;background:var(--dd-bg);color:var(--dd-text);pointer-events:none;font-size:12px;}
+.dd-comparer .dd-preview-progress progress {display:block;width:100%;height:7px;margin-top:6px;accent-color:var(--dd-accent);}

web/editor.js

@@ -94,7 +94,8 @@ export class InlineMaskEditor {
         head.append(this.sizeLabel);
         this.root.append(head);
         const files = el("div", "dd-row dd-files");
-        this.loadButton = this.button(files, T("Upload"), T("Upload an image, or drop / paste one"), () => this.imageInput.click(), "dd-primary");
+        this.loadButton = this.button(files, "↑", T("Upload an image, or drop / paste one"), () => this.imageInput.click(), "dd-primary");
+        this.downloadButton = this.button(files, "↓", T("Download the source image"), () => this.downloadSource(), "dd-primary");
         this.filename = el("span", "dd-filename", T("Drop an image onto the canvas"));
         files.append(this.filename);
         this.root.append(files);
@@ -121,7 +122,7 @@ export class InlineMaskEditor {
         const tools = el("div", "dd-row dd-tools");
         this.toolButtons = {};
         for (const [name, label, title] of [
-            ["brush", "Brush", "Brush (B)"], ["erase", "Eraser", "Eraser (E); hold Alt or right-click for temporary erase"],
+            ["brush", "Brush", "Brush (B)"], ["erase", "Eraser", "Eraser (E); hold Alt for temporary erase"],
             ["rect", "Rectangle", "Rectangle (R)"], ["lasso", "Lasso", "Freehand lasso (L)"],
             ["pan", "Pan", "Pan (H), or hold Space"],
         ]) this.toolButtons[name] = this.button(tools, T(label), T(title), () => this.setTool(name));
@@ -138,8 +139,17 @@ export class InlineMaskEditor {
         this.empty = el("div", "dd-empty");
         this.empty.append(el("span", "dd-empty-symbol", "+"), el("strong", "", T("Drop an image here")), el("span", "", T("Paint the areas to repaint directly in this node")));
         this.stage.append(this.empty);
-        this.root.append(this.stage, settings);
-        this.canvas.oncontextmenu = event => event.preventDefault();
+        this.uploadIndicator = el("div", "dd-upload-progress");
+        this.uploadIndicator.hidden = true;
+        this.uploadIndicator.setAttribute("role", "status");
+        this.uploadIndicator.setAttribute("aria-live", "polite");
+        this.uploadLabel = el("span", "dd-upload-label");
+        this.uploadBar = el("progress", "dd-upload-bar");
+        this.uploadBar.max = 100;
+        this.uploadBar.setAttribute("aria-label", T("Image upload progress"));
+        this.uploadIndicator.append(this.uploadLabel, this.uploadBar);
+        this.root.append(this.stage, settings, this.uploadIndicator);
+        this.canvas.oncontextmenu = event => { event.preventDefault(); event.stopPropagation(); };
         this.canvas.addEventListener("pointerdown", event => this.pointerDown(event));
         this.canvas.addEventListener("pointermove", event => this.pointerMove(event));
         this.canvas.addEventListener("pointerup", event => this.pointerUp(event));
@@ -154,6 +164,7 @@ export class InlineMaskEditor {
             if (file) this.upload(file);
         });
         const actions = el("div", "dd-row dd-actions");
+        this.actions = actions;
         this.undoButton = this.button(actions, T("Undo"), T("Undo (Ctrl+Z)"), () => this.undo());
         this.redoButton = this.button(actions, T("Redo"), T("Redo (Ctrl+Shift+Z)"), () => this.redo());
         this.button(actions, T("Invert"), T("Invert the painted mask"), () => this.action({ type: "invert" }));
@@ -172,8 +183,8 @@ export class InlineMaskEditor {
         footer.append(spacer);
         this.zoomLabel = el("span", "dd-zoom", "100%");
         footer.append(this.zoomLabel);
-        this.fitButton = this.button(footer, T("Fit"), T("Fit the image to the canvas"), () => this.fit());
-        this.keepButton = this.button(footer, "", T("Keep"), event => { event.stopPropagation(); this.setKeepFit(!this.keepFit); });
+        this.fitButton = this.button(footer, "⛶", T("Fit the image to the canvas"), () => this.fit());
+        this.keepButton = this.button(footer, "", T("Pin the current view"), event => { event.stopPropagation(); this.setKeepFit(!this.keepFit); });
         this.keepButton.innerHTML = '<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="m16 3 5 5-4 1-4 4 1 4-3 3-3-6-6-3 3-3 4 1 4-4zM8 16l-6 6"/></svg>';
         // The inline button must not select/bring forward the ComfyUI node:
         // newer frontends can remeasure its DOM widget on that bubbling click.
@@ -197,8 +208,10 @@ export class InlineMaskEditor {
             const info = el("div", "dd-row dd-source-info"); info.append(this.filename, this.sizeLabel);
             head.remove();
             this.controls.append(info, tools, settings, actions, footer, embedRow);
+            actions.addEventListener("keydown", event => this.keydown(event));
             this.loadButton.after(this.fitButton);
             this.fitButton.after(this.keepButton);
+            this.loadButton.after(this.downloadButton);
             // These are the original live controls; the sampler mounts this
             // same DOM tree so brush state and callbacks are preserved.
         }
@@ -223,24 +236,57 @@ export class InlineMaskEditor {
     status(message, error = false) {
         this.message.textContent = message;
         this.message.classList.toggle("dd-error", error);
+        // Only errors stay visible; transient hints and progress lines hide.
         this.message.hidden = !error;
     }
 
+    notice(message) {
+        // An explicit non-error notice that should remain visible.
+        this.message.textContent = message;
+        this.message.classList.remove("dd-error");
+        this.message.hidden = false;
+    }
+
     async upload(file) {
         if (file.size > 64 * 1024 * 1024) { this.status(T("File is too large (max 64 MB)"), true); return; }
+        this.uploadController?.abort();
+        const controller=this.uploadController=new AbortController();
+        const version = this.uploadUiVersion = (this.uploadUiVersion || 0) + 1;
+        const progress = ({phase, loaded = 0, total = 0}) => {
+            if (version !== this.uploadUiVersion || this.destroyed) return;
+            this.uploadIndicator.hidden = false;
+            if (phase === "upload" && total > 0) {
+                const percent = Math.max(0, Math.min(100, Math.round(100 * loaded / total)));
+                this.uploadLabel.textContent = `${T("Uploading image")} ${percent}%`;
+                this.uploadBar.value = percent;
+            } else {
+                this.uploadLabel.textContent = T(phase === "processing" ? "Upload complete, loading image…" : phase === "authenticated" ? "Uploading with authentication…" : "Uploading image…");
+                this.uploadBar.removeAttribute("value");
+            }
+        };
+        progress({phase: "upload"});
         try {
             this.status(T("Loading image…"));
-            await this.options.onUpload?.(file);
-        } catch (error) { this.status(error.message, true); }
+            // Keep the painted mask when a new source replaces the current one,
+            // the same way adopting a result does. A first upload starts empty.
+            await this.options.onUpload?.(file, { preserveMask: !!this.image, onProgress: progress, signal:controller.signal });
+        } catch (error) {
+            if (version === this.uploadUiVersion && !this.destroyed) this.status(error.message, true);
+        } finally {
+            if(this.uploadController===controller)this.uploadController=null;
+            if (version === this.uploadUiVersion && !this.destroyed) this.uploadIndicator.hidden = true;
+        }
     }
 
     async setImage(url, maskData = "", name = "") {
+        this.stroke = null; this.pointerId = null;
         const version = ++this.loadVersion;
+        const wasReady=this.ready;
         this.ready = false;
         try {
             const [image, loadedMask] = await Promise.all([loadImage(url), maskData ? loadImage(maskData) : Promise.resolve(null)]);
             if (version !== this.loadVersion || this.destroyed) return;
-            if (image.width * image.height > 32 * 1024 * 1024 || Math.max(image.width, image.height) > 16384) {
+            if (image.width * image.height > 32_000_000 || Math.max(image.width, image.height) > 16384) {
                 throw new Error(T("Image is too large; keep it under 32 MP and 16384 px per side"));
             }
             this.image = image;
@@ -274,7 +320,7 @@ export class InlineMaskEditor {
             this.options.onImage?.(image);
             this.status(T("Orange is the painted mask · Alt = temporary erase · Ctrl+wheel = brush size"));
         } catch (error) {
-            if (version === this.loadVersion && !this.destroyed) this.status(error.message, true);
+            if (version === this.loadVersion && !this.destroyed) {this.ready=!!this.image && wasReady;this.updateButtons();this.schedule();this.status(error.message, true);}
             throw error;
         }
     }
@@ -282,10 +328,10 @@ export class InlineMaskEditor {
     async replaceSource(url, name = "") {
         const version = ++this.loadVersion;
         const image = await loadImage(url);
-        if (version !== this.loadVersion || this.destroyed) return;
-        if (image.width !== this.mask.width || image.height !== this.mask.height) {
-            throw new Error(T("The result size must match the source image to preserve the mask"));
-        }
+        if (version !== this.loadVersion || this.destroyed) return false;
+        // A different size cannot keep the painted mask; report it so the caller
+        // can fall back to a fresh load instead of failing the upload.
+        if (image.width !== this.mask.width || image.height !== this.mask.height) return false;
         // Replacing only the source preserves mask pixels, the base snapshot,
         // command history, redo history, active strokes and the current view.
         this.flush();
@@ -293,6 +339,7 @@ export class InlineMaskEditor {
         this.filename.textContent = name || T("Embedded image");
         this.filename.title = this.filename.textContent;
         this.updateButtons(); this.schedule();
+        return true;
     }
 
     importMask(ctx, image) {
@@ -322,6 +369,7 @@ export class InlineMaskEditor {
     updateButtons() {
         this.keepButton?.classList.toggle("dd-active", !!this.keepFit);
         this.keepButton?.setAttribute("aria-pressed", String(!!this.keepFit));
+        if (this.downloadButton) this.downloadButton.disabled = !this.image;
         this.stage?.classList.toggle("dd-erasing", !!(this.altErase || this.tool === "erase" || this.stroke?.type === "erase"));
         for (const [name, button] of Object.entries(this.toolButtons || {})) {
             const active = name === (this.altErase ? "erase" : this.tool);
@@ -338,7 +386,6 @@ export class InlineMaskEditor {
 
     dimensions() {
         const r = this.canvas.getBoundingClientRect();
-        if (this.keepFit) { this.zoom = 1; this.pan = { x: 0, y: 0 }; }
         if (this.viewport?.width && this.viewport?.height && r.width && r.height) {
             this.pan.x *= r.width / this.viewport.width;
             this.pan.y *= r.height / this.viewport.height;
@@ -355,8 +402,8 @@ export class InlineMaskEditor {
     }
 
     pointerDown(event) {
+        if (event.button === 2) { event.preventDefault(); event.stopPropagation(); return; }
         if (!this.ready || this.pointerId != null) return;
-        if (this.keepFit && (this.space || (!event.altKey && this.tool === "pan") || event.button === 1)) return;
         event.preventDefault(); event.stopPropagation(); this.root.focus({ preventScroll: true });
         this.altErase = !!event.altKey; this.updateButtons();
         this.pointerId = event.pointerId;
@@ -367,7 +414,7 @@ export class InlineMaskEditor {
             return;
         }
         if (this.view === "image") { this.view = "overlay"; this.updateButtons(); }
-        this.stroke = { type: event.button === 2 || event.altKey ? "erase" : this.tool, size: this.brushSize, hardness: this.hardness, opacity: this.brushOpacity, points: [this.cursor] };
+        this.stroke = { type: event.altKey ? "erase" : this.tool, size: this.brushSize, hardness: this.hardness, opacity: this.brushOpacity, points: [this.cursor] };
         this.updateButtons();
         if (["brush", "erase"].includes(this.stroke.type)) this.paintSegment(this.mask.getContext("2d"), this.stroke, this.cursor, this.cursor);
         this.schedule();
@@ -385,7 +432,7 @@ export class InlineMaskEditor {
             } else if (this.stroke) {
                 const events = event.getCoalescedEvents?.() || [event];
                 if (["brush", "erase"].includes(this.stroke.type)) {
-                    const type = event.altKey || (event.buttons & 2) ? "erase" : this.tool;
+                    const type = event.altKey ? "erase" : this.tool;
                     if (["brush", "erase"].includes(type) && type !== this.stroke.type) {
                         // Keep separate commands when Alt is pressed/released mid-stroke.
                         const pivot = this.stroke.points.at(-1);
@@ -426,6 +473,8 @@ export class InlineMaskEditor {
     paintSegment(ctx, command, a, b) {
         if (ctx.canvas === this.mask) this.maskRevision++;
         ctx.save();
+        const bounds = this.options.paintBounds?.();
+        if (bounds) { ctx.beginPath(); ctx.rect(bounds.x, bounds.y, bounds.width, bounds.height); ctx.clip(); }
         ctx.globalCompositeOperation = command.type === "erase" ? "destination-out" : "source-over";
         ctx.globalAlpha = command.opacity / 100;
         if (command.hardness >= 99 && a.p === 1 && b.p === 1) {
@@ -452,6 +501,9 @@ export class InlineMaskEditor {
     }
 
     drawCommand(ctx, command) {
+        ctx.save();
+        const bounds = this.options.paintBounds?.();
+        if (bounds) { ctx.beginPath(); ctx.rect(bounds.x, bounds.y, bounds.width, bounds.height); ctx.clip(); }
         const w = this.mask.width, h = this.mask.height;
         if (["brush", "erase"].includes(command.type)) {
             this.paintSegment(ctx, command, command.points[0], command.points[0]);
@@ -460,7 +512,14 @@ export class InlineMaskEditor {
         else if (command.type === "fill") { ctx.save(); ctx.fillStyle = "white"; ctx.fillRect(0, 0, w, h); ctx.restore(); }
         else if (command.type === "invert") {
             ctx.save(); ctx.globalCompositeOperation = "xor"; ctx.fillStyle = "white"; ctx.fillRect(0, 0, w, h); ctx.restore();
-        } else if (command.type === "import") this.importMask(ctx, command.image);
+        } else if (command.type === "import") {
+            if (!bounds) this.importMask(ctx, command.image);
+            else {
+                const imported = makeCanvas(); imported.width = w; imported.height = h;
+                this.importMask(imported.getContext("2d"), command.image);
+                ctx.clearRect(0, 0, w, h); ctx.drawImage(imported, 0, 0); imported.width = imported.height = 0;
+            }
+        }
         else if (["rect", "lasso"].includes(command.type)) {
             const a = command.points[0], b = command.points.at(-1);
             ctx.save(); ctx.fillStyle = "white"; ctx.globalAlpha = command.opacity / 100;
@@ -468,12 +527,18 @@ export class InlineMaskEditor {
             else { ctx.beginPath(); ctx.moveTo(a.x, a.y); for (const p of command.points) ctx.lineTo(p.x, p.y); ctx.closePath(); ctx.fill(); }
             ctx.restore();
         }
+        ctx.restore();
     }
 
     remember(command) {
         this.history.push(command); this.future = [];
         // Keep commands rather than full-resolution pixel snapshots; memory stays small.
-        if (this.history.length > 500) this.drawCommand(this.baseMask.getContext("2d"), this.history.shift());
+        let bitmapBytes = this.history.reduce((total, item) => total + (item.image ? item.image.width * item.image.height * 4 : 0), 0);
+        while (this.history.length > 500 || (this.history.length > 1 && bitmapBytes > 64 * 1024 * 1024)) {
+            const first = this.history.shift();
+            this.drawCommand(this.baseMask.getContext("2d"), first);
+            bitmapBytes -= first.image ? first.image.width * first.image.height * 4 : 0;
+        }
         this.updateButtons();
     }
 
@@ -534,11 +599,25 @@ export class InlineMaskEditor {
         const link = el("a"); link.download = "adachi-mask.png"; link.href = this.flush(); link.click();
     }
 
+    downloadSource() {
+        if (!this.image) return;
+        const out = makeCanvas();
+        out.width = this.image.naturalWidth || this.image.width;
+        out.height = this.image.naturalHeight || this.image.height;
+        out.getContext("2d").drawImage(this.image, 0, 0);
+        const label = String(this.filename.textContent || "").split(/[\\/]/).pop().replace(/\.[^.]+$/, "");
+        const link = el("a");
+        link.download = `${label || "adachi-source"}.png`;
+        link.href = out.toDataURL("image/png");
+        link.click();
+        out.width = out.height = 0;
+    }
+
     fit() { this.zoom = 1; this.pan = { x: 0, y: 0 }; this.schedule(); }
 
     setKeepFit(enabled, notify = true) {
+        // Pin freezes the view exactly as drawn; it must not re-fit the image.
         this.keepFit = !!enabled;
-        if (this.keepFit) this.fit();
         this.updateButtons(); this.schedule();
         if (notify) this.options.onKeepFit?.(this.keepFit);
     }
@@ -551,6 +630,12 @@ export class InlineMaskEditor {
     wheel(event) {
         if (!this.ready) return;
         event.preventDefault(); event.stopPropagation();
+        if (event.shiftKey) {
+            const delta = event.deltaY * (event.deltaMode === 1 ? 16 : event.deltaMode === 2 ? 100 : 1);
+            this.hardnessInput.value = Math.max(0, Math.min(100, this.hardness + (delta < 0 ? 5 : delta > 0 ? -5 : 0)));
+            this.hardnessInput.dispatchEvent(new Event("input"));
+            this.cursor = this.point(event); this.schedule(); return;
+        }
         if (this.keepFit && !event.ctrlKey) { this.options.onGraphWheel?.(event); return; }
         if (event.ctrlKey) {
             const delta = event.deltaY * (event.deltaMode === 1 ? 16 : event.deltaMode === 2 ? 100 : 1);
@@ -586,6 +671,7 @@ export class InlineMaskEditor {
     schedule() { if (!this.frame && !this.destroyed) this.frame = requestAnimationFrame(() => { this.frame = 0; this.render(); }); }
 
     render() {
+        const view = this.options.displayView?.() || this.view;
         const rect = this.canvas.getBoundingClientRect();
         if (!rect.width || !rect.height) return;
         const d = this.dimensions();
@@ -595,27 +681,28 @@ export class InlineMaskEditor {
         if (this.canvas.width !== w || this.canvas.height !== h) { this.canvas.width = w; this.canvas.height = h; this.overlay.width = w; this.overlay.height = h; }
         const display = this.canvas.getContext("2d");
         if (!this.ready) { display.clearRect(0, 0, w, h); this.sceneKey = null; return; }
-        const key = [this.loadVersion, this.maskRevision, w, h, d.scale * ratio, d.x * ratio, d.y * ratio, this.view, this.overlayOpacity].join(":");
+        const key = [this.loadVersion, this.maskRevision, w, h, d.scale * ratio, d.x * ratio, d.y * ratio, view, this.overlayOpacity, this.options.sceneRevision?.() || 0].join(":");
         if (key !== this.sceneKey) {
             if (this.scene.width !== w || this.scene.height !== h) { this.scene.width = w; this.scene.height = h; }
             const ctx = this.scene.getContext("2d");
             ctx.setTransform(1, 0, 0, 1, 0, 0); ctx.clearRect(0, 0, w, h);
             ctx.setTransform(ratio * d.scale, 0, 0, ratio * d.scale, ratio * d.x, ratio * d.y);
-            if (this.view !== "mask") ctx.drawImage(this.image, 0, 0);
+            if (view !== "mask") { if (!this.options.drawSource?.(ctx)) ctx.drawImage(this.image, 0, 0); }
             else { ctx.fillStyle = "black"; ctx.fillRect(0, 0, this.image.width, this.image.height); }
-            if (this.view !== "image") {
+            if (view !== "image" && (view === "mask" || this.options.showMask?.() !== false)) {
                 const overlay = this.overlay.getContext("2d");
                 overlay.setTransform(1, 0, 0, 1, 0, 0); overlay.clearRect(0, 0, w, h);
                 overlay.setTransform(ratio * d.scale, 0, 0, ratio * d.scale, ratio * d.x, ratio * d.y);
                 overlay.drawImage(this.mask, 0, 0);
-                if (this.view === "overlay") { overlay.globalCompositeOperation = "source-in"; overlay.fillStyle = "#ffad45"; overlay.fillRect(0, 0, this.mask.width, this.mask.height); overlay.globalCompositeOperation = "source-over"; }
-                ctx.setTransform(1, 0, 0, 1, 0, 0); ctx.globalAlpha = this.view === "mask" ? 1 : this.overlayOpacity / 100; ctx.drawImage(this.overlay, 0, 0); ctx.globalAlpha = 1;
+                if (view === "overlay") { overlay.globalCompositeOperation = "source-in"; overlay.fillStyle = "#ffad45"; overlay.fillRect(0, 0, this.mask.width, this.mask.height); overlay.globalCompositeOperation = "source-over"; }
+                ctx.setTransform(1, 0, 0, 1, 0, 0); ctx.globalAlpha = view === "mask" ? 1 : this.overlayOpacity / 100; ctx.drawImage(this.overlay, 0, 0); ctx.globalAlpha = 1;
             }
             this.sceneKey = key;
         }
         const ctx = display;
         ctx.setTransform(1, 0, 0, 1, 0, 0); ctx.clearRect(0, 0, w, h); ctx.drawImage(this.scene, 0, 0);
         ctx.setTransform(ratio * d.scale, 0, 0, ratio * d.scale, ratio * d.x, ratio * d.y);
+        this.options.drawOverlay?.(ctx, d);
         if (this.stroke && ["rect", "lasso"].includes(this.stroke.type)) {
             ctx.save(); ctx.strokeStyle = "#ffcf85"; ctx.lineWidth = 1.5 / d.scale; ctx.setLineDash([5 / d.scale, 4 / d.scale]);
             const a = this.stroke.points[0], b = this.stroke.points.at(-1);
@@ -623,7 +710,7 @@ export class InlineMaskEditor {
             else { ctx.beginPath(); ctx.moveTo(a.x, a.y); for (const p of this.stroke.points) ctx.lineTo(p.x, p.y); ctx.stroke(); }
             ctx.restore();
         }
-        if (this.cursor && (this.altErase || ["brush", "erase"].includes(this.tool)) && !this.drag) {
+        if (this.cursor && (this.altErase || ["brush", "erase"].includes(this.tool)) && !this.drag && this.options.showCursor?.() !== false) {
             const erasing = this.altErase || this.tool === "erase" || this.stroke?.type === "erase";
             ctx.save();
             if (erasing) ctx.setLineDash([4 / d.scale, 3 / d.scale]);
@@ -635,6 +722,7 @@ export class InlineMaskEditor {
     }
 
     destroy() {
+        this.uploadController?.abort();++this.uploadUiVersion;
         this.destroyed = true; ++this.loadVersion;
         cancelAnimationFrame(this.frame); this.observer.disconnect();
         this.stopWheelCapture();

web/group_presets.js

@@ -19,7 +19,15 @@ class GroupPresetsPanel {
             this.refresh(true);
         });
         const toolbar = el("div", "dd-gp-toolbar"); toolbar.append(el("strong", "", T("Group presets")));
-        this.add = this.button(toolbar, T("+ New preset"), () => this.showForm("add"));
+        const tools = el("div", "dd-gp-toolbar-actions"); toolbar.append(tools);
+        this.toggleDetail = this.button(tools, "", () => {
+            app.canvas?.emitBeforeChange?.();
+            this.node.properties.adachi_group_detail_hidden = !this.node.properties.adachi_group_detail_hidden;
+            this.updateDetail();
+            app.canvas?.emitAfterChange?.();
+            this.node.graph?.setDirtyCanvas(true, true);
+        });
+        this.add = this.button(tools, T("+ New preset"), () => this.showForm("add"));
         this.form = el("form", "dd-gp-form"); this.form.hidden = true;
         this.nameInput = el("input"); this.nameInput.type = "text"; this.nameInput.maxLength = 120;
         this.nameInput.setAttribute("aria-label", T("Preset name"));
@@ -28,9 +36,9 @@ class GroupPresetsPanel {
         this.button(this.form, T("Cancel"), () => { this.form.hidden = true; });
         this.form.onsubmit = event => { event.preventDefault(); this.submitName(); };
         this.nameInput.onkeydown = event => { event.stopPropagation(); if (event.key === "Escape") { event.preventDefault(); this.form.hidden = true; } };
-        const body = el("div", "dd-gp-body");
+        const body = this.body = el("div", "dd-gp-body");
         this.presets = el("nav", "dd-gp-presets dd-gp-scroll"); this.presets.setAttribute("aria-label", T("Presets"));
-        const detail = el("section", "dd-gp-detail");
+        const detail = this.detail = el("section", "dd-gp-detail");
         const head = el("div", "dd-gp-detail-head"); this.title = el("strong"); head.append(this.title);
         this.rename = this.button(head, T("Rename"), () => this.showForm("rename"));
         this.groups = el("div", "dd-gp-groups dd-gp-scroll"); this.groups.setAttribute("aria-label", T("Group states"));
@@ -50,7 +58,16 @@ class GroupPresetsPanel {
         };
         window.addEventListener("wheel", this.onWheel, { capture: true, passive: false });
         this.timer = setInterval(() => { if (this.root.isConnected && !document.hidden) this.refresh(); }, 500);
-        this.refresh(true);
+        this.updateDetail(); this.refresh(true);
+    }
+    updateDetail() {
+        const hidden = !!this.node.properties.adachi_group_detail_hidden;
+        this.detail.hidden = hidden;
+        this.body.classList.toggle("dd-gp-collapsed", hidden);
+        this.toggleDetail.textContent = hidden ? "▶" : "◀";
+        this.toggleDetail.title = T(hidden ? "Show group details" : "Hide group details");
+        this.toggleDetail.setAttribute("aria-label", this.toggleDetail.title);
+        this.toggleDetail.setAttribute("aria-expanded", String(!hidden));
     }
     button(parent, label, callback) { const button = el("button", "", label); button.type = "button"; button.onclick = event => { event.stopPropagation(); callback(); }; parent.append(button); return button; }
     showForm(mode) {
@@ -122,7 +139,7 @@ app.registerExtension({
                 widget.serialize = false; this.setSize([620, 410]);
             }
             onAdded() { this.adachiGroupPresets.refresh(true); }
-            onConfigure() { this.adachiGroupPresets.model.restore(); this.adachiGroupPresets.refresh(true); }
+            onConfigure() { this.adachiGroupPresets.model.restore(); this.adachiGroupPresets.updateDetail(); this.adachiGroupPresets.refresh(true); }
             onRemoved() { this.adachiGroupPresets.destroy(); }
             getExtraMenuOptions(_, options) {
                 const panel = this.adachiGroupPresets;

web/i18n.js

@@ -6,6 +6,185 @@
 // buttons are drawn by this plugin, so they translate here by reading Comfy.Locale.
 
 const ZH = {
+  "Image upload timed out; check the connection and try again": "图片上传超时，请检查连接后重试",
+  "Load a source image first": "请先载入原图",
+  "Finish the current edit first": "请先结束当前编辑",
+  "The source image node was removed": "原图节点已移除",
+  "The image changed during restoration; try again": "图片在恢复过程中发生变化，请重试",
+  "The saved mask state is invalid": "历史蒙版状态无效",
+  "Source image": "原图",
+  "Edited source": "编辑中的原图",
+  "Processing the source image; please wait": "正在处理原图，请稍候",
+  "The source image history entry no longer exists": "原图历史不存在",
+  // Workbench panels, execution status and accessibility labels.
+  "Reading the current workflow…": "正在读取当前工作流…",
+  "Switch theme": "切换主题",
+  "Check connections": "检查连接",
+  "Return to workflow": "返回工作流",
+  "Edit mask": "编辑蒙版",
+  "Slide to compare": "滑动对比",
+  "Show before": "查看前图",
+  "Show after": "查看后图",
+  "Adopt": "采用",
+  "Use this result as the next source image and keep the mask; restore earlier images below": "采用当前结果作为下一轮原图，保留蒙版；下方可回退或选择之前的原图",
+  "Selected source image or generated result": "当前选择的前图或重绘结果",
+  "↶ Undo": "↶ 回退",
+  "Initial source": "初始原图",
+  "Choose an adopted source image": "选择已采用原图",
+  "Restore the previous source image, mask and crop": "回退到之前的原图，同时恢复蒙版和裁切状态",
+  "Restore the initial source image of this session": "恢复本次工作流会话中的初始原图",
+  "Choose an earlier adopted or uploaded source image": "选择之前采用或上传的原图",
+  "Current stage progress": "当前执行阶段进度",
+  "Actual progress from the current node; stages without step counts show an indeterminate bar": "显示当前节点的真实进度；没有步数的阶段显示等待动画",
+  "Workflow settings": "工作流设置",
+  "Generate": "生成",
+  "Generating": "生成中",
+  "Generation complete": "生成完成",
+  "Controls are not ready; reopen the workbench shortly": "控件尚未初始化，请稍后重新打开",
+  "This workflow needs Inline Mask, Adachi Inpaint Workbench and Image Compare nodes": "当前工作流需要内联蒙版、安达重绘工作台和图像对比节点",
+  "Source when opened": "打开时原图",
+  "Adopted result": "采用重绘结果",
+  "Uploaded source": "上传原图",
+  "Generation stopped": "生成已停止",
+  "Generating…": "生成中…",
+  "Workflow connections": "工作流连接",
+  "Multiple candidates found; select the nodes to use": "存在多个候选，请选择本次使用的节点",
+  "Inpaint node": "重绘节点",
+  "Image editor": "图像编辑",
+  "Image comparison": "图像对比",
+  "No preset controller": "不使用预设",
+  "Use these connections": "使用此连接",
+  "Select nodes from this workflow; this inpaint node remembers your choices": "选择当前工作流中的节点；同一重绘节点会记住本次选择",
+  "The workflow changed": "工作流已切换",
+  "Select the nodes to use in Workflow connections first": "请先在工作流连接中选择本次使用的节点",
+  "Upload a source image first": "请先上传原图",
+  "Checking in the background…": "后台检查中…",
+  "Workflow validation failed": "工作流检查未通过",
+  "Ready to generate": "检查通过 · 可以生成",
+  "Checking…": "检查中…",
+  "Queued for generation": "已进入生成队列",
+  "Generation failed; check the ComfyUI error log": "生成失败，请查看 ComfyUI 错误信息",
+  "Switch to light theme": "切换为白色主题",
+  "Switch to dark theme": "切换为黑色主题",
+  "Current workflow": "当前工作流",
+  "Reading workflow connections": "正在读取工作流连接",
+  "Actual sampler connections": "实际采样器连线",
+  "Main model": "主模型",
+  "Text encoder": "文本编码器",
+  "The group preset no longer exists": "分组预设不存在",
+  "Reading workflow connections; please wait": "正在读取工作流连接，请稍候",
+  "Some model lists are unavailable; existing workflow selections remain usable": "部分模型列表不可用，可保留工作流中的现有选择",
+  "Multiple group preset controllers found; choose one in Workflow connections.": "存在多个分组预设控制器，请在工作流连接中选择。",
+  "No enabled Group Bypass Presets node; using the current workflow settings.": "未启用组忽略预设节点，当前使用原工作流设置。",
+  "Main model and LoRA": "主模型与 LoRA",
+  "This profile uses the model from the workflow": "当前方案的模型由原工作流控制",
+  "Inherit workflow LoRAs": "沿用工作流 LoRA",
+  "Keep existing LoRA nodes and both CLIP branches; uncheck to use the custom list below. An empty custom list disables recognized LoRAs.": "保留原有 LoRA 节点及正负 CLIP 分支；取消后使用下方自定义列表，空列表即禁用已识别的 LoRA",
+  "No active LoRAs in the original workflow": "原工作流没有启用的 LoRA",
+  "No LoRA": "不使用 LoRA",
+  "None": "不使用",
+  "Remove this LoRA": "移除此 LoRA",
+  "＋ Add LoRA": "＋ 添加 LoRA",
+  "Use the workflow Refiner (must be compatible with the main model)": "使用工作流中的 Refiner（需与主模型兼容）",
+  "Positive prompt": "正面提示词",
+  "Negative prompt": "负面提示词",
+  "Controlled by the original workflow": "由原工作流控制",
+  "Reading connections…": "正在读取连接…",
+  "Encoders are identified from the execution graph; unrecognized conditioning stays controlled by the original workflow": "通过实际执行图查找编码器；无法明确识别的条件保留原工作流",
+  "Inherit upstream dynamic text; type to override": "沿用上游动态文本；输入后覆盖",
+  "Text comes from upstream; typing creates an override.": "文本来自上游；主动输入才覆盖。",
+  "Changes affect only this encoder; downstream conditioning is preserved.": "修改仅影响所选编码器，保留下游条件处理。",
+  "Loading main model": "加载主模型",
+  "Loading text encoder": "加载文本编码器",
+  "Loading VAE": "加载 VAE",
+  "Encoding prompts": "编码提示词",
+  "Loading LoRA": "加载 LoRA",
+  "Reading source image and mask": "读取原图与蒙版",
+  "Preparing inpaint sampling": "重绘采样准备",
+  "KSampler": "K 采样器",
+  "VAE decode": "VAE 解码",
+  "VAE encode": "VAE 编码",
+  "Preparing comparison images": "准备对比图片",
+  "Queued…": "排队中…",
+  "Executing workflow": "正在执行工作流",
+  "Reconnected; waiting for execution progress…": "已重连，等待执行进度…",
+  "Connection lost; recovering execution progress…": "连接中断，正在恢复进度…",
+  "Starting execution…": "开始执行…",
+  "Inpaint sampling (KSampler)": "重绘采样（K 采样器）",
+  "Resize canvas and settings": "调整画布与配置宽度",
+  "Drag to resize the canvas and settings; double-click to reset, or use arrow keys for fine adjustments": "拖动调整画布与配置宽度；双击恢复，方向键微调",
+  "The workbench backend is not loaded. Stop and restart ComfyUI, then check connections again; refreshing the browser is not enough.": "工作台后端尚未加载：请在 ComfyUI 启动器中停止后重新启动服务，再点击检查连接（仅刷新网页不够）",
+  "Submission timed out. Check the task queue before retrying to avoid duplicate generations.": "提交超时，请先检查任务队列，避免重复生成",
+  "Connection check timed out; check your network and retry": "连接检查超时，请检查网络后重试",
+  "The workflow changed; reopen the workbench": "工作流已切换，请重新打开工作台",
+  "Workflow connections could not be resolved; check bypass states, Get/Set and broadcast connections": "工作流连线无法解析，请检查忽略状态、Get/Set 或广播连线",
+  "The workflow changed; check connections again": "工作流已切换，请重新检查",
+  "The selected image, inpaint or comparison node is missing from the execution graph": "选中的图像、重绘或对比节点未出现在执行图中",
+  "The selected inpaint node is missing from the execution graph": "所选重绘节点未出现在执行图中",
+  "Connect a main model before adding LoRAs": "请连接主模型后再添加 LoRA",
+  "Open Adachi Workbench": "打开安达工作台",
+  "Text encoder {index}": "文本编码器 {index}",
+  "Model list request failed ({status})": "模型列表加载失败（{status}）",
+  "Missing {count} model files": "缺少 {count} 份模型文件",
+  "{label}: {value}": "{label}：{value}",
+  "LoRA {index} strength": "LoRA {index} 强度",
+  "LoRA {index}": "LoRA {index}",
+  "{label} · {count} encoders": "{label} · {count} 个编码器",
+  "Actual sampler connections · {node} #{id} · {field}. {description}": "实际采样器连线 · {node} #{id} · {field}。{description}",
+  "Restore {label} from the workflow": "恢复工作流中的{label}",
+  "Reading cache · {count} nodes": "读取缓存 · {count} 个节点",
+  "Settings panel {width} pixels wide": "右侧配置 {width} 像素",
+  "Invalid workbench response ({status}); check your proxy settings or restart ComfyUI": "工作台服务响应无效（{status}），请检查代理配置或重启 ComfyUI",
+  "{label} not found: {value}": "未找到{label}：{value}",
+  "The selected parameter node is missing from the execution graph #{id}": "所选参数节点未出现在执行图中 #{id}",
+
+  "Comparison image loading progress": "对比图片加载进度",
+  "Waiting for the generated image…": "等待生成结果，完成后自动找回…",
+  "Generation failed; check the ComfyUI task history": "生成失败，请查看 ComfyUI 任务历史",
+  "Loading comparison images": "正在加载对比图片",
+  "Retrying comparison images": "正在重试对比图片",
+  "Decoding comparison images…": "下载完成，正在解码图片…",
+  "Waiting to retry comparison images": "等待自动重试图片",
+  "Image loading failed; retrying automatically. You can also click reload.": "图片加载失败，正在自动重试；也可点击刷新。",
+  "Show group details": "展开分组详情",
+  "Hide group details": "收起分组详情",
+  "Reload comparison images": "重新加载对比图片",
+  "Image upload progress": "图片上传进度",
+  "Uploading image": "正在上传图片",
+  "Uploading image…": "正在上传图片…",
+  "Upload complete, loading image…": "上传完成，正在加载图片…",
+  "Uploading with authentication…": "正在通过身份验证上传…",
+  "Network error while uploading image": "上传图片时网络连接失败",
+  "Upload was cancelled": "上传已取消",
+  "Invalid upload response": "上传响应无效",
+  "Download the latest generated image": "下载最新生成的图片",
+  "Mask: paint the full image. Crop & Blend: select a region, inpaint it, then adopt and blend the result.": "普通蒙版：在整图涂抹。裁切融合：选定局部、重绘，再从图像对比采用结果并融合图层。",
+  "Crop resolution presets. Changing size releases the current selection; left-click the image to place a new region. Inpaint dimensions follow when selected.": "裁切分辨率预设。切换尺寸会取消当前选定，再左键点击图片定位新区域；选定后自动填写 Inpaint 宽高。",
+  "Swap width and height, release the current selection, then left-click to place the new region.": "交换宽高并取消当前选定，再左键点击图片选择新区域。",
+  "Edit layer blending: brush keeps the new result; Alt restores original pixels. Hold right-click to preview the original, release to return.": "编辑融合蒙版：画笔保留重绘结果，Alt 擦回原图。按住右键查看原图，松开恢复当前工作区。",
+  "Merge the visible result and blend mask into the source image. After merging, right-click releases the selected region.": "将当前可见结果和融合蒙版合并到原图。合并后，右键恢复为取消选区。",
+  "Undo the last stroke in the active repaint or blend mask (Ctrl+Z). Redo with Ctrl+Shift+Z.": "撤销当前重绘蒙版或融合蒙版的上一笔（Ctrl+Z）；Ctrl+Shift+Z 可重做。",
+  "Custom crop dimension in 64-pixel steps. Changing it releases the selected region.": "自定义裁切宽高，按 64 像素递增；修改尺寸会取消当前选定。",
+  "Paint inside the selected region. Right-click releases the selection; left-click places a new one. Alt: erase. Ctrl+wheel: brush size; Shift+wheel: hardness.": "在选区内涂蒙版。右键取消选定，再左键选择新区域；Alt 擦除；Ctrl+滚轮调笔径，Shift+滚轮调硬度。",
+  "Select a crop region before adopting a result": "请先选择裁切区域再采用结果",
+  "Finish editing before adopting a result": "请结束当前编辑后再采用结果",
+  "This result belongs to another crop; generate a result for the current region": "该结果来自其他选区，请为当前选区重新生成",
+  "Result dimensions must match the selected crop": "结果宽高必须与当前裁切区域一致",
+  "The crop changed while adopting the result": "采用过程中选区已改变，请重新生成结果",
+  "The result was not applied; the crop or layer state changed": "结果未应用，选区或图层状态已改变",
+  "The result preview is still loading": "结果预览仍在加载中",
+  "Select an inline mask to adopt into": "请选择要回填的内联蒙版",
+  "Painting tools and mask settings": "绘图工具与蒙版设置",
+  "Browser draft storage was full. The source was saved as an input file; the image, masks and layers are preserved.": "浏览器草稿空间不足，已将原图保存为图片文件；图像、蒙版和图层均已保留。",
+  "Editing mode": "编辑模式",
+  "Crop & Blend": "裁切融合",
+  "Crop size": "裁切尺寸",
+  "Custom": "自定义",
+  "Swap crop dimensions": "交换宽高",
+  "Choose another region": "重新选区（自动合并当前图层）",
+  "Repaint mask": "重绘蒙版",
+  "Blend layers: brush keeps new pixels; Alt restores original": "图层融合：画笔保留新图，Alt 擦回原图",
+  "Merge the visible layers": "合并当前可见图层",
   "Group Bypass Presets": "组忽略预设 / Group Bypass Presets",
   "Switch saved group enable/bypass presets; no connections required.": "切换已保存的分组启用/忽略预设，无需接线。",
   "Group presets": "分组预设",
@@ -56,7 +235,7 @@ const ZH = {
   "Brush": "画笔",
   "Brush (B)": "画笔 (B)",
   "Eraser": "橡皮",
-  "Eraser (E); hold Alt or right-click for temporary erase": "橡皮 (E)，按住 Alt 或鼠标右键临时擦除",
+  "Eraser (E); hold Alt for temporary erase": "橡皮 (E)，按住 Alt 临时擦除",
   "Rectangle": "矩形",
   "Rectangle (R)": "矩形 (R)",
   "Lasso": "套索",
@@ -87,6 +266,8 @@ const ZH = {
   "Mask": "蒙版",
   "Fit": "适应",
   "Keep": "保持",
+  "Pin the current view": "固定当前视图",
+  "Download the source image": "下载原图",
   "Upload": "上传",
   "Fit the image to the canvas": "适应画布",
   "File is too large (max 64 MB)": "图片文件不能超过 64 MB",
@@ -98,8 +279,8 @@ const ZH = {
   // inline_inpaint.js
   "The inline mask node was removed": "内联蒙版节点已移除",
   "Adopted the result as the new source image; the mask and undo history were preserved": "已采用重绘结果作为新原图，蒙版与撤销记录已保留",
-  "The result size must match the source image to preserve the mask": "为保留蒙版，重绘结果尺寸必须与原图一致",
-  "Wait for the source image to finish loading before adopting a result": "请等待原图加载完成后再采用结果",
+  "The new image has a different size, so the mask was reset": "新图尺寸与原图不同，蒙版已重置",
+  "Wait for the current image to finish loading": "请等待当前图片加载完成",
   "Drag to resize canvas height": "上下拖动调整画布高度",
   "Resize canvas height": "调整画布高度",
   "Upload failed": "上传失败",
@@ -109,9 +290,9 @@ const ZH = {
   "The painting tools live in the other inpaint node connected to this image": "画笔工具位于连接此原图的另一个重绘节点",
 
   // inpaint_panel.js
-  "WebUI Inpaint": "WebUI 局部重绘",
+  "Adachi Inpaint Workbench": "安达重绘工作台",
   "Controlled WebUI inpaint node": "控制的局部重绘节点",
-  "Connect a WebUI inpaint node to edit its parameters here": "连接 WebUI 局部重绘节点后，在这里调整参数",
+  "Connect a WebUI inpaint node to edit its parameters here": "连接安达重绘工作台后，在这里调整参数",
   "Mask mode": "重绘区域",
   "Fill content": "重绘参考内容",
   "Sampling area": "重绘画布",
@@ -165,28 +346,33 @@ const ZH = {
 const DICTS = { zh: ZH };
 
 function readLocale() {
-  try {
-    const api = globalThis.comfyAPI;
-    const value = api?.app?.ui?.settings?.getSettingValue?.("Comfy.Locale");
-    if (value) return String(value);
-  } catch (_) { /* settings unavailable before the app boots */ }
-  try {
-    const value = globalThis.app?.ui?.settings?.getSettingValue?.("Comfy.Locale");
-    if (value) return String(value);
-  } catch (_) { /* ignore */ }
+  // Modern frontends expose an ES-module namespace at comfyAPI.app, while
+  // earlier builds exposed the app directly. Never infer the UI language from
+  // the browser: ComfyUI's selected locale is the user's explicit preference.
+  const api = globalThis.comfyAPI;
+  for (const app of [api?.app?.app, api?.app, globalThis.app]) {
+    try {
+      const value = app?.ui?.settings?.getSettingValue?.("Comfy.Locale");
+      if (value) return String(value);
+    } catch (_) { /* settings may be unavailable before the app boots */ }
+  }
   return "en";
 }
 
 export function localeDict() {
-  const base = readLocale().split("-")[0].toLowerCase();
-  return DICTS[base] || null;
+  const base = readLocale().split(/[-_]/)[0].toLowerCase();
+  return Object.hasOwn(DICTS, base) ? DICTS[base] : null;
 }
 
-// Translate a user-facing string. `context` disambiguates identical English strings
-// that need different translations (for example the Fill button and the Fill enum).
-export function T(text, context) {
+// English is the fallback for every locale. The optional string context retains
+// enum disambiguation; an object supplies named values without concatenating
+// translated fragments. Values remain text and are never interpreted as HTML.
+export function T(text, contextOrParams, params = {}) {
+  const context = typeof contextOrParams === "string" ? contextOrParams : null;
+  const values = contextOrParams && typeof contextOrParams === "object" ? contextOrParams : params;
   const dict = localeDict();
-  if (!dict) return text;
   const key = context ? `${context}:${text}` : text;
-  return dict[key] ?? text;
+  const translated = dict && Object.hasOwn(dict, key) ? dict[key] : text;
+  return translated.replace(/\{([A-Za-z][A-Za-z0-9_]*)\}/g, (placeholder, name) =>
+    Object.hasOwn(values, name) ? String(values[name]) : placeholder);
 }

web/inline_crop.js

@@ -0,0 +1,690 @@
+import { CROP_PRESETS, cropAt } from "./crop_geometry.js?v=0.6.2";
+import { T } from "./i18n.js";
+import { syncCropDimensions, syncCurrentDimensions } from "./parameter_bridge.js?v=0.8.2";
+import { installResultAdoption } from "./result_adoption.js?v=stability1";
+
+const fields = ["mask", "baseMask", "history", "future", "lastMask", "dirty"];
+const canvas = (w, h) =>
+  Object.assign(document.createElement("canvas"), { width: w, height: h });
+const load = async (url) => {
+  const image = new Image();
+  image.src = url;
+  await image.decode();
+  return image;
+};
+const make = (tag, cls) => {
+  const e = document.createElement(tag);
+  e.className = cls;
+  return e;
+};
+
+export class InlineCropController {
+  constructor(
+    node,
+    editor,
+    { find, hide, changed, imageUrl, beforeModeChange },
+  ) {
+    Object.assign(this, { node, editor, changed, imageUrl, beforeModeChange });
+    this.widgets = Object.fromEntries(
+      ["edit_mode", "crop_state", "blend_data", "layer_image"].map((name) => [
+        name,
+        find(name),
+      ]),
+    );
+    Object.values(this.widgets).forEach(hide);
+    this.contexts = {};
+    this.target = "mask";
+    this.layer = null;
+    this.revision = 0;
+    this.acceptVersion = 0;
+    this.cropWidth = 1024;
+    this.cropHeight = 1536;
+    this.patchCanvas = canvas(1, 1);
+    this.build();
+    installResultAdoption(node, this);
+    // Only user uploads go through editor.upload; internal layer merges and
+    // adopting an already-sized result call onUpload directly instead.
+    const upload = editor.upload.bind(editor);
+    editor.upload = async file => {
+      const previous = editor.image;
+      await upload(file);
+      if (!editor.destroyed && editor.ready && editor.image !== previous && this.mode === "mask") {
+        syncCurrentDimensions(node);
+      }
+    };
+    const originalFlush = editor.flush.bind(editor);
+    editor.flush = () => {
+      if (this.target !== "blend") return originalFlush();
+      if (!editor.ready || !this.region?.locked)
+        return this.widgets.blend_data.value;
+      if (!editor.dirty && !editor.stroke) return this.widgets.blend_data.value;
+      const r = this.region,
+        out = canvas(r.width, r.height),
+        ctx = out.getContext("2d");
+      ctx.fillStyle = "black";
+      ctx.fillRect(0, 0, r.width, r.height);
+      ctx.drawImage(
+        editor.mask,
+        r.x,
+        r.y,
+        r.width,
+        r.height,
+        0,
+        0,
+        r.width,
+        r.height,
+      );
+      this.widgets.blend_data.value = out.toDataURL("image/png");
+      out.width = out.height = 0;
+      editor.dirty = false;
+      return this.widgets.blend_data.value;
+    };
+    editor.options.sceneRevision = () =>
+      `${this.mode}:${this.target}:${this.revision}`;
+    editor.options.drawSource = (ctx) => this.drawSource(ctx, {preview:this.previewOriginal});
+    editor.options.displayView = () => this.previewOriginal ? "image" : editor.view;
+    editor.options.showCursor = () => !this.previewOriginal;
+    editor.options.showMask = () =>
+      this.mode !== "crop" || (this.target === "mask" && this.region?.locked);
+    editor.options.drawOverlay = (ctx, d) => this.drawOverlay(ctx, d);
+    editor.options.paintBounds = () =>
+      this.mode === "crop" && this.region?.locked ? this.region : undefined;
+    const down = editor.pointerDown.bind(editor),
+      move = editor.pointerMove.bind(editor);
+    editor.pointerDown = (e) => {
+      if (!editor.ready) return;
+      if (this.previewOriginal) { e.preventDefault(); e.stopPropagation(); return; }
+      if (this.busy) {
+        e.preventDefault();
+        e.stopPropagation();
+        return;
+      }
+      if (this.mode === "crop" && e.button === 2) {
+        e.preventDefault();
+        e.stopPropagation();
+        editor.root.focus({ preventScroll: true });
+        // Hold-right preview only makes sense while adjusting the blend layer.
+        // After adoption the layer is already baked, so right-click reselects
+        // (reselect flattens any remaining layer first).
+        if (this.target === "blend" && this.layer && this.region.locked && !editor.stroke && editor.pointerId == null) {
+          this.previewOriginal = true;
+          this.originalPreviewPointer = e.pointerId;
+          editor.canvas.setPointerCapture(e.pointerId);
+          editor.cursor = null;
+          this.revision++;
+          editor.schedule();
+        } else if (this.region.locked) this.run(() => this.reselect());
+        return;
+      }
+      if (
+        this.mode === "crop" &&
+        !this.region?.locked &&
+        e.button === 0 &&
+        !editor.space
+      ) {
+        e.preventDefault();
+        e.stopPropagation();
+        editor.root.focus({ preventScroll: true });
+        const p = editor.point(e),
+          r = cropAt(
+            editor.image.width,
+            editor.image.height,
+            this.cropWidth,
+            this.cropHeight,
+            p.x,
+            p.y,
+          );
+        if (r) this.lock(r);
+        return;
+      }
+      down(e);
+    };
+    editor.pointerMove = (e) => {
+      if (this.previewOriginal) { editor.cursor = null; return; }
+      if (
+        this.mode === "crop" &&
+        editor.ready &&
+        !this.region?.locked &&
+        !editor.drag
+      ) {
+        const p = editor.point(e);
+        this.hover = cropAt(
+          editor.image.width,
+          editor.image.height,
+          this.cropWidth,
+          this.cropHeight,
+          p.x,
+          p.y,
+        );
+      }
+      move(e);
+    };
+    this.finishPreview = event => {
+      if (event.pointerId !== this.originalPreviewPointer) return;
+      if (event.type === "pointerup" && (event.buttons & 2)) return;
+      event.preventDefault(); event.stopPropagation();
+      this.stopOriginalPreview();
+    };
+    for (const name of ["pointerup", "pointercancel"]) {
+      window.addEventListener(name, this.finishPreview, true);
+    }
+    editor.canvas.addEventListener("lostpointercapture", this.finishPreview, true);
+    this.previewBlur = () => this.stopOriginalPreview();
+    window.addEventListener("blur", this.previewBlur);
+    editor.canvas.addEventListener("pointerleave", () => {
+      this.hover = null;
+      editor.schedule();
+    });
+    editor.root.addEventListener("keydown", (e) => {
+      if (e.key === "Escape" && this.mode === "crop" && !editor.stroke) {
+        e.preventDefault();
+        e.stopPropagation();
+        this.reselect();
+      }
+    });
+  }
+
+  get mode() {
+    return this.widgets.edit_mode.value || "mask";
+  }
+  stopOriginalPreview() {
+    if (!this.previewOriginal) return;
+    const pointer = this.originalPreviewPointer;
+    this.previewOriginal = false;
+    this.originalPreviewPointer = null;
+    if (pointer != null && this.editor.canvas.hasPointerCapture(pointer)) {
+      this.editor.canvas.releasePointerCapture(pointer);
+    }
+    this.revision++;
+    this.editor.schedule();
+  }
+  get region() {
+    try {
+      return JSON.parse(this.widgets.crop_state.value || "{}");
+    } catch {
+      return {};
+    }
+  }
+  state(value) {
+    this.widgets.crop_state.value = JSON.stringify(value);
+    this.revision++;
+  }
+  snapshot() {
+    return Object.fromEntries(fields.map((k) => [k, this.editor[k]]));
+  }
+  activate(target) {
+    if (this.target === target) return;
+    this.editor.flush();
+    this.contexts[this.target] = this.snapshot();
+    const next = this.contexts[target];
+    if (!next) return;
+    Object.assign(this.editor, next);
+    this.target = target;
+    this.editor.maskRevision++;
+    this.editor.updateButtons();
+    this.editor.schedule();
+    this.sync();
+  }
+  build() {
+    const e = this.editor;
+    this.modeSelect = make("select", "dd-mode-select");
+    this.modeSelect.setAttribute("aria-label", T("Editing mode"));
+    this.modeSelect.title = T("Mask: paint the full image. Crop & Blend: select a region, inpaint it, then adopt and blend the result.");
+    for (const [value, label] of [
+      ["mask", "Mask"],
+      ["crop", "Crop & Blend"],
+    ]) {
+      const option = document.createElement("option");
+      option.value = value;
+      option.textContent = T(label);
+      this.modeSelect.append(option);
+    }
+    this.modeSelect.onchange = () => this.setMode(this.modeSelect.value);
+    e.root.append(this.modeSelect);
+    this.bar = make("div", "dd-crop-bar");
+    this.preset = make("select", "");
+    this.preset.setAttribute("aria-label", T("Crop size"));
+    this.preset.title = T("Crop resolution presets. Changing size releases the current selection; left-click the image to place a new region. Inpaint dimensions follow when selected.");
+    for (const [w, h] of CROP_PRESETS) {
+      const o = document.createElement("option");
+      o.value = `${w}x${h}`;
+      o.textContent = `${w} × ${h}`;
+      this.preset.append(o);
+    }
+    const custom = document.createElement("option");
+    custom.value = "custom";
+    custom.textContent = T("Custom");
+    this.preset.append(custom);
+    this.preset.onchange = () => this.run(() => {
+      const custom = this.preset.value === "custom";
+      const [width, height] = custom
+        ? [this.cropWidth, this.cropHeight]
+        : this.preset.value.split("x").map(Number);
+      return this.changeSize(width, height, {custom});
+    });
+    this.bar.append(this.preset);
+    const button = (text, title, action) =>
+      e.button(this.bar, text, T(title), () => this.run(action));
+    this.swapButton = button("⇄", "Swap width and height, release the current selection, then left-click to place the new region.", () =>
+      this.changeSize(this.cropHeight, this.cropWidth),
+    );
+    // Shown in both modes: one click clears the mask of the active target.
+    // It sits right after the swap control in Crop & Blend, and is the only
+    // visible button in Mask mode.
+    this.clearButton = button("🗑", "Clear the mask", () => e.action({ type: "clear" }));
+    this.swapButton.after(this.clearButton);
+    this.maskButton = button("◌", "Paint inside the selected region. Right-click releases the selection; left-click places a new one. Alt: erase. Ctrl+wheel: brush size; Shift+wheel: hardness.", () => {
+      this.activate("mask");
+      this.changed();
+    });
+    this.blendButton = button(
+      "◐",
+      "Edit layer blending: brush keeps the new result; Alt restores original pixels. Hold right-click to preview the original, release to return.",
+      () =>
+        this.run(async () => {
+          // Blending is opt-in after adoption: build the blend mask on first
+          // use, then switch the editor into the blend layer.
+          if (!this.contexts.blend) await this.makeBlend(this.widgets.blend_data.value);
+          this.activate("blend");
+          this.changed();
+        }),
+    );
+    // Move the original live button, retaining its history and disabled state.
+    this.undoButton = e.undoButton;
+    this.undoButton.textContent = "↶";
+    this.undoButton.title = T("Undo the last stroke in the active repaint or blend mask (Ctrl+Z). Redo with Ctrl+Shift+Z.");
+    this.custom = make("div", "dd-crop-custom");
+    this.custom.hidden = true;
+    for (const key of ["cropWidth", "cropHeight"]) {
+      const input = make("input", "");
+      input.type = "number";
+      input.min = 64;
+      input.max = 16384;
+      input.step = 64;
+      input.value = this[key];
+      input.setAttribute(
+        "aria-label",
+        T(key === "cropWidth" ? "Width" : "Height"),
+      );
+      input.title = T("Custom crop dimension in 64-pixel steps. Changing it releases the selected region.");
+      input.onchange = () => this.run(() => {
+        const value = Math.max(64, Math.min(16384, Math.round(Number(input.value) / 64) * 64));
+        return this.changeSize(key === "cropWidth" ? value : this.cropWidth,
+          key === "cropHeight" ? value : this.cropHeight, {custom:true});
+      });
+      this.custom.append(input);
+    }
+    this.bar.append(this.custom);
+    e.root.append(this.bar);
+    this.sync();
+  }
+  async run(action) {
+    if (
+      this.busy ||
+      this.editor.stroke
+    )
+      return;
+    try {
+      await action();
+    } catch (error) {
+      this.editor.status(error.message, true);
+    }
+  }
+  sync() {
+    this.editor.root.classList.toggle("dd-crop-mode", this.mode === "crop");
+    this.modeSelect.value = this.mode;
+    // No hover text on the painting canvas; the bottom controls carry the hints.
+    this.editor.canvas.removeAttribute("title");
+    for (const control of [this.preset, this.swapButton, this.maskButton, this.blendButton]) {
+      control.hidden = this.mode !== "crop";
+    }
+    this.custom.hidden = this.mode !== "crop" || this.preset.value !== "custom";
+    this.undoButton.disabled = !!this.busy || !this.editor.history.length;
+    this.clearButton.disabled = !!this.busy || !this.editor.ready;
+    this.blendButton.disabled = !this.layer;
+    this.maskButton.classList.toggle("dd-active", this.target === "mask");
+    this.blendButton.classList.toggle("dd-active", this.target === "blend");
+    this.modeSelect.disabled = !!this.busy;
+    this.preset.disabled = !!this.busy;
+    this.swapButton.disabled = !!this.busy;
+    for (const input of this.custom.children)
+      input.disabled = !!this.busy;
+  }
+  setMode(mode) {
+    if (
+      this.editor.stroke ||
+      this.busy
+    ) {
+      this.sync();
+      return;
+    }
+    if (mode === this.mode) {
+      this.sync();
+      return;
+    }
+    const apply = () => {
+      this.stopOriginalPreview();
+      this.activate("mask");
+      this.widgets.edit_mode.value = mode;
+      this.state({ ...this.region, job: crypto.randomUUID() });
+      this.editor.fit();
+      this.sync();
+      if (mode === "mask") syncCurrentDimensions(this.node);
+      this.changed();
+    };
+    const recovery = this.beforeModeChange?.();
+    if (!recovery?.then) {
+      if (recovery !== false) apply();
+      return;
+    }
+    this.busy = true;
+    this.sync();
+    return recovery
+      .then((ok) => {
+        this.busy = false;
+        if (ok !== false && !this.editor.destroyed) apply();
+        else this.sync();
+      })
+      .catch((error) => {
+        this.busy = false;
+        this.sync();
+        this.editor.status(error.message, true);
+      });
+  }
+  syncPreset() {
+    const value = `${this.cropWidth}x${this.cropHeight}`;
+    this.preset.value = [...this.preset.options].some((o) => o.value === value)
+      ? value
+      : "custom";
+    this.custom.hidden = this.preset.value !== "custom";
+    [...this.custom.children].forEach(
+      (e, i) => (e.value = i ? this.cropHeight : this.cropWidth),
+    );
+  }
+  sizeChanged() {
+    this.hover = null;
+    this.editor.schedule();
+    this.syncPreset();
+  }
+  async changeSize(width, height, {custom = false} = {}) {
+    if (this.busy || this.editor.stroke) { this.syncPreset(); return; }
+    try {
+      // Use the same merge-before-reselect behavior as the reselect button.
+      if (this.region.locked) await this.reselect();
+      this.cropWidth = width;
+      this.cropHeight = height;
+      this.sizeChanged();
+      if (custom) { this.preset.value = "custom"; this.custom.hidden = false; }
+      this.sync();
+      this.changed();
+    } catch (error) {
+      this.syncPreset();
+      throw error;
+    }
+  }
+  reset() {
+    this.stopOriginalPreview();
+    this.layer = null;
+    this.widgets.layer_image.value = "";
+    this.widgets.blend_data.value = "";
+    this.contexts = {};
+    this.target = "mask";
+    this.hover = null;
+    this.state({ locked: false, job: crypto.randomUUID() });
+    const { width, height } = this.editor.image;
+    for (const o of this.preset.options)
+      if (o.value !== "custom") {
+        const [w, h] = o.value.split("x").map(Number);
+        o.disabled = w > width || h > height;
+      }
+    if (this.cropWidth > width || this.cropHeight > height) {
+      this.cropWidth = Math.min(1024, Math.floor(width / 64) * 64);
+      this.cropHeight = Math.min(1536, Math.floor(height / 64) * 64);
+    }
+    this.syncPreset();
+    this.sync();
+    this.editor.schedule();
+  }
+  lock(region) {
+    this.activate("mask");
+    this.contexts.blend = null;
+    this.layer = null;
+    this.widgets.layer_image.value = "";
+    this.widgets.blend_data.value = "";
+    this.state({ ...region, locked: true, job: crypto.randomUUID() });
+    const e = this.editor;
+    // A new crop only limits the painting area; the repaint mask starts empty.
+    for (const c of [e.mask, e.baseMask])
+      c.getContext("2d").clearRect(0, 0, c.width, c.height);
+    e.history = [];
+    e.future = [];
+    e.changed();
+    syncCropDimensions(this.node, region.width, region.height);
+    this.sync();
+    this.changed();
+  }
+  makeBlend(data = "") {
+    const r = this.region,
+      e = this.editor,
+      mask = canvas(e.image.width, e.image.height),
+      baseMask = canvas(e.image.width, e.image.height);
+    const ctx = mask.getContext("2d");
+    ctx.fillStyle = "white";
+    ctx.fillRect(r.x, r.y, r.width, r.height);
+    baseMask.getContext("2d").drawImage(mask, 0, 0);
+    this.contexts.blend = {
+      mask,
+      baseMask,
+      history: [],
+      future: [],
+      lastMask: "",
+      dirty: false,
+    };
+    if (!data) return Promise.resolve();
+    return load(data).then((image) => {
+      const out = canvas(r.width, r.height),
+        c = out.getContext("2d");
+      c.drawImage(image, 0, 0);
+      const pixels = c.getImageData(0, 0, r.width, r.height);
+      for (let i = 0; i < pixels.data.length; i += 4) {
+        pixels.data[i + 3] = pixels.data[i];
+        pixels.data[i] = pixels.data[i + 1] = pixels.data[i + 2] = 255;
+      }
+      ctx.clearRect(0, 0, mask.width, mask.height);
+      ctx.putImageData(pixels, r.x, r.y);
+      const base = baseMask.getContext("2d");
+      base.clearRect(0, 0, baseMask.width, baseMask.height);
+      base.drawImage(mask, 0, 0);
+      out.width = out.height = 0;
+    });
+  }
+  async restore() {
+    this.stopOriginalPreview();
+    this.target = "mask";
+    this.contexts = {};
+    this.layer = null;
+    const r = this.region;
+    if (r.locked) {
+      this.cropWidth = r.width;
+      this.cropHeight = r.height;
+      this.syncPreset();
+      if (this.widgets.layer_image.value) {
+        this.layer = await load(this.imageUrl(this.widgets.layer_image.value));
+        await this.makeBlend(this.widgets.blend_data.value);
+        // Restore the state the user left: a non-empty blend mask means they were
+        // still adjusting the blend layer, otherwise the adopted layer is already
+        // baked and we stay on the repaint mask.
+        if (this.widgets.blend_data.value && this.mode === "crop") this.activate("blend");
+      }
+    }
+    this.revision++;
+    this.sync();
+    this.editor.schedule();
+  }
+  async accept(record, {allowBusy = false, image = null} = {}) {
+    if (
+      this.editor.destroyed ||
+      this.mode !== "crop" ||
+      !this.region.locked ||
+      record.job !== this.region.job ||
+      (this.busy && !allowBusy)
+    )
+      return;
+    if (this.editor.stroke) {
+      this.pendingResult = record;
+      clearTimeout(this.resultTimer);
+      this.resultTimer = setTimeout(() => {
+        const next = this.pendingResult;
+        this.pendingResult = null;
+        this.accept(next, {allowBusy}).catch((error) =>
+          this.editor.status(error.message, true),
+        );
+      }, 50);
+      return;
+    }
+    const name = record.image.subfolder
+      ? `${record.image.subfolder}/${record.image.filename}`
+      : record.image.filename;
+    if (name === this.widgets.layer_image.value) return;
+    const version = ++this.acceptVersion;
+    // The adoption path already decoded this image once; reuse it instead of
+    // downloading and decoding the full-size result a second time.
+    const decoded = image || await load(this.imageUrl(name));
+    if (
+      this.editor.destroyed ||
+      version !== this.acceptVersion ||
+      record.job !== this.region.job ||
+      this.mode !== "crop"
+    )
+      return;
+    this.layer = decoded;
+    this.widgets.layer_image.value = name;
+    // Adopt then return: with an empty blend mask the backend hard-composites
+    // the whole crop from layer_image, so the result is immediately baked into
+    // the source and the repaint mask is reused on the next generation. Blending
+    // is no longer entered automatically; the ◐ button opens it on demand.
+    this.activate("mask");
+    this.revision++;
+    this.sync();
+    this.editor.schedule();
+    this.changed();
+  }
+  drawSource(ctx, {preview = false} = {}) {
+    if (this.mode !== "crop") return false;
+    const e = this.editor,
+      r = this.region;
+    ctx.drawImage(e.image, 0, 0);
+    if (!preview && this.layer && r.locked) {
+      const key = `${this.revision}:${e.maskRevision}`;
+      if (this.patchKey !== key) {
+        this.patchCanvas.width = r.width;
+        this.patchCanvas.height = r.height;
+        const c = this.patchCanvas.getContext("2d");
+        c.drawImage(this.layer, 0, 0);
+        const mask =
+          this.target === "blend" ? e.mask : this.contexts.blend?.mask;
+        if (mask) {
+          c.globalCompositeOperation = "destination-in";
+          c.drawImage(
+            mask,
+            r.x,
+            r.y,
+            r.width,
+            r.height,
+            0,
+            0,
+            r.width,
+            r.height,
+          );
+          c.globalCompositeOperation = "source-over";
+        }
+        this.patchKey = key;
+      }
+      ctx.drawImage(this.patchCanvas, r.x, r.y);
+    }
+    return true;
+  }
+  drawOverlay(ctx, d) {
+    if (this.mode !== "crop") return;
+    const r = this.region.locked ? this.region : this.hover;
+    if (!r) return;
+    ctx.save();
+    if (this.region.locked) {
+      ctx.fillStyle = "#0008";
+      ctx.beginPath();
+      ctx.rect(0, 0, this.editor.image.width, this.editor.image.height);
+      ctx.rect(r.x, r.y, r.width, r.height);
+      ctx.fill("evenodd");
+    }
+    ctx.strokeStyle = "#000000";
+    ctx.lineWidth = 1 / d.scale;
+    ctx.setLineDash([6 / d.scale, 4 / d.scale]);
+    ctx.strokeRect(r.x, r.y, r.width, r.height);
+    ctx.restore();
+  }
+  async flatten() {
+    if (!this.layer) return;
+    this.stopOriginalPreview();
+    this.editor.flush();
+    this.busy = true;
+    this.sync();
+    try {
+      const e = this.editor,
+        c = canvas(e.image.width, e.image.height);
+      this.drawSource(c.getContext("2d"));
+      const blob = await new Promise((resolve) =>
+        c.toBlob(resolve, "image/png"),
+      );
+      c.width = c.height = 0;
+      this.activate("mask");
+      await e.options.onUpload(
+        new File([blob], `merged-${crypto.randomUUID()}.png`, {
+          type: "image/png",
+        }),
+        { preserveMask: true, preserveCrop: true },
+      );
+      this.layer = null;
+      this.widgets.layer_image.value = "";
+      this.widgets.blend_data.value = "";
+      for(const key of ['mask','baseMask']){const raster=this.contexts.blend?.[key];if(raster && raster!==e.mask && raster!==e.baseMask)raster.width=raster.height=0;}
+      this.contexts.blend = null;
+      this.state({ ...this.region, job: crypto.randomUUID() });
+    } finally {
+      this.busy = false;
+      this.sync();
+      this.editor.schedule();
+      this.changed();
+    }
+  }
+  async reselect() {
+    if (
+      this.busy ||
+      this.editor.stroke
+    )
+      return;
+    await this.flatten();
+    this.activate("mask");
+    this.state({ locked: false, job: crypto.randomUUID() });
+    this.hover = null;
+    this.sync();
+    this.editor.schedule();
+    this.changed();
+  }
+  destroy() {
+    this.adoptController?.abort();
+    this.stopOriginalPreview();
+    window.removeEventListener("blur", this.previewBlur);
+    for (const name of ["pointerup", "pointercancel"]) {
+      window.removeEventListener(name, this.finishPreview, true);
+    }
+    this.editor.canvas.removeEventListener("lostpointercapture", this.finishPreview, true);
+    clearTimeout(this.resultTimer);
+    this.pendingResult = null;
+    this.acceptVersion++;
+    for (const context of Object.values(this.contexts))
+      for (const key of ["mask", "baseMask"])
+        if (context?.[key] !== this.editor[key] && context?.[key])
+          context[key].width = context[key].height = 0;
+    this.patchCanvas.width = this.patchCanvas.height = 0;
+  }
+}

web/inline_inpaint.js

@@ -1,15 +1,19 @@
 import { app } from "../../scripts/app.js";
 import { api } from "../../scripts/api.js";
-import { InlineMaskEditor } from "./editor.js?v=0.5.10";
-import { installInpaintWorkbench } from "./inpaint_node.js?v=0.5.5";
+import { InlineMaskEditor } from "./editor.js?v=stability1";
+import { installInpaintWorkbench } from "./inpaint_node.js?v=2.0.0";
 import { bindWidgetToNodeWidth } from "./widget_layout.js?v=0.5.5";
 import { routeWheelOutsideImage } from "./wheel_routing.js";
+import { InlineCropController } from "./inline_crop.js?v=stability1";
+import { canStoreDraft, prepareInlineDraft, syncInlineWidgets } from "./draft_guard.js?v=0.6.3";
 import { T } from "./i18n.js";
+import { removeLegacyFeedback } from "./legacy_feedback.js?v=0.6.4";
+import { uploadImageWithProgress } from "./upload_transport.js?v=stability1";
 
 const styleId = "adachi-inpaint-style";
 const stylesheet = document.getElementById(styleId) || document.createElement("link");
 stylesheet.id = styleId; stylesheet.rel = "stylesheet";
-stylesheet.href = new URL("./editor.css?v=0.5.5", import.meta.url).href;
+stylesheet.href = new URL("./editor.css?v=workspace-network2", import.meta.url).href;
 if (!stylesheet.isConnected) document.head.append(stylesheet);
 
 function hide(widget) {
@@ -36,9 +40,44 @@ function imageUrl(value) {
     return api.apiURL(`/view?${params}`);
 }
 
+function draftOptions() {
+    let workspace = "personal";
+    try { const current = JSON.parse(sessionStorage.getItem("Comfy.Workspace.Current") || "null"); if (current?.type !== "personal" && current?.id) workspace = current.id; } catch { /* Personal workspace. */ }
+    return { workspace, path: app.extensionManager?.workflow?.activeWorkflow?.path || "" };
+}
+
+async function uploadEmbeddedSource(embedded) {
+    const blob = await (await fetch(embedded)).blob();
+    if (blob.size > 64 * 1024 * 1024) throw Error(T("File is too large (max 64 MB)"));
+    const body = new FormData();
+    body.append("image", blob, `source-${crypto.randomUUID()}.png`);
+    body.append("type", "input"); body.append("subfolder", "adachi-inpaint");
+    const response = await api.fetchApi("/upload/image", {method:"POST", body});
+    if (!response.ok) throw Error(`${T("Upload failed")} (${response.status})`);
+    const uploaded = await response.json();
+    if (typeof uploaded.name !== "string" || !uploaded.name) throw Error(T("Upload failed"));
+    return uploaded.subfolder ? `${uploaded.subfolder}/${uploaded.name}` : uploaded.name;
+}
+
 app.registerExtension({
     name: "Adachi.InlineInpaint",
+    async beforeConfigureGraph(graphData) {
+        removeLegacyFeedback(graphData);
+        if (app.ui.settings.getSettingValue("Comfy.Workflow.Persist") === false) return;
+        try {
+            // The active workflow can still be the previous tab at this point;
+            // do not claim its replacement credit for an incoming workflow.
+            const options = { ...draftOptions(), path: "" };
+            const count = await prepareInlineDraft(graphData, {
+                fits: graph => canStoreDraft(graph, options), upload: uploadEmbeddedSource,
+            });
+            if (count) console.info(`[Adachi Inpaint] Saved ${count} oversized embedded source(s) as exact input files before workflow draft initialization. Masks and layers were preserved.`);
+        } catch (error) {
+            console.warn("[Adachi Inpaint] Draft preflight failed; embedded source data was retained.", error);
+        }
+    },
     async beforeRegisterNodeDef(nodeType, nodeData) {
+        if (nodeData.name === "AdachiLayerCommit") { nodeType.skip_list = true; return; }
         if (["AdachiWebUIInpaint", "AdachiInpaintPrepare"].includes(nodeData.name)) { installInpaintWorkbench(nodeType); return; }
         if (nodeData.name !== "AdachiInlineMask") return;
         const originalCreated = nodeType.prototype.onNodeCreated;
@@ -56,6 +95,9 @@ app.registerExtension({
             const find = name => node.widgets?.find(widget => widget.name === name);
             const imageWidget = find("image"), maskWidget = find("mask_data"), sourceWidget = find("source_data");
             for (const widget of [imageWidget, maskWidget, sourceWidget]) hide(widget);
+            let crop;
+            let draftRecovery = null;
+            let draftNoticeTimer = 0;
             let restoring = false;
             let restoreFrame = 0;
             let uploadVersion = 0;
@@ -63,7 +105,7 @@ app.registerExtension({
             const fitNodeToImage = () => {
                 cancelAnimationFrame(fitFrame);
                 fitFrame = requestAnimationFrame(() => {
-                    if (!editor.ready || editor.destroyed || !node.graph) return;
+                    if (!editor.ready || editor.destroyed || !node.graph || editor.root.closest(".adachi-workspace")) return;
                     const width = editor.canvas.clientWidth, height = editor.canvas.clientHeight;
                     if (!width || !height) return;
                     const overhead = node.size[1] - height;
@@ -79,7 +121,7 @@ app.registerExtension({
                 if (!embedded && !filename) return;
                 restoring = true;
                 editor.embed.checked = !!embedded;
-                try { await editor.setImage(embedded || imageUrl(filename), maskWidget?.value || "", embedded ? T("Embedded image") : filename); }
+                try { crop?.activate("mask"); await editor.setImage(embedded || imageUrl(filename), maskWidget?.value || "", embedded ? T("Embedded image") : filename); await crop?.restore(); }
                 catch (error) { console.warn("[Adachi Inpaint]", error); }
                 finally { restoring = false; }
             };
@@ -90,38 +132,77 @@ app.registerExtension({
                 onKeepFit(enabled) { node.properties.adachi_keep_fit = enabled; markChanged(); },
                 onImage() { if (!restoring || !node.properties.adachi_compact_ui) fitNodeToImage(); },
                 onGraphWheel: event => app.canvas?.processMouseWheel(event),
-                async onUpload(file, { preserveMask = false } = {}) {
-                    if (preserveMask && editor.image && !editor.ready) throw new Error(T("Wait for the source image to finish loading before adopting a result"));
+                async onUpload(file, { preserveMask = false, preserveCrop = false, onProgress = null, signal } = {}) {
+                    if (preserveMask && editor.image && !editor.ready) throw new Error(T("Wait for the current image to finish loading"));
                     const version = ++uploadVersion;
                     const body = new FormData();
                     body.append("image", file, file.name);
                     body.append("type", "input");
                     body.append("subfolder", "adachi-inpaint");
-                    const response = await api.fetchApi("/upload/image", { method: "POST", body });
-                    if (!response.ok) throw new Error(`${T("Upload failed")} (${response.status})`);
-                    const uploaded = await response.json();
+                    const uploaded = await uploadImageWithProgress(api, body, onProgress, {signal});
+                    if (typeof uploaded?.name !== "string" || !uploaded.name) throw new Error(T("Upload failed"));
                     if (version !== uploadVersion || editor.destroyed) return;
+                    signal?.throwIfAborted();
                     const name = uploaded.subfolder ? `${uploaded.subfolder}/${uploaded.name}` : uploaded.name;
                     restoring = false;
-                    if (preserveMask && editor.ready) await editor.replaceSource(imageUrl(name), file.name);
-                    else await editor.setImage(imageUrl(name), "", file.name);
+                    crop?.activate("mask");
+                    let kept = false;
+                    if (preserveMask && editor.ready) {
+                        kept = await editor.replaceSource(imageUrl(name), file.name);
+                        if (version !== uploadVersion || editor.destroyed) return;
+                        // A different-sized replacement cannot reuse the old mask.
+                        // Actually load it as a fresh image before clearing state.
+                        if (!kept) await editor.setImage(imageUrl(name), "", file.name);
+                    } else await editor.setImage(imageUrl(name), "", file.name);
                     if (version !== uploadVersion || editor.destroyed) return;
                     imageWidget.value = name;
-                    maskWidget.value = preserveMask ? editor.flush() : "";
+                    maskWidget.value = kept ? editor.flush() : "";
                     sourceWidget.value = editor.embed.checked ? editor.sourceData() : "";
+                    if (!preserveCrop) crop?.reset();
+                    await beforeModeChange();
+                    if (preserveMask && !kept) editor.notice(T("The new image has a different size, so the mask was reset"));
                     markChanged();
                 },
                 onEmbed(enabled) {
                     if (sourceWidget) sourceWidget.value = enabled ? editor.sourceData() : "";
+                    const recovery = enabled ? beforeModeChange() : null;
+                    if (recovery?.then) return recovery.then(() => markChanged()).catch(error => editor.status(error.message, true));
                     markChanged();
                 },
             });
-            editor.root.dataset.adachiVersion = "0.5.10";
+            editor.root.dataset.adachiVersion = "0.6.7";
+            const beforeModeChange = () => {
+                if (draftRecovery) return draftRecovery;
+                const embedded = sourceWidget?.value;
+                if (!editor.ready || !embedded?.startsWith("data:") || embedded.length < 256 * 1024) return;
+                if (app.ui.settings.getSettingValue("Comfy.Workflow.Persist") === false) return;
+                const graph = app.rootGraph || node.graph;
+                if (canStoreDraft(graph.serialize(), draftOptions()) !== false) return;
+                // Persist a fresh, exact copy before dropping any embedded bytes.
+                // This also repairs portable workflows whose original input path is absent.
+                draftRecovery = (async () => {
+                    const version = editor.loadVersion;
+                    const name = await uploadEmbeddedSource(embedded);
+                    if (editor.destroyed || editor.loadVersion !== version || sourceWidget.value !== embedded) return false;
+                    imageWidget.value = name;
+                    sourceWidget.value = ""; editor.embed.checked = false;
+                    editor.message.textContent = T("Browser draft storage was full. The source was saved as an input file; the image, masks and layers are preserved.");
+                    editor.message.classList.remove("dd-error"); editor.message.hidden = false;
+                    const notice = editor.message.textContent;
+                    clearTimeout(draftNoticeTimer);
+                    draftNoticeTimer = setTimeout(() => { if (!editor.destroyed && editor.message.textContent === notice) editor.message.hidden = true; }, 8000);
+                    return true;
+                })().finally(() => { draftRecovery = null; });
+                return draftRecovery;
+            };
+            crop = new InlineCropController(node, editor, {find, hide, changed:markChanged, imageUrl, beforeModeChange});
+            node.adachiCrop = crop;
             node.adachiEditor = editor;
             const stopWheelRouting = routeWheelOutsideImage(editor.root, editor.canvas, () => editor.ready);
-            node.adachiUseResult = async file => {
+            node.adachiUseResult = async (file, origin, options={}) => {
                 if (editor.destroyed) throw new Error(T("The inline mask node was removed"));
-                await editor.options.onUpload(file, { preserveMask: true });
+                if (crop.mode === "crop") return node.adachiAdoptCrop(file, origin, options);
+                await editor.options.onUpload(file, { preserveMask: true, signal:options.signal });
                 editor.view = "overlay"; editor.updateButtons(); editor.schedule();
                 editor.status(T("Adopted the result as the new source image; the mask and undo history were preserved"));
             };
@@ -183,10 +264,10 @@ app.registerExtension({
             };
             const previousSerializeNode = node.onSerialize;
             node.onSerialize = function (data) {
+                const result = previousSerializeNode?.apply(this, arguments);
                 editor.flush();
-                const index = node.widgets.indexOf(maskWidget);
-                if (data?.widgets_values && index >= 0) data.widgets_values[index] = maskWidget.value;
-                return previousSerializeNode?.apply(this, arguments);
+                if (data) syncInlineWidgets(data, node.widgets);
+                return result;
             };
             const previousQueued = maskWidget.beforeQueued;
             maskWidget.beforeQueued = function () { editor.flush(); return previousQueued?.apply(this, arguments); };
@@ -200,7 +281,7 @@ app.registerExtension({
             const previousImageCallback = imageWidget.callback;
             imageWidget.callback = function () { previousImageCallback?.apply(this, arguments); restore(); };
             const previousRemoved = node.onRemoved;
-            node.onRemoved = function () { cancelAnimationFrame(restoreFrame); cancelAnimationFrame(fitFrame); stopWheelRouting(); editor.destroy(); return previousRemoved?.apply(this, arguments); };
+            node.onRemoved = function () { node.adachiWorkspaceHistory?.clear();node.adachiWorkspaceHistory=null;clearTimeout(draftNoticeTimer); cancelAnimationFrame(restoreFrame); cancelAnimationFrame(fitFrame); stopWheelRouting(); crop.destroy(); editor.destroy(); return previousRemoved?.apply(this, arguments); };
             node.setSize([Math.max(380, node.size[0]), Math.max(480, node.size[1])]);
             restoreFrame = requestAnimationFrame(restore);
             return result;

web/inpaint_node.js

@@ -1,6 +1,6 @@
 import { app } from "../../scripts/app.js";
-import { InpaintParameterPanel } from "./inpaint_panel.js";
-import { connectedEditor, connectedSamplers, relocateWidgets } from "./parameter_bridge.js";
+import { InpaintParameterPanel } from "./inpaint_panel.js?v=adachi-name";
+import { connectedEditor, connectedSamplers, relocateWidgets, syncCurrentDimensions } from "./parameter_bridge.js?v=0.8.2";
 import { routeWheelOutsideImage } from "./wheel_routing.js";
 import { T } from "./i18n.js";
 import { bindWidgetToNodeWidth } from "./widget_layout.js?v=0.5.5";
@@ -10,10 +10,24 @@ export function installInpaintWorkbench(nodeType) {
     nodeType.prototype.onNodeCreated = function () {
         const result = created?.apply(this, arguments), node = this;
         const root = document.createElement("div"); root.className = "dd-inpaint dd-workbench";
+        if (node.comfyClass === "AdachiWebUIInpaint" || node.type === "AdachiWebUIInpaint") {
+            const launch = document.createElement("button");
+            launch.type = "button"; launch.textContent = T("Open Adachi Workbench");
+            launch.className = "adachi-workspace-launch";
+            launch.onclick = event => { event.preventDefault(); event.stopPropagation(); document.dispatchEvent(new CustomEvent("adachi-open-workspace", {detail:{sampler:node}})); };
+            root.append(launch);
+        }
+        const fold = document.createElement("details"); fold.className = "dd-workbench-fold";
+        const summary = document.createElement("summary"); summary.textContent = T("Painting tools and mask settings");
+        summary.addEventListener("pointerdown", event => event.stopPropagation());
+        summary.addEventListener("click", event => event.stopPropagation());
+        fold.append(summary);
         const tools = document.createElement("div"); tools.className = "dd-workbench-tools";
         const placeholder = document.createElement("div"); placeholder.className = "dd-param-placeholder dd-tools-placeholder";
         placeholder.textContent = T("Connect an inline mask to show the painting tools");
-        tools.append(placeholder); root.append(tools);
+        tools.append(placeholder); fold.append(tools);
+        const actionHost = document.createElement("div"); actionHost.className = "dd-workbench-actions";
+        root.append(fold, actionHost);
         let source = null, mountedEditor = null, frame = 0, removed = false;
         const markChanged = () => {
             app.canvas?.emitBeforeChange?.(); app.canvas?.emitAfterChange?.();
@@ -28,47 +42,80 @@ export function installInpaintWorkbench(nodeType) {
             const next = canMount ? editor : null;
             if (next !== mountedEditor) {
                 if (mountedEditor?.controls.parentElement === tools) mountedEditor.controls.remove();
+                if (mountedEditor?.actions.parentElement === actionHost) mountedEditor.actions.remove();
                 mountedEditor = next;
-                if (next) tools.append(next.controls);
+                if (next) { tools.append(next.controls); actionHost.append(next.actions); }
                 scheduleSize();
             }
             placeholder.hidden = !!next;
             placeholder.textContent = editor ? T("The painting tools live in the other inpaint node connected to this image") : T("Connect an inline mask to show the painting tools");
         };
-        const contentHeight = () => (tools.scrollHeight || 30) + (panel?.root.scrollHeight || 620);
+        const contentHeight = () => (root.querySelector('.adachi-workspace-launch')?.offsetHeight || 0) + fold.offsetHeight + actionHost.offsetHeight + (panel?.root.scrollHeight || 620);
         const scheduleSize = () => {
             cancelAnimationFrame(frame);
             frame = requestAnimationFrame(() => {
-                if (removed || !node.graph || !root.isConnected) return;
+                if (removed || !node.graph || !root.isConnected || root.closest(".adachi-workspace")) return;
                 const required = node.computeSize()[1];
                 if (Math.abs(node.size[1] - required) > 1) node.setSize([node.size[0], required]);
                 node.graph.setDirtyCanvas(true, true);
             });
         };
         let panel;
+        fold.open = !!node.properties.adachi_tools_open;
+        fold.ontoggle = () => {
+            if (node.properties.adachi_tools_open !== fold.open) {
+                node.properties.adachi_tools_open = fold.open; markChanged();
+            }
+            scheduleSize();
+        };
         panel = new InpaintParameterPanel({
             getTargets: () => { refreshEditor(); return [node]; },
-            getImage: () => source?.adachiEditor?.ready ? source.adachiEditor.image : null,
+            getImage: () => source?.adachiEditor?.ready ? (source.adachiCrop?.mode === "crop" && source.adachiCrop.region.locked ? source.adachiCrop.region : source.adachiEditor.image) : null,
             onTargetChange: (previous, next) => relocateWidgets(node, previous, next),
             onDirty: markChanged, onHeight: scheduleSize,
+            onMaskSettings: element => {
+                fold.querySelector(".dd-param-mask-settings")?.remove();
+                fold.append(element); scheduleSize();
+            },
         });
         root.append(panel.root);
-        node.adachiParameters = panel; node.adachiWorkbench = { root, tools, get source() { return source; } };
+        node.adachiParameters = panel; node.adachiWorkbench = { root, tools, fold, actionHost, get source() { return source; } };
         const stopWheelRouting = routeWheelOutsideImage(root, null, () => false);
         const widget = node.addDOMWidget("adachi_workbench", "adachi_inpaint_controls", root, {
             serialize: false, hideOnZoom: false, getMinHeight: contentHeight, getMaxHeight: () => Math.max(4096, contentHeight()),
         });
         widget.serialize = false;
         bindWidgetToNodeWidth(node, widget, root);
-        const observer = new ResizeObserver(scheduleSize); observer.observe(tools);
+        const observer = new ResizeObserver(scheduleSize); observer.observe(tools); observer.observe(fold); observer.observe(actionHost);
         const configured = node.onConfigure;
-        node.onConfigure = function () { const result = configured?.apply(this, arguments); queueMicrotask(() => { if (!removed) { panel.refresh(); scheduleSize(); } }); return result; };
+        node.onConfigure = function () {
+            const result = configured?.apply(this, arguments);
+            const legacyTitles = ["WebUI Inpaint", "WebUI 局部重绘", "WebUI 局部重绘 / WebUI Inpaint", "安达局部重绘工作台"];
+            if (node.type === "AdachiWebUIInpaint" && legacyTitles.includes(node.title)) {
+                node.title = T("Adachi Inpaint Workbench");
+            }
+            fold.open = !!node.properties.adachi_tools_open;
+            queueMicrotask(() => { if (!removed) { panel.refresh(); scheduleSize(); } });
+            return result;
+        };
         const connections = node.onConnectionsChange;
-        node.onConnectionsChange = function () { const result = connections?.apply(this, arguments); queueMicrotask(() => { if (!removed) panel.refresh(); }); return result; };
+        node.onConnectionsChange = function (type, slot, connected) {
+            const result = connections?.apply(this, arguments);
+            // Graph restoration also connects wires: never overwrite the sizes
+            // saved in a workflow while it is being configured.
+            const userImageConnection = !app.configuringGraph && type === 1 && connected && node.inputs?.[slot]?.name === "image";
+            queueMicrotask(() => {
+                if (removed) return;
+                panel.refresh();
+                if (userImageConnection) syncCurrentDimensions(connectedEditor(node), node);
+            });
+            return result;
+        };
         const previousRemoved = node.onRemoved;
         node.onRemoved = function () {
             removed = true; cancelAnimationFrame(frame); observer.disconnect();
             if (mountedEditor?.controls.parentElement === tools) mountedEditor.controls.remove();
+            if (mountedEditor?.actions.parentElement === actionHost) mountedEditor.actions.remove();
             panel.destroy(); stopWheelRouting(); root.remove();
             return previousRemoved?.apply(this, arguments);
         };

web/inpaint_panel.js

@@ -13,7 +13,7 @@ const choices = {
 export class InpaintParameterPanel {
     constructor(options) {
         this.options = options; this.bindings = []; this.root = el("section", "dd-parameters");
-        const header = el("div", "dd-param-header"); header.append(el("strong", "", T("WebUI Inpaint")));
+        const header = el("div", "dd-param-header"); header.append(el("strong", "", T("Adachi Inpaint Workbench")));
         this.targetSelect = el("select"); this.targetSelect.setAttribute("aria-label", T("Controlled WebUI inpaint node"));
         this.targetSelect.onchange = () => { this.preferred = this.targetSelect.value; this.options.onPreferred?.(this.preferred); this.refresh(); };
         header.append(this.targetSelect); this.root.append(header);
@@ -31,7 +31,7 @@ export class InpaintParameterPanel {
         const signature = targets.map(n => `${n.id}:${n.title}`).join("|");
         if (signature !== this.signature) {
             this.signature = signature; this.targetSelect.replaceChildren();
-            for (const n of targets) { const option = el("option", "", `${n.title || T("WebUI Inpaint")} #${n.id}`); option.value = String(n.id); this.targetSelect.append(option); }
+            for (const n of targets) { const option = el("option", "", `${n.title || T("Adachi Inpaint Workbench")} #${n.id}`); option.value = String(n.id); this.targetSelect.append(option); }
         }
         this.targetSelect.hidden = targets.length < 2;
         if (next) this.targetSelect.value = String(next.id);
@@ -114,8 +114,11 @@ export class InpaintParameterPanel {
     grid(parent) { const grid = el("div", "dd-param-grid"); parent.append(grid); return grid; }
 
     build() {
-        this.enum(this.body, "mask_mode", "Mask mode", true);
-        this.enum(this.body, "fill", "Fill content", true);
+        this.maskSettings = el("div", "dd-param-mask-settings");
+        this.enum(this.maskSettings, "mask_mode", "Mask mode", true);
+        this.enum(this.maskSettings, "fill", "Fill content", true);
+        this.body.append(this.maskSettings);
+        this.options.onMaskSettings?.(this.maskSettings);
         this.enum(this.body, "area", "Sampling area", true);
         const resolution = el("div", "dd-param-resolution"); resolution.append(el("span", "dd-param-label", T("Sampling resolution")));
         this.sourceSize = el("button", "", T("Use source size")); this.sourceSize.type = "button";

web/legacy_feedback.js

@@ -0,0 +1,32 @@
+// Remove obsolete feedback edges before LiteGraph assigns saved socket indices.
+export function removeLegacyFeedback(graph, seen = new Set()) {
+    if (!graph || seen.has(graph)) return;
+    seen.add(graph);
+    const nodes = new Map((graph.nodes || []).map(node => [String(node.id), node]));
+    const removed = new Set(), indices = new Map();
+    for (const node of nodes.values()) {
+        if (node.type !== "AdachiInlineMask") continue;
+        const slots = (node.inputs || []).flatMap((input, index) => input.name === "result" ? [index] : []);
+        indices.set(String(node.id), slots);
+        for (const slot of slots) if (node.inputs[slot].link != null) removed.add(String(node.inputs[slot].link));
+    }
+    const entries = Array.isArray(graph.links) ? graph.links.map((link, index) => [index, link]) : Object.entries(graph.links || {});
+    for (const [, link] of entries) {
+        const id = Array.isArray(link) ? link[0] : link.id;
+        const target = String(Array.isArray(link) ? link[3] : link.target_id);
+        const slot = Array.isArray(link) ? link[4] : link.target_slot;
+        if (!indices.has(target)) continue;
+        const input = nodes.get(target).inputs?.[slot], slots = indices.get(target);
+        // Also repair orphaned feedback edges from a previously migrated save.
+        if (slots.includes(slot) || String(input?.link) !== String(id)) removed.add(String(id));
+        else if (Array.isArray(link)) link[4] -= slots.filter(index => index < slot).length;
+        else link.target_slot -= slots.filter(index => index < slot).length;
+    }
+    if (Array.isArray(graph.links)) graph.links = graph.links.filter(link => !removed.has(String(Array.isArray(link) ? link[0] : link.id)));
+    else for (const [key, link] of entries) if (removed.has(String(Array.isArray(link) ? link[0] : link.id))) delete graph.links[key];
+    for (const node of nodes.values()) {
+        if (node.type === "AdachiInlineMask") node.inputs = (node.inputs || []).filter(input => input.name !== "result");
+        for (const output of node.outputs || []) if (Array.isArray(output.links)) output.links = output.links.filter(id => !removed.has(String(id)));
+    }
+    for (const subgraph of graph.definitions?.subgraphs || []) removeLegacyFeedback(subgraph, seen);
+}

web/parameter_bridge.js

@@ -35,6 +35,34 @@ export function connectedSamplers(source) {
         .sort((a, b) => Number(b.type === "AdachiWebUIInpaint") - Number(a.type === "AdachiWebUIInpaint"));
 }
 
+// Use the same rounding, range and linked-input rules as Use source size.
+export function syncCropDimensions(source, width, height, onlyTarget = null) {
+    const aligned = value => Math.max(64, Math.min(8192, Math.round(value / 8) * 8));
+    for (const target of connectedSamplers(source)) {
+        if (onlyTarget && target !== onlyTarget) continue;
+        target.adachiParameters?.refresh();
+        for (const [name, value] of [["width", width], ["height", height]]) {
+            if (target.inputs?.some(input => input.name === name && input.link != null)) continue;
+            const widget = target.widgets?.find(widget => widget.name === name);
+            if (widget && widget.value !== aligned(value)) {
+                widget.value = aligned(value);
+                widget.callback?.(widget.value);
+            }
+        }
+        target.adachiParameters?.refresh();
+    }
+}
+
+// The crop output does not have dimensions until the user locks a region.
+export function syncCurrentDimensions(source, onlyTarget = null) {
+    const editor = source?.adachiEditor, crop = source?.adachiCrop;
+    if (!editor?.ready || !editor.image) return;
+    const region = crop?.mode === "crop" ? crop.region : null;
+    if (crop?.mode === "crop" && !region?.locked) return;
+    const image = region || editor.image;
+    syncCropDimensions(source, image.naturalWidth || image.width, image.naturalHeight || image.height, onlyTarget);
+}
+
 export function relocateWidgets(owner, previous, next) {
     if (previous) {
         for (const widget of previous.widgets || []) {

web/preview_images.js

@@ -0,0 +1,44 @@
+// A failed half of a comparison must not leave its sibling decoding in the
+// background, retain a corrupt blob forever, or leak a late object URL.
+export async function decodePreview(blob, signal, {ImageClass=globalThis.Image, urls=URL, timeoutMs=20000}={}) {
+    signal?.throwIfAborted();
+    const url=urls.createObjectURL(blob), image=new ImageClass();
+    let timer, abort;
+    const stopped=new Promise((_,reject)=>{
+        abort=()=>{image.src='';reject(new DOMException('Cancelled','AbortError'));};
+        signal?.addEventListener('abort',abort,{once:true});
+        timer=setTimeout(()=>{image.src='';reject(Error('Preview image decoding timed out'));},timeoutMs);
+    });
+    try {
+        image.src=url;
+        await Promise.race([image.decode(),stopped]);
+        signal?.throwIfAborted();
+        const width=image.naturalWidth||image.width,height=image.naturalHeight||image.height;
+        if(!width || !height || width*height>32_000_000 || Math.max(width,height)>16384)throw Error('Preview image exceeds supported dimensions');
+        return {image,url};
+    } catch(error) {urls.revokeObjectURL(url);throw error;}
+    finally {clearTimeout(timer);signal?.removeEventListener('abort',abort);}
+}
+
+export async function loadPreviewPair(refs,{loadPreview,cached,signal,onProgress,onDecoding,decode=decodePreview,revoke=url=>URL.revokeObjectURL(url)}) {
+    signal.throwIfAborted();
+    const pair=new AbortController(),abort=()=>pair.abort();
+    signal.addEventListener('abort',abort,{once:true});
+    let failure;
+    try {
+        const results=await Promise.allSettled(refs.map(async(ref,index)=>{
+            try {
+                const blob=cached[index]||await loadPreview(ref,{signal:pair.signal,onProgress:state=>{if(!pair.signal.aborted)onProgress?.(index,state);}});
+                pair.signal.throwIfAborted();cached[index]=blob;
+                onDecoding?.();
+                try {return await decode(blob,pair.signal);}
+                catch(error){if(error.name!=='AbortError')cached[index]=null;throw error;}
+            } catch(error) {failure??=error;pair.abort();throw error;}
+        }));
+        if(failure || signal.aborted) {
+            for(const result of results)if(result.status==='fulfilled')revoke(result.value.url);
+            signal.throwIfAborted();throw failure;
+        }
+        return results.map(result=>result.value);
+    } finally {signal.removeEventListener('abort',abort);}
+}

web/preview_transport.js

@@ -0,0 +1,51 @@
+// Read through ComfyUI's authenticated API and report received bytes. The
+// deadline measures inactivity, so a slow but advancing cloud download works.
+export async function readPreview(fetchApi, path, {signal, onProgress, idleMs=30000, maxBytes=128*1024*1024}={}) {
+    const controller=new AbortController();let timer,timedOut=false;
+    const abort=()=>controller.abort();
+    const reset=()=>{clearTimeout(timer);timer=setTimeout(()=>{timedOut=true;controller.abort();},idleMs);};
+    if(signal?.aborted)controller.abort();else signal?.addEventListener('abort',abort,{once:true});
+    try {
+        controller.signal.throwIfAborted();
+        reset();
+        const response=await fetchApi(path,{signal:controller.signal,cache:'no-store'});
+        if(!response.ok)throw Object.assign(new Error(`Preview request failed (${response.status})`),{status:response.status});
+        const encoded=response.headers?.get('content-encoding');
+        const length=Number(response.headers?.get('content-length'));
+        const total=encoded && encoded!=='identity'?0:Number.isFinite(length)&&length>0?length:0;
+        if(total>maxBytes)throw Error('Preview exceeds the image size limit');
+        onProgress?.({loaded:0,total});
+        let blob;
+        if(response.body?.getReader) {
+            const reader=response.body.getReader(),chunks=[];let loaded=0;
+            const cancelRead=()=>{reader.cancel().catch(()=>{});};
+            controller.signal.addEventListener('abort',cancelRead,{once:true});
+            try {
+                while(true) {
+                    controller.signal.throwIfAborted();
+                    const {done,value}=await reader.read();
+                    controller.signal.throwIfAborted();
+                    if(done)break;
+                    if(!value.byteLength)continue;
+                    reset();loaded+=value.byteLength;
+                    if(loaded>maxBytes)throw Error('Preview exceeds the image size limit');
+                    chunks.push(value);onProgress?.({loaded,total});
+                }
+                if(total && loaded<total)throw Error('Preview download was interrupted');
+                blob=new Blob(chunks,{type:response.headers?.get('content-type')||'image/png'});
+                onProgress?.({loaded,total:loaded,complete:true});
+            } catch(error) {await reader.cancel().catch(()=>{});throw error;}
+            finally {controller.signal.removeEventListener('abort',cancelRead);reader.releaseLock();}
+        } else {
+            blob=await response.blob();
+            controller.signal.throwIfAborted();
+            if(blob.size>maxBytes)throw Error('Preview exceeds the image size limit');
+            onProgress?.({loaded:blob.size,total:blob.size,complete:true});
+        }
+        if(!blob.size)throw Error('Empty preview image');
+        return blob;
+    } catch(error) {
+        if(timedOut)throw Error('Preview download timed out');
+        throw error;
+    } finally {clearTimeout(timer);signal?.removeEventListener('abort',abort);controller.abort();}
+}

web/result_adoption.js

@@ -0,0 +1,60 @@
+import { api } from "../../scripts/api.js";
+import { app } from "../../scripts/app.js";
+import { executionId } from "./workspace_flow.js";
+import { T } from "./i18n.js";
+import {uploadImageWithProgress} from './upload_transport.js?v=stability1';
+
+export function installResultAdoption(node, crop) {
+    // Saved workflows may still contain the old feedback socket and wire.
+    const removeLegacyInput = () => {
+        const index = node.inputs?.findIndex(input => input.name === "result");
+        if (index >= 0) node.removeInput(index);
+    };
+    removeLegacyInput();
+    const configured = node.onConfigure;
+    node.onConfigure = function () {
+        const result = configured?.apply(this, arguments);
+        removeLegacyInput();
+        return result;
+    };
+    node.adachiAdoptCrop = async (file, origin, {signal}={}) => {
+        const editor = crop.editor, region = crop.region;
+        if (!editor.ready || !region.locked) throw Error(T("Select a crop region before adopting a result"));
+        if (crop.busy || editor.stroke) throw Error(T("Finish editing before adopting a result"));
+        if (origin && (origin.mode !== "crop" || String(origin.source) !== executionId(app.rootGraph||app.graph,node) || origin.job !== region.job)) {
+            throw Error(T("This result belongs to another crop; generate a result for the current region"));
+        }
+        if (file.size > 64 * 1024 * 1024) throw Error(T("File is too large (max 64 MB)"));
+        signal?.throwIfAborted();
+        const controller=crop.adoptController=new AbortController(),abort=()=>controller.abort();
+        signal?.addEventListener('abort',abort,{once:true});
+        crop.busy = true; crop.sync();
+        const objectUrl = URL.createObjectURL(file);
+        try {
+            const image = new Image(); image.src = objectUrl; await image.decode();
+            controller.signal.throwIfAborted();
+            if (image.width !== region.width || image.height !== region.height) {
+                throw Error(T("Result dimensions must match the selected crop"));
+            }
+            const body = new FormData();
+            body.append("image", file, `adopted-${crypto.randomUUID()}.png`);
+            body.append("type", "input"); body.append("subfolder", "adachi-inpaint-layers");
+            const uploaded = await uploadImageWithProgress(api,body,null,{signal:controller.signal});
+            if (typeof uploaded.name !== "string" || !uploaded.name) throw Error(T("Upload failed"));
+            if (editor.destroyed || crop.mode !== "crop" || crop.region.job !== region.job) {
+                throw Error(T("The crop changed while adopting the result"));
+            }
+            const descriptor = {filename:uploaded.name, subfolder:uploaded.subfolder || "", type:"input"};
+            await crop.accept({job:region.job, image:descriptor}, {allowBusy:true, image});
+            const name = descriptor.subfolder ? `${descriptor.subfolder}/${descriptor.filename}` : descriptor.filename;
+            if (crop.widgets.layer_image.value !== name) {
+                // accept() may have skipped silently (busy / changed job / duplicate).
+                throw Error(T("The result was not applied; the crop or layer state changed"));
+            }
+            editor.status("");
+        } finally {
+            signal?.removeEventListener('abort',abort);if(crop.adoptController===controller)crop.adoptController=null;
+            URL.revokeObjectURL(objectUrl); crop.busy = false; crop.sync();
+        }
+    };
+}

web/result_recovery.js

@@ -0,0 +1,93 @@
+import {T} from './i18n.js';
+// Execution notifications are an optimization, not the sole delivery channel.
+// Poll only IDs queued by this browser; never guess from another workflow's
+// latest history entry or from a coincidentally matching node ID.
+import {executionNode,graphNodeEntries} from './workspace_flow.js';
+export class ResultRecovery {
+    constructor({fetchApi,getGraph,onOutput,onPending=()=>{},onTerminal=()=>{},onDiscard=()=>{},interval=2500}) {
+        Object.assign(this,{fetchApi,getGraph,onOutput,onPending,onTerminal,onDiscard,interval});
+        this.pending=new Map();this.delivered=new Set();this.controllers=new Set();
+    }
+    track(graph,promptId,nodeId) {
+        const node=executionNode(graph,nodeId),key=`${promptId}:${nodeId}`;
+        if(this.destroyed || !promptId || node?.type!=='AdachiImageCompare' || this.delivered.has(key) || this.pending.has(key))return;
+        this.pending.set(key,{graph,node,promptId,nodeId:String(nodeId),key});
+        this.onPending(node);this.schedule(0);
+    }
+    schedule(ms=this.interval) {clearTimeout(this.timer);if(!this.destroyed && this.pending.size)this.timer=setTimeout(()=>this.recover(),ms);}
+    discard(record){if(this.pending.delete(record.key))this.onDiscard(record.node);}
+    async recover() {
+        if(this.destroyed || this.busy)return;this.busy=true;clearTimeout(this.timer);
+        const groups=new Map();
+        for(const record of this.pending.values()) {
+            if(record.graph!==this.getGraph() || executionNode(record.graph,record.nodeId)!==record.node){this.discard(record);continue;}
+            const records=groups.get(record.promptId)||[];records.push(record);groups.set(record.promptId,records);
+        }
+        try {
+            for(const [promptId,records] of groups) {
+                if(this.destroyed)break;
+                const controller=new AbortController(),timer=setTimeout(()=>controller.abort(),15000);
+                this.controllers.add(controller);
+                try {
+                    const response=await this.fetchApi(`/history/${encodeURIComponent(promptId)}`,{cache:'no-store',signal:controller.signal});
+                    if(!response.ok)continue;
+                    const entry=(await response.json())[promptId];if(!entry)continue;
+                    for(const r of records) {
+                        if(this.destroyed || !this.pending.has(r.key))continue;
+                        if(r.graph!==this.getGraph() || executionNode(r.graph,r.nodeId)!==r.node){this.discard(r);continue;}
+                        const output=entry.outputs?.[r.nodeId];
+                        if(output?.adachi_originals?.length && output?.adachi_results?.length) {
+                            await this.onOutput(r.node,output,promptId,entry);
+                            this.pending.delete(r.key);this.delivered.add(r.key);
+                        } else if(entry.status?.completed) {
+                            this.pending.delete(r.key);this.onTerminal(r.node,entry,promptId);
+                        }
+                    }
+                } catch {} finally {clearTimeout(timer);this.controllers.delete(controller);controller.abort();}
+            }
+        } finally {
+            while(this.delivered.size>256)this.delivered.delete(this.delivered.values().next().value);
+            this.busy=false;this.schedule();
+        }
+    }
+    destroy(){this.destroyed=true;clearTimeout(this.timer);for(const controller of this.controllers)controller.abort();for(const record of this.pending.values())this.discard(record);this.delivered.clear();}
+}
+
+const installed=new WeakMap();
+export function installComparisonRecovery(api,getGraph) {
+    if(installed.has(api))return installed.get(api);
+    const recovery=new ResultRecovery({fetchApi:api.fetchApi.bind(api),getGraph,
+        onDiscard:node=>{const view=node.adachiCompare;if(view){view.pendingOutputs=Math.max(0,(view.pendingOutputs||0)-1);view.updateButton();if(!view.pendingOutputs&&!view.loadingActive)view.progressBox.hidden=true;}},
+        onPending:node=>{const view=node.adachiCompare;if(view){view.pendingOutputs=(view.pendingOutputs||0)+1;view.updateButton();view.loading(view.options.waitingText||T("Waiting for the generated image…"));}},
+        onOutput(node,output,promptId,entry){
+            if(node.adachiCompare)node.adachiCompare.pendingOutputs=Math.max(0,(node.adachiCompare.pendingOutputs||0)-1);
+            const known=new Set((node.properties.adachi_compare?.results||[]).map(ref=>JSON.stringify([ref.type,ref.subfolder,ref.filename])));
+            if(output.adachi_results.some(ref=>!known.has(JSON.stringify([ref.type,ref.subfolder,ref.filename]))))node.onExecuted?.(output);
+            else if(node.adachiCompare?.ready)node.adachiCompare.progressBox.hidden=!(node.adachiCompare.pendingOutputs>0);
+            node.adachiCompare?.updateButton();
+            document.dispatchEvent(new CustomEvent('adachi-result-recovered',{detail:{node,promptId,entry}}));
+        },
+        onTerminal(node,entry,promptId){
+            const view=node.adachiCompare;if(view){view.pendingOutputs=Math.max(0,(view.pendingOutputs||0)-1);view.updateButton();view.progressBox.hidden=!(view.pendingOutputs>0);view.status(view.options.failedText||T("Generation failed; check the ComfyUI task history"),true);}
+            document.dispatchEvent(new CustomEvent('adachi-result-recovered',{detail:{node,promptId,entry}}));
+        },
+    });
+    const queued=api.queuePrompt;
+    api.queuePrompt=async function(...args) {
+        const graph=getGraph(),nodes=Object.entries(args[1]?.output||{}).filter(([,n])=>n.class_type==='AdachiImageCompare').map(([id])=>id);
+        const result=await queued.apply(this,args);
+        for(const id of nodes)recovery.track(graph,result.prompt_id,id);
+        return result;
+    };
+    const wake=()=>{
+        recovery.schedule(0);
+        for(const {node} of graphNodeEntries(getGraph())) {
+            const view=node.adachiCompare;
+            if(view && !view.ready && !view.loadingActive && view.results.length)view.loadSelected();
+        }
+    };
+    for(const event of ['reconnected','execution_success','execution_error','execution_interrupted'])api.addEventListener(event,()=>recovery.schedule(0));
+    api.addEventListener('reconnected',wake);window.addEventListener('online',wake);
+    document.addEventListener('visibilitychange',()=>{if(!document.hidden)wake();});
+    installed.set(api,recovery);return recovery;
+}

web/upload_transport.js

@@ -0,0 +1,48 @@
+import { T } from "./i18n.js";
+
+// XMLHttpRequest reports bytes sent, unlike fetch. Keep ComfyUI's API URL and
+// per-user header; the browser automatically sends same-origin session cookies.
+export async function uploadImageWithProgress(api, body, onProgress, {signal,idleMs=60000}={}) {
+    const controller=new AbortController();let timer,timedOut=false;
+    const abort=()=>controller.abort();
+    const reset=(ms=idleMs)=>{clearTimeout(timer);timer=setTimeout(()=>{timedOut=true;controller.abort();},ms);};
+    if(signal?.aborted)controller.abort();else signal?.addEventListener('abort',abort,{once:true});
+    try {
+    controller.signal.throwIfAborted();reset();
+    const response = await new Promise((resolve, reject) => {
+        const xhr = new XMLHttpRequest();
+        let settled=false,lastLoaded=0;
+        const finish=(fn,value)=>{if(settled)return;settled=true;controller.signal.removeEventListener('abort',cancel);xhr.onload=xhr.onerror=xhr.onabort=null;xhr.upload.onprogress=xhr.upload.onload=null;fn(value);};
+        const cancel=()=>{xhr.abort();finish(reject,new DOMException(T('Upload was cancelled'),'AbortError'));};
+        controller.signal.addEventListener('abort',cancel,{once:true});
+        xhr.open("POST", api.apiURL("/upload/image"));
+        if (api.user) xhr.setRequestHeader("Comfy-User", api.user);
+        xhr.upload.onprogress = event => {if(event.loaded>lastLoaded){lastLoaded=event.loaded;reset();}onProgress?.({phase: "upload", loaded: event.loaded, total: event.lengthComputable ? event.total : 0});};
+        xhr.upload.onload = () => {reset();onProgress?.({phase: "processing"});};
+        xhr.onload = () => finish(resolve,xhr);
+        xhr.onerror = () => finish(reject,new Error(T("Network error while uploading image")));
+        xhr.onabort = () => finish(reject,new DOMException(T("Upload was cancelled"),'AbortError'));
+        try{xhr.send(body);}catch(error){finish(reject,error);}
+    });
+
+    // Some cloud frontends wrap fetchApi with an authentication mechanism that
+    // is not exposed to XHR. Retry rejected authentication through that same
+    // frontend API, without inventing upload percentages for the fallback.
+    if (response.status === 401 || response.status === 403) {
+        onProgress?.({phase: "authenticated"});
+        // fetch cannot expose upload byte progress. Give authenticated uploads
+        // a larger bounded budget rather than applying XHR's inactivity limit.
+        reset(idleMs*5);
+        const fallback = await api.fetchApi("/upload/image", {method: "POST", body, signal:controller.signal});
+        if (!fallback.ok) throw new Error(`${T("Upload failed")} (${fallback.status})`);
+        onProgress?.({phase: "processing"});
+        return await fallback.json();
+    }
+    if (response.status < 200 || response.status >= 300) throw new Error(`${T("Upload failed")} (${response.status})`);
+    try { return JSON.parse(response.responseText); }
+    catch { throw new Error(T("Invalid upload response")); }
+    } catch(error) {
+        if(timedOut)throw Error(T('Image upload timed out; check the connection and try again'));
+        throw error;
+    } finally {clearTimeout(timer);signal?.removeEventListener('abort',abort);controller.abort();}
+}

web/wheel_routing.js

@@ -3,6 +3,7 @@ import { app } from "../../scripts/app.js";
 export function routeWheelOutsideImage(root, canvas, isReady) {
     const route = event => {
         if (!root.contains(event.target)) return;
+        if (root.closest(".adachi-workspace")) return;
         if (event.target === canvas && isReady()) return;
         event.preventDefault(); event.stopPropagation();
         // Preserve the user's ComfyUI navigation mode (zoom or pan).

web/workbench_analysis.js

@@ -0,0 +1,44 @@
+// Read-only discovery for the inpainting workbench. The live graph remains
+// authoritative: virtual Get/Set routes and group bypasses are resolved by
+// ComfyUI and its installed extensions when a prompt is actually queued.
+import {graphNodeEntries,executionNode} from './workspace_flow.js';
+export function inspectWorkbench(graph) {
+    const nodes = graphNodeEntries(graph).map(entry=>entry.node);
+    const find = type => nodes.filter(node => node.type === type);
+    const editors = find("AdachiInlineMask");
+    const samplers = find("AdachiWebUIInpaint");
+    const comparisons = find("AdachiImageCompare");
+    const presets = find("AdachiGroupPresets");
+    const checkpoints = find("CheckpointLoaderSimple");
+    const loras = nodes.filter(node => node.type === "LoraLoader" || node.type === "Power Lora Loader (rgthree)");
+    const missing = [];
+    if (!editors.length) missing.push("inline_mask");
+    if (!samplers.length) missing.push("inpaint");
+    if (!comparisons.length) missing.push("compare");
+    return { editors, samplers, comparisons, presets, checkpoints, loras, missing };
+}
+
+export function upstreamNodes(node,inputName) {
+    const found=new Set(),queue=[];
+    const addInput=(current,input)=>{const g=current.graph,l=g?.getLink?.(input?.link)??g?.links?.get?.(input?.link)??g?.links?.[input?.link];if(l)queue.push(g.getNodeById(l.origin_id));};
+    for(const input of node?.inputs||[])if(!inputName||input.name===inputName)addInput(node,input);
+    while(queue.length){const current=queue.shift();if(!current||found.has(current))continue;found.add(current);
+        if(current.type==='GetNode'){try{queue.push(current.findSetter?.(current.graph));}catch{}}
+        for(const input of current.inputs||[])addInput(current,input);
+    }
+    return found;
+}
+export function selectWorkbenchPipeline(graph,preferred,options={}) {
+    const nodes=inspectWorkbench(graph),choose=(list,explicit,candidates=[])=>{
+        if(explicit&&list.includes(explicit))return {node:explicit,ambiguous:false};
+        const pool=candidates.length?candidates:list,active=pool.filter(n=>n.mode===0);
+        if(pool.length===1)return {node:pool[0],ambiguous:false};
+        if(active.length===1)return {node:active[0],ambiguous:false};
+        return {node:active[0]||pool[0],ambiguous:pool.length>1};
+    };
+    const sampler=choose(nodes.samplers,preferred||options.samplerNode);
+    const saved=sampler.node?.properties?.adachi_workspace_connections;
+    const editor=choose(nodes.editors,options.editorNode||executionNode(graph,saved?.editorId),nodes.editors.filter(n=>upstreamNodes(sampler.node,'image').has(n)));
+    const compare=choose(nodes.comparisons,options.compareNode||executionNode(graph,saved?.compareId),nodes.comparisons.filter(n=>upstreamNodes(n,'result').has(sampler.node)));
+    return {...nodes,editor:editor.node,sampler:sampler.node,compare:compare.node,ambiguous:sampler.ambiguous||editor.ambiguous||compare.ambiguous};
+}

web/workspace.css

@@ -0,0 +1,156 @@
+.adachi-workspace {
+  --aw-bg:#eef1f4; --aw-panel:#fff; --aw-input:#f6f8fa; --aw-text:#273747;
+  --aw-muted:#61758a; --aw-line:#ced7e1; --aw-hover:#edf3f9;
+  --aw-accent:#276da7; --aw-selected:#eaf3fb; --aw-selected-text:#205b8b;
+  --aw-stage:#e9edf1; --aw-checker:#e0e5eb; --aw-error:#a93e34; --aw-ok:#237459;
+  position:fixed; inset:0; z-index:12000; display:flex; flex-direction:column;
+  background:var(--aw-bg); color:var(--aw-text); color-scheme:light;
+  font:14px/1.5 'Segoe UI','Microsoft YaHei',sans-serif;
+}
+.adachi-workspace[data-theme=dark] {
+  --aw-bg:#171a1e; --aw-panel:#24282e; --aw-input:#2d333b; --aw-text:#e3e8ef;
+  --aw-muted:#a2b1c2; --aw-line:#46515e; --aw-hover:#343e49;
+  --aw-accent:#357fb8; --aw-selected:#263f55; --aw-selected-text:#b4dcff;
+  --aw-stage:#191d22; --aw-checker:#232931; --aw-error:#ffa69a; --aw-ok:#83d5b5;
+  color-scheme:dark;
+}
+.adachi-workspace * {box-sizing:border-box}
+.adachi-workspace [hidden] {display:none!important}
+body.adachi-workspace-open > :not(.adachi-workspace):not(script):not(style):not(link) {opacity:0!important;pointer-events:none!important}
+.adachi-workspace button,.adachi-workspace select,.adachi-workspace input,.adachi-workspace textarea {
+  font:inherit; color:var(--aw-text); background:var(--aw-panel); border:1px solid var(--aw-line);
+  border-radius:5px; padding:7px 10px;
+}
+.adachi-workspace button {cursor:pointer;white-space:nowrap}
+.adachi-workspace button:hover {border-color:var(--aw-accent);background:var(--aw-hover)}
+.adachi-workspace button:disabled {opacity:.4;cursor:not-allowed}
+.adachi-workspace button:focus-visible,.adachi-workspace select:focus-visible {outline:2px solid var(--aw-accent);outline-offset:2px}
+.adachi-workspace button.active,.adachi-workspace button[aria-pressed=true],.adachi-workspace .dd-inpaint button.dd-active {
+  background:var(--aw-selected);color:var(--aw-selected-text);border-color:var(--aw-accent);
+}
+.adachi-workspace button.primary {background:var(--aw-accent);border-color:var(--aw-accent);color:#fff}
+.adachi-workspace button.primary:hover {filter:brightness(1.1)}
+.adachi-workspace .aw-icon {display:inline-flex;align-items:center;justify-content:center;width:32px;height:32px;padding:5px;flex-shrink:0}
+.adachi-workspace header {position:absolute;z-index:6;top:8px;left:20px;right:20px;height:36px;display:flex;align-items:center;gap:8px;pointer-events:none}
+.adachi-workspace header>* {pointer-events:auto}
+.adachi-workspace h1 {font-size:18px;font-weight:650;margin:0 auto 0 0;padding:3px 0;background:transparent;color:rgba(0,0,0,.62);white-space:nowrap}
+.adachi-workspace[data-theme=dark] h1 {text-shadow:0 0 1px rgba(255,255,255,.9),0 0 3px rgba(255,255,255,.65)}
+.adachi-workspace .aw-status {font-size:12px;min-width:0;max-width:42vw;flex:0 1 auto;padding:4px 8px;border-radius:4px;background:var(--aw-panel);color:var(--aw-muted);overflow:hidden;white-space:nowrap;text-overflow:ellipsis}
+.adachi-workspace .aw-status[data-state=ok] {color:var(--aw-ok)}
+.adachi-workspace .aw-status[data-state=error] {color:var(--aw-error)}
+.adachi-workspace main {display:grid;grid-template-columns:minmax(0,1fr) 10px var(--aw-sidebar-width,360px);gap:0;padding:6px;flex:1;min-height:0}
+.adachi-workspace .aw-canvas {position:relative;min-height:0;overflow:hidden;border:1px solid var(--aw-line);border-radius:8px;background:var(--aw-panel)}
+.adachi-workspace .aw-surface {position:absolute;inset:0;min-height:0;background:var(--aw-stage)}
+.adachi-workspace .aw-toolbar {position:absolute;top:48px;left:8px;right:8px;z-index:5;display:flex;gap:6px;align-items:center;pointer-events:none}
+.adachi-workspace .aw-toolbar button {pointer-events:auto;padding:5px 9px}
+.adachi-workspace .aw-toolbar .aw-icon {padding:5px}
+.adachi-workspace .aw-toolbar .spacer {flex:1}
+.adachi-workspace .aw-footer {position:absolute;bottom:8px;left:8px;z-index:5;display:flex;align-items:center;flex-wrap:wrap;gap:6px;padding:5px;border:1px solid var(--aw-line);border-radius:5px;background:var(--aw-panel);max-width:calc(100% - 16px);font-size:12px;color:var(--aw-muted)}
+.adachi-workspace .aw-footer button {padding:4px 8px}
+.adachi-workspace .aw-footer select {max-width:260px;min-width:0;padding:4px 7px}
+.adachi-workspace .aw-note {display:none}
+.adachi-workspace .aw-result-image {width:100%;height:100%;position:absolute;object-fit:contain;background:var(--aw-stage)}
+.adachi-workspace aside {overflow:auto;background:var(--aw-panel);border:1px solid var(--aw-line);border-radius:8px;padding:48px 16px 16px}
+.adachi-workspace aside h2 {font-size:12px;letter-spacing:1px;color:var(--aw-muted);margin:0 0 12px}
+.adachi-workspace aside label {margin-top:10px}
+.adachi-workspace aside select,.adachi-workspace aside textarea {max-width:100%}
+.adachi-workspace aside textarea {width:100%;min-height:85px;resize:vertical}
+.adachi-workspace details {margin:8px 0}
+.adachi-workspace summary {cursor:pointer}
+.adachi-workspace .aw-presets {display:flex;gap:6px;flex-wrap:wrap;margin-bottom:12px}
+.adachi-workspace .aw-presets button {flex:1;padding:7px 5px;font-size:12px}
+.adachi-workspace .aw-profile-warning {border:1px solid var(--aw-line);border-left:3px solid var(--aw-accent);padding:8px;margin:0 0 12px;font-size:12px;color:var(--aw-muted)}
+.adachi-workspace .aw-bindings label {display:block;font-size:12px;color:var(--aw-muted)}
+.adachi-workspace .aw-bindings label select {width:100%;display:block;margin:5px 0 10px}
+.adachi-workspace .aw-bindings textarea {display:block;margin-top:5px;line-height:1.5}
+.adachi-workspace .aw-bindings label>button {font-size:12px;padding:1px 6px;margin-top:3px}
+.adachi-workspace .aw-bindings details {border-bottom:1px solid var(--aw-line);padding:5px 0 10px}
+.adachi-workspace .aw-bindings summary {font-size:12px;color:var(--aw-muted)}
+.adachi-workspace .aw-lora-row {display:grid;grid-template-columns:minmax(0,1fr) 65px 30px;gap:4px;margin:6px 0}
+.adachi-workspace .aw-lora-row input {min-width:0;width:100%;padding:5px}
+.adachi-workspace .aw-lora-row button {padding:3px}
+.adachi-workspace .aw-embedded {position:absolute!important;inset:0!important;width:100%!important;height:100%!important;max-height:none!important;transform:none!important;visibility:visible!important}
+.adachi-workspace .dd-inpaint {--dd-bg:var(--aw-panel);--dd-panel:var(--aw-input);--dd-line:var(--aw-line);--dd-text:var(--aw-text);--dd-muted:var(--aw-muted);--dd-accent:var(--aw-accent);color:var(--aw-text);color-scheme:inherit;background:var(--aw-panel)}
+.adachi-workspace .dd-inpaint button {background:var(--aw-input);color:var(--aw-text)}
+.adachi-workspace .dd-inpaint .dd-stage {background-color:var(--aw-stage);background-image:linear-gradient(45deg,var(--aw-checker) 25%,transparent 25%),linear-gradient(-45deg,var(--aw-checker) 25%,transparent 25%),linear-gradient(45deg,transparent 75%,var(--aw-checker) 75%),linear-gradient(-45deg,transparent 75%,var(--aw-checker) 75%)}
+.adachi-workspace .dd-inpaint input,.adachi-workspace .dd-inpaint select {background:var(--aw-input);color:var(--aw-text)}
+.adachi-workspace .dd-inline-minimal {background:transparent}
+.adachi-workspace .dd-crop-bar {background:var(--aw-panel)}
+.adachi-workspace .dd-inpaint .dd-upload-progress {background:var(--aw-panel);color:var(--aw-text)}
+.adachi-workspace .aw-parameters {position:relative;min-height:100px}
+.adachi-workspace .aw-parameters>.dd-inpaint {position:relative!important;transform:none!important;inset:auto!important;height:auto!important;max-height:none!important;overflow:visible!important;width:100%!important}
+.adachi-workspace .dd-param-advanced-body {padding:8px 0}
+.adachi-workspace .dd-workbench {border:0}
+.adachi-workspace .dd-editor-controls {background:var(--aw-panel)}
+.adachi-workspace .dd-inpaint .dd-actions {padding:8px 0}
+.adachi-workspace .dd-inpaint .dd-settings {gap:6px;padding:6px 0}
+.adachi-workspace .dd-inpaint .dd-tools {padding:5px 0}
+.adachi-workspace .dd-inpaint .dd-tools button {padding:4px;font-size:11px}
+.adachi-workspace .dd-inpaint .dd-stage canvas {outline:none}
+.adachi-workspace .dd-inline-minimal>.dd-status:not([hidden]) {background:var(--aw-panel);color:var(--aw-error)}
+.adachi-workspace .dd-inpaint .dd-param-grid {gap:8px}
+.adachi-workspace .aw-canvas .dd-files,.adachi-workspace .aw-canvas .dd-mode-select {top:88px}
+.adachi-workspace .aw-canvas .dd-crop-bar {
+  bottom:calc(var(--aw-history-height,40px) + 16px);left:8px;right:8px;
+  background:var(--aw-panel);border:1px solid var(--aw-line);padding:5px;gap:5px;
+  box-shadow:0 2px 8px #0002;
+}
+.adachi-workspace .aw-canvas .dd-crop-bar button,.adachi-workspace .aw-canvas .dd-crop-bar select,.adachi-workspace .aw-canvas .dd-crop-bar input {
+  height:28px;min-height:28px;font-size:12px;color:var(--aw-text);
+}
+.adachi-workspace .aw-canvas .dd-files button {opacity:1}
+.adachi-workspace .aw-canvas .dd-adoption {display:none}
+.adachi-workspace .aw-canvas .dd-compare-stage {position:absolute;inset:0;min-height:0}
+.adachi-workspace .aw-canvas .dd-compare-toolbar {
+  position:absolute;left:8px;right:8px;bottom:calc(var(--aw-history-height,40px) + 16px);z-index:4;
+  padding:5px;border:1px solid var(--aw-line);border-radius:5px;background:var(--aw-panel);width:fit-content;max-width:calc(100% - 16px);
+}
+.adachi-workspace .aw-canvas .dd-status:not([hidden]) {
+  position:absolute;left:8px;right:8px;z-index:5;bottom:calc(var(--aw-history-height,40px) + var(--aw-crop-height,32px) + 24px);
+  padding:7px;border-radius:5px;background:var(--aw-panel);color:var(--aw-error);
+}
+.adachi-workspace .adachi-workspace-launch {display:none}
+@media(max-width:850px) {
+  .adachi-workspace main {grid-template-columns:1fr;overflow:auto}
+  .adachi-workspace .aw-canvas {min-height:78vh}
+  .adachi-workspace aside {overflow:visible;padding-top:14px}
+  .adachi-workspace header {left:12px;right:12px;height:36px;gap:5px}
+  .adachi-workspace header h1 {font-size:14px}
+  .adachi-workspace header button {padding:4px 6px;font-size:11px}
+  .adachi-workspace header .aw-status {max-width:25vw;font-size:10px}
+}
+
+.adachi-workspace .aw-canvas .dd-preview-progress {top:100px;bottom:auto;background:var(--aw-panel);color:var(--aw-text);border-color:var(--aw-line);}
+
+.adachi-workspace .aw-generation-progress {padding:9px 10px;margin:0 0 14px;border:1px solid var(--aw-line);border-radius:5px;background:var(--aw-input);color:var(--aw-text);font-size:12px;overflow-wrap:anywhere;}
+.adachi-workspace .aw-generation-progress progress {display:block;width:100%;height:7px;margin:6px 0 0;accent-color:var(--aw-accent);}
+.adachi-workspace .aw-generation-progress[data-state=error] {color:var(--aw-error);}
+.adachi-workspace .aw-generation-progress[data-state=complete] {color:var(--aw-ok);}
+
+.adachi-workspace .aw-generation-progress progress {appearance:none;border:0;border-radius:4px;overflow:hidden;background:var(--aw-line);}
+.adachi-workspace .aw-generation-progress progress::-webkit-progress-bar {background:var(--aw-line);border-radius:4px;}
+.adachi-workspace .aw-generation-progress progress::-webkit-progress-value {background:var(--aw-accent);border-radius:4px;}
+.adachi-workspace .aw-generation-progress progress::-moz-progress-bar {background:var(--aw-accent);border-radius:4px;}
+.adachi-workspace .aw-model-warning {padding:7px 9px;border:1px solid var(--aw-line);border-left:3px solid var(--aw-error);border-radius:4px;color:var(--aw-error);font-size:12px;}
+.adachi-workspace .aw-model-warning ul {margin:7px 0;padding-left:17px;overflow-wrap:anywhere;}
+.adachi-workspace .aw-model-warning li {margin:5px 0;}
+.adachi-workspace .aw-model-warning small {display:block;color:var(--aw-muted);}
+
+@keyframes aw-progress-wait {from{background-position:100% 0}to{background-position:-100% 0}}
+.adachi-workspace .aw-generation-progress progress:not([value]) {background:linear-gradient(90deg,var(--aw-line) 20%,var(--aw-accent) 50%,var(--aw-line) 80%);background-size:200% 100%;animation:aw-progress-wait 1.5s linear infinite;}
+.adachi-workspace .aw-generation-progress progress:not([value])::-webkit-progress-bar,.adachi-workspace .aw-generation-progress progress:not([value])::-webkit-progress-value {background:transparent;}
+@media(prefers-reduced-motion:reduce){.adachi-workspace .aw-generation-progress progress:not([value]){animation:none}}
+
+.adachi-workspace .aw-splitter {position:relative;cursor:col-resize;touch-action:none;outline:none;}
+.adachi-workspace .aw-splitter::before {content:"";position:absolute;top:12px;bottom:12px;left:4px;width:2px;background:var(--aw-line);border-radius:2px;}
+.adachi-workspace .aw-splitter:hover::before,.adachi-workspace .aw-splitter:focus-visible::before,.adachi-workspace[data-resizing] .aw-splitter::before {background:var(--aw-accent);}
+.adachi-workspace[data-resizing],.adachi-workspace[data-resizing] * {cursor:col-resize!important;user-select:none!important;}
+@media(max-width:850px){.adachi-workspace .aw-splitter{display:none}.adachi-workspace main{gap:8px}}
+
+.adachi-workspace .aw-connections {font-size:12px;border-bottom:1px solid var(--aw-line);padding:3px 0 8px;margin:0 0 10px;}
+.adachi-workspace .aw-connections summary {color:var(--aw-muted);}
+.adachi-workspace .aw-connections label {display:block;margin-top:8px;color:var(--aw-muted);}
+.adachi-workspace .aw-connections select {display:block;width:100%;margin-top:4px;}
+.adachi-workspace .aw-connections button {padding:4px 8px;margin-top:8px;}
+.adachi-workspace .aw-connections p {color:var(--aw-error);}
+.adachi-workspace .aw-bindings details small {display:block;overflow-wrap:anywhere;}

web/workspace_activation.js

@@ -0,0 +1,27 @@
+const virtualTypes=new Set(['GetNode','SetNode','Reroute','AdachiExclusiveSwitch']);
+function members(group,seen=new Set()) {
+    if(seen.has(group))return [];seen.add(group);group.recomputeInsideNodes?.();
+    return [...new Set(Array.from(group._children||group._nodes||group.nodes||[]).flatMap(child=>
+        typeof child.recomputeInsideNodes==='function'?members(child,seen):[child]))];
+}
+export function workspaceRoutingNodes(roots,controller) {
+    const allowed=new Set();
+    const plan=controller?.active&&controller.planStates?controller.planStates(controller.active).plan:null;
+    for(const root of roots)if(root.mode===4){
+        const groups=Array.from(root.graph?.groups||root.graph?._groups||[]).map(group=>{
+            const nodes=members(group);return {nodes,area:Math.abs((group.size?.[0]||1)*(group.size?.[1]||1))};
+        }).filter(group=>group.nodes.includes(root)).sort((a,b)=>a.area-b.area);
+        const group=groups[0];if(!group)continue;
+        const inherited=plan?.has(root)?plan.get(root)===4:group.nodes.filter(n=>!['AdachiGroupPresets','Note','MarkdownNote'].includes(n.type)).every(n=>n.mode===4);
+        if(inherited)for(const node of group.nodes)allowed.add(node);
+    }
+    const found=new Set(),seen=new Set();
+    const visit=node=>{
+        if(!node||seen.has(node))return;seen.add(node);
+        if(!roots.includes(node)&&!virtualTypes.has(node.type))return;
+        if(virtualTypes.has(node.type)&&node.mode===4&&allowed.has(node))found.add(node);
+        if(node.type==='GetNode'){const setter=node.findSetter?.(node.graph);if(setter)visit(setter);}
+        for(const input of node.inputs||[]){const g=node.graph,l=g?.getLink?.(input.link)??g?.links?.get?.(input.link)??g?.links?.[input.link];if(l)visit(g.getNodeById(l.origin_id));}
+    };
+    roots.forEach(visit);return found;
+}

web/workspace_bindings.js

@@ -0,0 +1,198 @@
+import {T} from './i18n.js';
+import {inspectExecutionFlow,executionNode,executionId,graphNodeEntries} from './workspace_flow.js';
+// Bind the workbench to existing graph widgets; no model or LoRA is loaded here.
+const widget = (node, name) => node?.widgets?.find(w => w.name === name);
+const nodeList = graph => graphNodeEntries(graph).map(entry=>entry.node);
+const ref = (node, slot = 0) => [String(node.id), slot];
+const valueList = w => {
+    const values = typeof w?.options?.values === 'function' ? w.options.values() : w?.options?.values;
+    return Array.isArray(values) ? values.map(String) : [];
+};
+const stateKey = 'adachi_workspace_bindings';
+const presetController=(nodes,sampler,graph)=>{
+    const saved=sampler.properties?.adachi_workspace_connections;
+    if(saved && Object.hasOwn(saved,'controllerId'))return nodes.find(n=>n.type==='AdachiGroupPresets'&&n.mode===0&&executionId(graph,n)===saved.controllerId)?.adachiGroupPresets?.model;
+    const active=nodes.filter(n=>n.type==='AdachiGroupPresets'&&n.mode===0),local=active.filter(n=>n.graph===sampler.graph),pool=local.length?local:active;
+    return pool.length===1?pool[0].adachiGroupPresets?.model:undefined;
+};
+
+export class WorkspaceBindings {
+    constructor({graph, sampler, canvas, onChange = () => {}, markChanged = () => {}}) {
+        Object.assign(this, {graph, sampler, canvas, onChange, markChanged});
+        this.nodes = nodeList(graph);
+        this.controller = presetController(this.nodes,sampler,graph);
+        this.owner = this.controller?.node || sampler;
+        this.owner.properties ??= {};
+        this.owner.properties[stateKey] ??= {profiles:{}};
+        this.catalog = {checkpoints:[], unets:[], ggufs:[], clipggufs:[], clips:[], vaes:[], loras:[]};
+        this.catalogReady = new Set();
+        this.profiles = this.controller?.data?.presets?.map(p => ({id:p.id, name:p.name, preset:p})) || [];
+        if (!this.profiles.length) this.profiles = [{id:'current',name:T("Current workflow")}];
+        this.profileId = this.controller?.data?.activeId || this.profiles[0].id;
+        this.resolve();
+    }
+    resolve() {
+        const profile=this.profiles.find(p=>p.id===this.profileId)||this.profiles[0];this.profileId=profile.id;
+        const all=this.owner.properties[stateKey].profiles;
+        if(!all[this.profileId])all[this.profileId]={loraMode:'inherit',loras:[],useRefiner:!!this.sampler.inputs?.find(i=>i.name==='refiner_model'&&i.link!=null)};
+        this.settings=all[this.profileId];this.settings.loras??=[];
+        // Retain explicit selections made in older versions without retaining
+        // any of the old template IDs or model-family name matching.
+        this.settings.loraMode??='override';
+        this.settings.textOverrides??={};
+        if(this.compiledProfileId===this.profileId)this.updateExecutionGraph(this.compiled,this.samplerId,false);
+        else {this.flow=null;this.model=this.clip=this.vae=this.positive=this.negative=this.lora=null;this.promptSources={positive:T("Reading workflow connections"),negative:T("Reading workflow connections")};}
+        return profile;
+    }
+    updateExecutionGraph(output,samplerId=executionId(this.graph,this.sampler),render=true) {
+        this.compiled=output;this.compiledProfileId=this.profileId;this.samplerId=String(samplerId);
+        this.flow=inspectExecutionFlow(output,this.samplerId,id=>executionNode(this.graph,id));
+        this.model=this.flow.modelNodes[0]?.node||null;this.clip=this.flow.clipNodes[0]?.node||null;this.vae=this.flow.vaeNodes[0]?.node||null;
+        this.positive=this.flow.encoders.positive.length===1?this.flow.encoders.positive[0].node:null;
+        this.negative=this.flow.encoders.negative.length===1?this.flow.encoders.negative[0].node:null;
+        this.lora=this.flow.loras[0]?.node||null;
+        this.promptSources={positive:T("Actual sampler connections"),negative:T("Actual sampler connections")};
+        const signature=JSON.stringify([this.profileId,this.flow.refs,[...this.flow.modelNodes,...this.flow.clipNodes,...this.flow.vaeNodes].map(e=>e.id),Object.entries(this.flow.encoders).map(([kind,items])=>[kind,items.map(e=>[e.id,e.fields])])]);
+        // Do not replace a focused textarea for ordinary background validation.
+        if(render && signature!==this.routingSignature){this.routingSignature=signature;this.render();}
+    }
+    changed() { this.markChanged(); this.onChange(this); }
+    modelDependencies() {
+        const specs={CheckpointLoader:['ckpt_name','checkpoints',T("Main model"),'checkpoints'],CheckpointLoaderSimple:['ckpt_name','checkpoints',T("Main model"),'checkpoints'],
+            UNETLoader:['unet_name','unets',T("Main model"),'diffusion_models'],
+            UNETLoaderGGUF:['unet_name','ggufs',T("Main model"),'diffusion_models'],
+            CLIPLoader:['clip_name','clips',T("Text encoder"),'text_encoders'],
+            CLIPLoaderGGUF:['clip_name','clipggufs',T("Text encoder"),'text_encoders'],
+            VAELoader:['vae_name','vaes','VAE','vae']};
+        const entries=this.flow?[...this.flow.modelNodes,...this.flow.clipNodes,...this.flow.vaeNodes]:[];
+        const unique=entries.filter((e,i,all)=>all.findIndex(x=>x.id===e.id)===i);
+        return unique.flatMap(entry=>{
+            const node=entry.node;
+            if(['DualCLIPLoader','TripleCLIPLoader','DualCLIPLoaderGGUF'].includes(node?.type))return [1,2,3].flatMap(index=>{const w=widget(node,'clip_name'+index);return w?[{...entry,widget:w,key:node.type.endsWith('GGUF')?'clipggufs':'clips',label:T("Text encoder {index}",{index}),folder:'text_encoders'}]:[];});
+            const spec=specs[node?.type];if(!spec)return [];
+            const [name,key,label,folder]=spec,w=widget(node,name);
+            if(w)return [{...entry,widget:w,key,label,folder}];
+            return [];
+        });
+    }
+    missingDependencies() {
+        return this.modelDependencies().filter(d=>this.catalogReady.has(d.key) && d.widget.value && !this.catalog[d.key].includes(String(d.widget.value)));
+    }
+    refreshProfiles() {
+        this.nodes=nodeList(this.graph);
+        const controller = presetController(this.nodes,this.sampler,this.graph);
+        const signature = JSON.stringify([controller?.node?.id, controller?.data?.activeId, controller?.data?.presets?.map(p=>[p.id,p.name])]);
+        if(signature===this.presetSignature)return;
+        this.presetSignature=signature;
+        this.controller=controller;
+        this.owner=controller?.node || this.sampler;
+        this.owner.properties ??= {};
+        this.owner.properties[stateKey] ??= {profiles:{}};
+        this.profiles=controller?.data?.presets?.map(p=>({id:p.id,name:p.name,preset:p})) || [];
+        if(!this.profiles.length)this.profiles=[{id:'current',name:T("Current workflow")}];
+        this.profileId=controller?.data?.activeId || (this.profiles.some(p=>p.id===this.profileId)?this.profileId:this.profiles[0].id);
+        this.resolve();this.render();this.onChange(this);
+    }
+    selectProfile(id) {
+        if (!this.profiles.some(p=>p.id===id)) throw Error(T("The group preset no longer exists"));
+        this.profileId=id;
+        this.controller?.apply(id);
+        this.resolve(); this.render(); this.changed();
+    }
+    getExecutionBinding() {
+        const binding={samplerInputs:{},inputOverrides:{},removeInputs:{},
+            loraMode:this.settings.loraMode,loras:this.settings.loras,clipConsumers:this.flow?.clipConsumers||[]};
+        if(!this.flow)throw Error(T("Reading workflow connections; please wait"));
+        // Final MODEL and CONDITIONING refs remain untouched, including custom
+        // model patches, ControlNet and combinations of multiple encoders.
+        for(const [kind,entries] of Object.entries(this.flow.encoders))for(const entry of entries)for(const field of entry.fields){
+            const target=this.settings.textOverrides[entry.id]||{};
+            const legacy=entries.length===1 && field==='text' && Object.hasOwn(this.settings,kind);
+            if(Object.hasOwn(target,field)||legacy){binding.inputOverrides[entry.id]??={};binding.inputOverrides[entry.id][field]=Object.hasOwn(target,field)?target[field]:this.settings[kind];}
+        }
+        if(!this.settings.useRefiner || Number(widget(this.sampler,'refiner_switch')?.value??1)>=1)binding.removeInputs[this.samplerId]=['refiner_model','refiner_positive','refiner_negative'];
+        return binding;
+    }
+    async loadCatalog(fetchApi) {
+        this.catalogController?.abort();const controller=this.catalogController=new AbortController(),timer=setTimeout(()=>controller.abort(),15000);
+        const get = async path => {const r=await fetchApi(path,{signal:controller.signal});if(!r.ok)throw Error(T("Model list request failed ({status})",{status:r.status}));return r.json();};
+        const sources=[['checkpoints','/object_info/CheckpointLoaderSimple','CheckpointLoaderSimple','ckpt_name'],
+            ['unets','/object_info/UNETLoader','UNETLoader','unet_name'],
+            ['clips','/object_info/CLIPLoader','CLIPLoader','clip_name'],
+            ['vaes','/object_info/VAELoader','VAELoader','vae_name'],['loras','/models/loras']];
+        let results;try{results=await Promise.allSettled(sources.map(([,path])=>get(path)));}finally{clearTimeout(timer);}
+        if(this.destroyed || this.catalogController!==controller)return;
+        results.forEach((result,i)=>{if(result.status!=='fulfilled')return;const [key,,type,input]=sources[i];
+            const values=type?result.value?.[type]?.input?.required?.[input]?.[0]:result.value;
+            if(Array.isArray(values)){this.catalog[key]=values.map(String);this.catalogReady.add(key);}
+        });
+        this.catalogError=results.some(r=>r.status==='rejected')?T("Some model lists are unavailable; existing workflow selections remain usable"):'';
+        this.render();
+    }
+    createPanel() { this.root=document.createElement('section');this.root.className='aw-bindings';this.render();return this.root; }
+    render() {
+        if(!this.root || this.destroyed)return;
+        const doc=this.root.ownerDocument;
+        const el=(tag,text)=>{const e=doc.createElement(tag);if(text)e.textContent=text;return e;};
+        const row=(label,control)=>{const l=el('label',label);l.append(control);return l;};
+        const select=(values,current,change)=>{const s=el('select');for(const [value,label] of values){const o=el('option',label);o.value=value;s.append(o);}s.value=current;s.onchange=()=>change(s.value);return s;};
+        this.root.replaceChildren();
+        const presets=el('div');presets.className='aw-presets';
+        const controllerActive = this.controller && this.controller.node.mode === 0;
+        if (!controllerActive) {const multiple=this.nodes.filter(n=>n.type==='AdachiGroupPresets'&&n.mode===0).length>1;const warning=el('p',multiple?T("Multiple group preset controllers found; choose one in Workflow connections."):T("No enabled Group Bypass Presets node; using the current workflow settings."));warning.className='aw-profile-warning';warning.setAttribute('role','status');this.root.append(warning);}
+        for(const p of controllerActive ? this.profiles : []){const b=el('button',p.name);b.type='button';b.setAttribute('aria-pressed',String(p.id===this.profileId));b.onclick=()=>this.selectProfile(p.id);presets.append(b);}
+        this.root.append(presets);
+        const missing=this.missingDependencies();
+        if(missing.length){const warning=el('details'),head=el('summary',T("Missing {count} model files",{count:missing.length}));warning.className='aw-model-warning';warning.title=missing.map(d=>T("{label}: {value}",{label:d.label,value:d.widget.value})).join('\n');warning.append(head);
+            const list=el('ul');for(const d of missing){const item=el('li',T("{label}: {value}",{label:d.label,value:d.widget.value}));item.append(el('small',`ComfyUI/models/${d.folder}/`));list.append(item);}warning.append(list);this.root.append(warning);}
+        const details=el('details'),summary=el('summary',T("Main model and LoRA"));details.append(summary);details.open=!!this.detailsOpen;details.ontoggle=()=>{this.detailsOpen=details.open;};
+        const dependencies=this.modelDependencies();
+        for(const d of dependencies){const w=d.widget,current=String(w.value||'');const values=[...new Set([...valueList(w),...this.catalog[d.key],current])].filter(Boolean);
+            const label=dependencies.filter(x=>x.label===d.label).length>1?`${d.label} · ${d.node.title||d.node.type} #${d.id}`:d.label;const field=row(label,select(values.map(v=>[v,v]),current,v=>{w.value=v;w.callback?.(v,this.canvas,d.node);this.render();this.changed();}));field.title=`ComfyUI/models/${d.folder}/`;details.append(field);}
+        if(!dependencies.some(d=>d.node===this.model))details.append(el('p',T("This profile uses the model from the workflow")));
+        const loraTitle=el('div','LoRA');details.append(loraTitle);
+        const inherit=el('input');inherit.type='checkbox';inherit.checked=this.settings.loraMode==='inherit';inherit.onchange=()=>{this.settings.loraMode=inherit.checked?'inherit':'override';this.render();this.changed();};const inheritRow=row(T("Inherit workflow LoRAs"),inherit);inheritRow.title=T("Keep existing LoRA nodes and both CLIP branches; uncheck to use the custom list below. An empty custom list disables recognized LoRAs.");details.append(inheritRow);
+        if(inherit.checked){
+            const active=[];for(const entry of this.flow?.loras||[]){const input=entry.api.inputs||{};if(entry.api.class_type==='Power Lora Loader (rgthree)'){for(const value of Object.values(input))if(value?.on && value.lora)active.push(`${value.lora} · ${value.strength??1}`);}else if(input.lora_name && (Number(input.strength_model)||Number(input.strength_clip)))active.push(`${input.lora_name} · ${input.strength_model??1}`);}
+            const info=el('small',active.length?active.join(' / '):T("No active LoRAs in the original workflow"));info.title=active.join('\n');details.append(info);
+        }
+        const customLoras=el('div');customLoras.hidden=inherit.checked;details.append(customLoras);
+
+        if(!this.settings.loras.length)customLoras.append(el('small',T("No LoRA")));
+        this.settings.loras.forEach((l,index)=>{
+            const line=el('div');line.className='aw-lora-row';
+            const names=[...new Set([...this.catalog.loras,l.name].filter(Boolean))];
+            const pick=select([['',T("None")],...names.map(n=>[n,n])],l.name||'',v=>{l.name=v;this.changed();});pick.setAttribute('aria-label',T("LoRA {index}",{index:index+1}));
+            const strength=el('input');strength.type='number';strength.min='-10';strength.max='10';strength.step='0.05';strength.value=l.strength??1;strength.setAttribute('aria-label',T("LoRA {index} strength",{index:index+1}));strength.onchange=()=>{const n=Number(strength.value);if(Number.isFinite(n)){l.strength=Math.max(-10,Math.min(10,n));strength.value=l.strength;this.changed();}};
+            const remove=el('button','×');remove.type='button';remove.title=T("Remove this LoRA");remove.onclick=()=>{this.settings.loras.splice(index,1);this.render();this.changed();};
+            line.append(pick,strength,remove);customLoras.append(line);
+        });
+        const add=el('button',T("＋ Add LoRA"));add.type='button';add.disabled=!this.flow?.refs.model;add.onclick=()=>{this.settings.loras.push({name:'',strength:1});this.render();this.changed();};customLoras.append(add);
+        const refiner=el('input');refiner.type='checkbox';refiner.checked=!!this.settings.useRefiner;
+        refiner.onchange=()=>{this.settings.useRefiner=refiner.checked;this.changed();};
+        details.append(row(T("Use the workflow Refiner (must be compatible with the main model)"),refiner));
+        if(this.catalogError)details.append(el('small',this.catalogError));
+        this.root.append(details);
+        for(const [kind,label] of [['positive',T("Positive prompt")],['negative',T("Negative prompt")]]) {
+            const entries=this.flow?.encoders[kind]||[];
+            if(!entries.length){const info=el('p',label+": "+(this.flow?T("Controlled by the original workflow"):T("Reading connections…")));info.title=T("Encoders are identified from the execution graph; unrecognized conditioning stays controlled by the original workflow");this.root.append(info);continue;}
+            const container=entries.length>1?el('details'):this.root;
+            if(entries.length>1){container.append(el('summary',T("{label} · {count} encoders",{label,count:entries.length})));this.root.append(container);}
+            for(const entry of entries)for(const name of entry.fields){
+                const w=widget(entry.node,name),linked=Array.isArray(entry.api.inputs?.[name]);
+                const saved=this.settings.textOverrides[entry.id]||{},legacy=entries.length===1&&name==='text'?this.settings[kind]:undefined;
+                const text=el('textarea');text.rows=kind==='positive'?6:3;
+                text.value=saved[name]??legacy??(linked?'':w?.value??entry.api.inputs?.[name]??'');
+                text.placeholder=linked?T("Inherit upstream dynamic text; type to override"):'';
+                text.dataset.executionId=entry.id;text.dataset.widget=name;
+                text.oninput=()=>{this.settings.textOverrides[entry.id]??={};this.settings.textOverrides[entry.id][name]=text.value;this.changed();};
+                const title=entries.length>1||entry.fields.length>1?`${label} · ${entry.node?.title||entry.api.class_type} · ${name}`:label;
+                text.setAttribute('aria-label',title);const field=row(title,text);
+                field.title=T("Actual sampler connections · {node} #{id} · {field}. {description}",{node:entry.node?.title||entry.api.class_type,id:entry.id,field:name,description:linked?T("Text comes from upstream; typing creates an override."):T("Changes affect only this encoder; downstream conditioning is preserved.")});
+                const reset=el('button','↶');reset.type='button';reset.title=T("Restore {label} from the workflow",{label:title});reset.onclick=()=>{delete this.settings.textOverrides[entry.id]?.[name];if(entries.length===1&&name==='text')delete this.settings[kind];this.render();this.changed();};field.append(reset);container.append(field);
+            }
+        }
+
+    }
+    destroy() {this.destroyed=true;this.catalogController?.abort();this.root?.remove();}
+}

web/workspace_execution.js

@@ -0,0 +1,48 @@
+import {T} from './i18n.js';
+import {app} from '../../scripts/app.js';
+import {api} from '../../scripts/api.js';
+import {workspaceRequest} from './workspace_transport.js';
+import {workspaceRoutingNodes} from './workspace_activation.js';
+import {applyExecutionBinding,graphNodeEntries,executionId,executionNode} from './workspace_flow.js';
+const graphTasks=new WeakMap();
+const link=value=>Array.isArray(value)&&value.length===2&&typeof value[0]==='string'&&Number.isInteger(value[1]);
+function depends(prompt,ref,target,seen=new Set()) {
+ if(!link(ref)||seen.has(ref[0]))return false;if(ref[0]===target)return true;seen.add(ref[0]);
+ return Object.values(prompt[ref[0]]?.inputs||{}).some(input=>depends(prompt,input,target,seen));
+}
+export function prepareWorkspacePrompt(options) {
+ const graph=app.rootGraph||app.graph;
+ const previous=graphTasks.get(graph)||Promise.resolve();
+ const task=previous.catch(()=>{}).then(()=>prepareSnapshot(graph,options));graphTasks.set(graph,task);return task;
+}
+async function prepareSnapshot(graph,{editorNode,samplerNode,compareNode,binding={},getBinding,allowEmpty=false,controller}) {
+ if(!editorNode?.adachiEditor?.ready && !allowEmpty)throw Error(T("Upload a source image first"));
+ const entries=graphNodeEntries(graph),roots=[editorNode,samplerNode,compareNode];
+ if(graph!==(app.rootGraph||app.graph)||roots.some(n=>!entries.some(e=>e.node===n)))throw Error(T("The workflow changed; reopen the workbench"));
+ const ids=roots.map(n=>executionId(graph,n)),[editorId,samplerId,compareId]=ids;
+ const parents=roots.flatMap(node=>{const path=executionId(graph,node).split(':');return path.slice(0,-1).map((_,i)=>executionNode(graph,path.slice(0,i+1).join(':'))).filter(Boolean);});
+ const routers=workspaceRoutingNodes(roots,controller),modes=new Map([...parents,...roots,...routers].map(n=>[n,n.mode]));
+ if(editorNode.adachiEditor.ready)editorNode.adachiEditor.flush();let prepared;
+ try {
+  for(const node of modes.keys())node.mode=0;
+  prepared=typeof app.ue_modified_prompt==='function'?await app.ue_modified_prompt():await app.graphToPrompt();
+  if(!prepared?.output)throw Error(T("Workflow connections could not be resolved; check bypass states, Get/Set and broadcast connections"));
+ } finally {for(const [node,mode]of modes)node.mode=mode;}
+ if(graph!==(app.rootGraph||app.graph))throw Error(T("The workflow changed; check connections again"));
+ const output=prepared.output,sampler=output[samplerId],compare=output[compareId];
+ if(!sampler||!compare||!output[editorId])throw Error(T("The selected image, inpaint or comparison node is missing from the execution graph"));
+ // Existing image processors and result postprocessing remain connected. An
+ // unconnected workbench can use its explicitly selected inline source/output.
+ if(!depends(output,sampler.inputs.image,editorId)){sampler.inputs.image=[editorId,0];sampler.inputs.mask=[editorId,1];}
+ else if(!link(sampler.inputs.mask))sampler.inputs.mask=[editorId,1];
+ if(!link(compare.inputs.original))compare.inputs.original=[editorId,0];
+ if(!depends(output,compare.inputs.result,samplerId))compare.inputs.result=[samplerId,0];
+ const resolved=getBinding?await getBinding(output,samplerId):binding;
+ const prompt=applyExecutionBinding(output,samplerId,resolved);
+ // Keep the editable workflow, including ignored groups and original routers,
+ // in PNG metadata. Only the separate API execution graph is specialized.
+ const workflow=JSON.parse(JSON.stringify(graph.serialize()));
+ return {prompt,compare_id:compareId,client_id:api.clientId||'',extra_data:{extra_pnginfo:{workflow},auth_token_comfy_org:api.authToken,api_key_comfy_org:api.apiKey}};
+}
+export const validateWorkspace=(prepared,options)=>workspaceRequest(api,'validate',prepared,options);
+export const submitWorkspace=prepared=>workspaceRequest(api,'submit',prepared);

web/workspace_flow.js

@@ -0,0 +1,133 @@
+import {T} from './i18n.js';
+const link=value=>Array.isArray(value) && value.length===2 && ['string','number'].includes(typeof value[0]) && Number.isInteger(value[1]);
+const asRef=value=>link(value)?[String(value[0]),value[1]]:null;
+const loraTypes=new Set(['LoraLoader','LoraLoaderModelOnly','Power Lora Loader (rgthree)']);
+const staticRouter=api=>api.class_type==='AdachiExclusiveSwitch'||(['ComfySwitchNode','ComfySoftSwitchNode'].includes(api.class_type)&&typeof api.inputs?.switch==='boolean');
+const loaderTypes={MODEL:new Set(['CheckpointLoaderSimple','CheckpointLoader','UNETLoader','UNETLoaderGGUF','CheckpointLoaderNF4']),
+    CLIP:new Set(['CheckpointLoaderSimple','CheckpointLoader','CLIPLoader','DualCLIPLoader','TripleCLIPLoader','CLIPLoaderGGUF','DualCLIPLoaderGGUF']),
+    VAE:new Set(['CheckpointLoaderSimple','CheckpointLoader','VAELoader'])};
+
+function sourceInputs(api) {
+    if(['ComfySwitchNode','ComfySoftSwitchNode'].includes(api.class_type)&&typeof api.inputs?.switch==='boolean'){
+        const inputs=api.inputs;if(api.class_type==='ComfySoftSwitchNode'){
+            if(!Object.hasOwn(inputs,'on_false'))return [inputs.on_true];if(!Object.hasOwn(inputs,'on_true'))return [inputs.on_false];
+        }
+        return [inputs[inputs.switch?'on_true':'on_false']];
+    }
+    if(api.class_type==='AdachiExclusiveSwitch') {
+        const mask=Number(api.inputs?.active_inputs);
+        if(mask===1)return [api.inputs.input_a];if(mask===2)return [api.inputs.input_b];
+    }
+    return Object.values(api.inputs||{});
+}
+function ancestors(prompt,refs) {
+    const found=new Set(),queue=refs.map(asRef).filter(Boolean);
+    while(queue.length){const [id]=queue.pop();if(found.has(id)||!prompt[id])continue;found.add(id);queue.push(...sourceInputs(prompt[id]).map(asRef).filter(Boolean));}
+    return found;
+}
+function origins(prompt,start,type,lookup) {
+    const found=[],seen=new Set(),queue=[asRef(start)].filter(Boolean);
+    while(queue.length){const [id,slot]=queue.shift(),key=`${id}:${slot}`;if(seen.has(key))continue;seen.add(key);
+        const api=prompt[id],node=lookup(id);if(!api)continue;
+        if(loaderTypes[type].has(api.class_type)){found.push({id,node,api});continue;}
+        if(staticRouter(api)){queue.push(...sourceInputs(api).map(asRef).filter(Boolean));continue;}
+        if(loraTypes.has(api.class_type)){const ref=asRef(api.inputs?.[type==='MODEL'?'model':type==='CLIP'?'clip':'vae']);if(ref)queue.push(ref);continue;}
+        const typed=(node?.inputs||[]).filter(i=>i.type===type).map(i=>i.name);
+        const names=typed.length?typed:Object.keys(api.inputs||{}).filter(name=>new RegExp(`^${type.toLowerCase()}(?:$|[_\\d])`,'i').test(name));
+        queue.push(...names.map(name=>asRef(api.inputs[name])).filter(Boolean));
+    }
+    return found.filter((item,i,all)=>all.findIndex(x=>x.id===item.id)===i);
+}
+function encoders(prompt,start,lookup) {
+    const found=[],seen=new Set(),queue=[asRef(start)].filter(Boolean);
+    while(queue.length){const [id,slot]=queue.shift(),key=`${id}:${slot}`;if(seen.has(key))continue;seen.add(key);
+        const api=prompt[id],node=lookup(id);if(!api)continue;
+        const fields=(node?.widgets||[]).filter(w=>/^(text(?:_[gl])?|prompt)$/i.test(w.name) && typeof w.value==='string').map(w=>w.name);
+        const output=node?.outputs?.[slot];
+        if(fields.length && (output?.type==='CONDITIONING'||/^CLIPTextEncode/.test(api.class_type))){found.push({id,node,api,fields});continue;}
+        if(staticRouter(api)){queue.push(...sourceInputs(api).map(asRef).filter(Boolean));continue;}
+        const selected=output?.name && /^(positive|negative)$/i.test(output.name) && asRef(api.inputs?.[output.name]);
+        if(selected){queue.push(selected);continue;}
+        const names=(node?.inputs||[]).filter(i=>i.type==='CONDITIONING').map(i=>i.name);
+        const fallback=Object.keys(api.inputs||{}).filter(name=>/^(positive|negative|conditioning(?:_\d+)?)$/i.test(name));
+        queue.push(...(names.length?names:fallback).map(name=>asRef(api.inputs[name])).filter(Boolean));
+    }
+    return found.filter((item,i,all)=>all.findIndex(x=>x.id===item.id)===i);
+}
+
+export function inspectExecutionFlow(prompt,samplerId,lookup=()=>null) {
+    const api=prompt[String(samplerId)],refs={};
+    for(const name of ['model','positive','negative','vae'])refs[name]=asRef(api?.inputs?.[name]);
+    const positive=encoders(prompt,refs.positive,lookup),negative=encoders(prompt,refs.negative,lookup);
+    const allEncoders=[...positive,...negative].filter((item,i,all)=>all.findIndex(x=>x.id===item.id)===i);
+    const consumers=allEncoders.filter(e=>asRef(e.api.inputs?.clip)).map(e=>({id:e.id,input:'clip'}));
+    const clipNodes=consumers.flatMap(e=>origins(prompt,prompt[e.id].inputs[e.input],'CLIP',lookup)).filter((e,i,all)=>all.findIndex(x=>x.id===e.id)===i);
+    return {refs,modelNodes:origins(prompt,refs.model,'MODEL',lookup),clipNodes,
+        vaeNodes:origins(prompt,refs.vae,'VAE',lookup),encoders:{positive,negative},clipConsumers:consumers,
+        loras:[...ancestors(prompt,Object.values(refs))].filter(id=>loraTypes.has(prompt[id].class_type)).map(id=>({id,node:lookup(id),api:prompt[id]}))};
+}
+
+export function executionNode(graph,id) {
+    const path=String(id).split(':');let node;
+    for(let i=0;i<path.length;i++){node=graph?.getNodeById(path[i]);if(!node && /^-?\d+$/.test(path[i]))node=graph?.getNodeById(Number(path[i]));if(!node)return null;if(i<path.length-1)graph=node.subgraph;}
+    return node;
+}
+
+export function graphNodeEntries(graph,prefix=[],seen=new Set()) {
+    if(!graph || seen.has(graph))return [];seen.add(graph);
+    return Array.from(graph._nodes||[]).flatMap(node=>{
+        const path=[...prefix,String(node.id)];
+        return [{id:path.join(':'),node,graph},...graphNodeEntries(node.subgraph,path,seen)];
+    });
+}
+export const executionId=(graph,node)=>graphNodeEntries(graph).find(entry=>entry.node===node)?.id;
+
+// Rewrite only the chosen sampler's ancestry. Other output branches keep their
+// graph definitions; the backend subsequently queues only the comparison target.
+export function applyExecutionBinding(original,samplerId,binding={}) {
+    // Native widget values can be Vue proxies (for example rgthree LoRA
+    // rows). Copy using the same JSON boundary as ComfyUI's queue request.
+    const prompt=JSON.parse(JSON.stringify(original)),id=String(samplerId),sampler=prompt[id];
+    if(!sampler)throw Error(T("The selected inpaint node is missing from the execution graph"));
+    Object.assign(sampler.inputs,binding.samplerInputs||{});
+    for(const [key,inputs] of Object.entries(binding.inputOverrides||{})){
+        if(!prompt[key])throw Error(T("The selected parameter node is missing from the execution graph #{id}",{id:key}));Object.assign(prompt[key].inputs,inputs);
+    }
+    for(const [key,names] of Object.entries(binding.removeInputs||{}))for(const name of names)if(prompt[key])delete prompt[key].inputs[name];
+    const scope=()=>new Set([id,...ancestors(prompt,Object.values(sampler.inputs))]);
+    const collapse=(candidates,choose)=>{
+        const resolve=(value,seen=new Set())=>{
+            const ref=asRef(value);if(!ref || !candidates.has(ref[0]))return ref;
+            if(seen.has(ref[0]))return null;seen=new Set(seen);seen.add(ref[0]);
+            const next=choose(prompt[ref[0]],ref[1]);return next?resolve(next,seen):null;
+        };
+        for(const key of scope())if(!candidates.has(key))for(const [name,value] of Object.entries(prompt[key]?.inputs||{})){
+            const ref=asRef(value);if(ref && candidates.has(ref[0])){const result=resolve(ref);if(result && prompt[result[0]])prompt[key].inputs[name]=result;}
+        }
+    };
+    const switches=new Set([...scope()].filter(key=>prompt[key]?.class_type==='AdachiExclusiveSwitch' && [1,2].includes(Number(prompt[key].inputs.active_inputs))));
+    collapse(switches,(api,slot)=>slot===0?api.inputs[Number(api.inputs.active_inputs)===1?'input_a':'input_b']:null);
+    if(binding.loraMode==='override') {
+        const loras=new Set([...scope()].filter(key=>loraTypes.has(prompt[key]?.class_type)));
+        collapse(loras,(api,slot)=>slot===0?api.inputs.model:slot===1?api.inputs.clip:null);
+        const selected=(binding.loras||[]).filter(l=>l.name && l.on!==false);
+        if(selected.length){
+            const baseModel=asRef(sampler.inputs.model);if(!baseModel)throw Error(T("Connect a main model before adding LoRAs"));
+            let nextId=-1000000;
+            const fresh=()=>{while(prompt[String(nextId)])nextId--;return String(nextId--);};
+            const consumers=(binding.clipConsumers||[]).filter(c=>asRef(prompt[c.id]?.inputs?.[c.input]));
+            const unique=new Map();for(const c of consumers){const ref=asRef(prompt[c.id].inputs[c.input]),key=JSON.stringify(ref);if(!unique.has(key))unique.set(key,{ref,consumers:[]});unique.get(key).consumers.push(c);}
+            const chains=unique.size?[...unique.values()]:[{ref:null,consumers:[]}];
+            chains.forEach((chain,index)=>{
+                let model=baseModel,clip=chain.ref;
+                for(const l of selected){const key=fresh(),inputs={model,lora_name:l.name,strength_model:Number(l.strength??1)};
+                    if(clip)Object.assign(inputs,{clip,strength_clip:Number(l.clipStrength??l.strength??1)});
+                    prompt[key]={class_type:clip?'LoraLoader':'LoraLoaderModelOnly',inputs};model=[key,0];if(clip)clip=[key,1];
+                }
+                if(index===0)sampler.inputs.model=model;
+                for(const c of chain.consumers)prompt[c.id].inputs[c.input]=clip;
+            });
+        }
+    }
+    return prompt;
+}

web/workspace_history.js

@@ -0,0 +1,164 @@
+import {T} from './i18n.js';
+// Session-only source checkpoints for the full-screen workspace. Capturing and
+// restoring never uploads a new image or serializes the rest of the workflow.
+const WIDGETS = ["image", "mask_data", "source_data", "edit_mode", "crop_state", "blend_data", "layer_image"];
+const copyCommands = commands => (commands || []).map(command => ({ ...command,
+    ...(command.points ? {points: command.points.map(point => ({...point}))} : {}),
+}));
+
+function contextSnapshot(context) {
+    return {
+        mask: context.mask.toDataURL("image/png"),
+        baseMask: context.baseMask.toDataURL("image/png"),
+        history: copyCommands(context.history), future: copyCommands(context.future),
+        lastMask: context.lastMask || "", dirty: !!context.dirty,
+    };
+}
+
+async function decodeCanvas(data) {
+    const image = new Image(); image.src = data;
+    await image.decode();
+    const canvas = document.createElement("canvas");
+    canvas.width = image.naturalWidth || image.width; canvas.height = image.naturalHeight || image.height;
+    canvas.getContext("2d").drawImage(image, 0, 0);
+    return canvas;
+}
+
+export function captureSourceState(node) {
+    const editor = node.adachiEditor, crop = node.adachiCrop;
+    if (!editor?.ready || !editor.image || editor.destroyed) throw Error(T("Load a source image first"));
+    if (editor.stroke || editor.pointerId != null || crop?.busy) throw Error(T("Finish the current edit first"));
+    editor.flush();
+    const target = crop?.target || "mask", contexts = {};
+    for (const [name, context] of Object.entries(crop?.contexts || {})) {
+        if (name !== target && context?.mask && context?.baseMask) contexts[name] = contextSnapshot(context);
+    }
+    contexts[target] = contextSnapshot(editor);
+    const state = {
+        widgets: Object.fromEntries(WIDGETS.map(name => [name, node.widgets?.find(widget => widget.name === name)?.value])),
+        image: editor.image, layer: crop?.layer || null, contexts, target,
+        cropWidth: crop?.cropWidth, cropHeight: crop?.cropHeight,
+        filename: editor.filename?.textContent || "", embed: !!editor.embed?.checked,
+        view: editor.view, zoom: editor.zoom, pan: {...editor.pan}, keepFit: !!editor.keepFit,
+    };
+    state.key = JSON.stringify([state.widgets, state.image.src, state.layer?.src, state.contexts, state.target]);
+    return state;
+}
+
+function releaseContexts(contexts, active) {
+    const canvases = new Set();
+    for (const context of Object.values(contexts || {})) for (const name of ["mask", "baseMask"]) {
+        if (context?.[name] && context[name] !== active?.mask && context[name] !== active?.baseMask) canvases.add(context[name]);
+    }
+    for (const canvas of canvases) canvas.width = canvas.height = 0;
+}
+
+export async function restoreSourceState(node, state, {decode = decodeCanvas, onRestore} = {}) {
+    const editor = node.adachiEditor, crop = node.adachiCrop;
+    if (!editor || editor.destroyed || !node.graph) throw Error(T("The source image node was removed"));
+    if (editor.stroke || editor.pointerId != null || crop?.busy) throw Error(T("Finish the current edit first"));
+    const loadVersion = editor.loadVersion, maskRevision = editor.maskRevision, cropRevision=crop?.revision;
+    const contexts = {};
+    try {
+        // Decode all masks before altering live state. A failed decode leaves
+        // the working image, mask and crop completely untouched.
+        for (const [name, context] of Object.entries(state.contexts)) {
+            const decoded=await Promise.allSettled([decode(context.mask), decode(context.baseMask)]);
+            const failed=decoded.find(item=>item.status==='rejected');
+            if(failed){for(const item of decoded)if(item.status==='fulfilled')item.value.width=item.value.height=0;throw failed.reason;}
+            const [mask,baseMask]=decoded.map(item=>item.value);
+            contexts[name] = {...context, mask, baseMask, history:copyCommands(context.history), future:copyCommands(context.future)};
+        }
+        if (editor.destroyed || !node.graph || editor.loadVersion !== loadVersion || editor.maskRevision !== maskRevision || crop?.revision !== cropRevision || editor.stroke || editor.pointerId != null || crop?.busy) {
+            throw Error(T("The image changed during restoration; try again"));
+        }
+    } catch (error) { releaseContexts(contexts); throw error; }
+    const active = contexts[state.target];
+    if (!active) { releaseContexts(contexts); throw Error(T("The saved mask state is invalid")); }
+    const oldContexts = {...(crop?.contexts || {}), live: {mask:editor.mask, baseMask:editor.baseMask}};
+    crop?.stopOriginalPreview?.();
+    for (const [name, value] of Object.entries(state.widgets)) {
+        const widget = node.widgets?.find(widget => widget.name === name);
+        if (widget && value !== undefined) widget.value = value;
+    }
+    editor.loadVersion++;
+    editor.image = state.image;
+    Object.assign(editor, active);
+    editor.view = state.view; editor.zoom = state.zoom; editor.pan = {...state.pan};
+    editor.keepFit = state.keepFit; editor.ready = true;
+    editor.stroke = null; editor.pointerId = null; editor.drag = null; editor.cursor = null;
+    editor.sceneKey = null; editor.maskRevision++;
+    if (editor.filename) { editor.filename.textContent = state.filename; editor.filename.title = state.filename; }
+    if (editor.sizeLabel) editor.sizeLabel.textContent = `${state.image.width} × ${state.image.height}`;
+    if (editor.embed) editor.embed.checked = state.embed;
+    node.properties ??= {}; node.properties.adachi_keep_fit = state.keepFit;
+    if (editor.empty) editor.empty.hidden = true;
+    if (crop) {
+        clearTimeout(crop.resultTimer); crop.pendingResult = null; crop.acceptVersion++;
+        crop.contexts = contexts; crop.target = state.target; crop.layer = state.layer;
+        crop.cropWidth = state.cropWidth; crop.cropHeight = state.cropHeight;
+        crop.hover = null; crop.revision++; crop.patchKey = null;
+        crop.syncPreset?.(); crop.sync?.();
+    }
+    releaseContexts(oldContexts, active);
+    editor.updateButtons?.(); editor.schedule?.();
+    node.graph.setDirtyCanvas?.(true, true);
+    await onRestore?.(state);
+    return state;
+}
+
+function estimateBytes(entries) {
+    const images = new Set(); let bytes = 0;
+    for (const {state} of entries) {
+        bytes += state.key.length * 2;
+        for (const image of [state.image, state.layer]) if (image && !images.has(image)) {
+            images.add(image); bytes += image.width * image.height * 4;
+        }
+    }
+    return bytes;
+}
+
+export class WorkspaceHistory {
+    constructor(node, {maxEntries = 12, maxBytes = 256 * 1024 * 1024, onChange, onRestore} = {}) {
+        Object.assign(this, {node, maxEntries:Math.max(2,maxEntries), maxBytes, onChange, onRestore});
+        this.entries = []; this.index = -1; this.busy = false;
+    }
+    get current() { return this.entries[this.index] || null; }
+    get canUndo() { return this.index > 0 && !this.busy; }
+    checkpoint(label = T("Source image")) {
+        const state = captureSourceState(this.node);
+        if (this.current?.state.key === state.key) return this.current;
+        this.entries.splice(this.index + 1);
+        const entry = {id:crypto.randomUUID(), label:String(label), state};
+        this.entries.push(entry); this.index = this.entries.length - 1;
+        // Keep the original and current source reachable; discard intermediate
+        // checkpoints first when the session history reaches its budget.
+        while (this.entries.length > 2 && (this.entries.length > this.maxEntries || estimateBytes(this.entries) > this.maxBytes)) {
+            this.entries.splice(1,1); this.index--;
+        }
+        this.onChange?.(this); return entry;
+    }
+    async transact(label, operation) {
+        if (this.busy) throw Error(T("Processing the source image; please wait"));
+        this.busy = true; this.onChange?.(this);
+        try {
+            this.checkpoint(this.entries.length ? T("Edited source") : T("Initial source"));
+            const result = await operation();
+            this.checkpoint(label);
+            return result;
+        } finally { this.busy = false; this.onChange?.(this); }
+    }
+    async restore(index) {
+        if (this.busy) throw Error(T("Processing the source image; please wait"));
+        if (!Number.isInteger(index) || !this.entries[index]) throw Error(T("The source image history entry no longer exists"));
+        this.busy = true; this.onChange?.(this);
+        try {
+            await restoreSourceState(this.node, this.entries[index].state, {onRestore:this.onRestore});
+            this.index = index;
+            return this.current;
+        } finally { this.busy = false; this.onChange?.(this); }
+    }
+    undo() { return this.restore(this.index - 1); }
+    original() { return this.restore(0); }
+    clear() { this.entries = []; this.index = -1; this.onChange?.(this); }
+}

web/workspace_progress.js

@@ -0,0 +1,71 @@
+import {T} from './i18n.js';
+const stageNames={
+    CheckpointLoaderSimple:"Loading main model",UNETLoader:"Loading main model",UNETLoaderGGUF:"Loading main model",
+    CLIPLoader:"Loading text encoder",DualCLIPLoader:"Loading text encoder",VAELoader:"Loading VAE",
+    CLIPTextEncode:"Encoding prompts",LoraLoader:"Loading LoRA",'Power Lora Loader (rgthree)':"Loading LoRA",
+    AdachiInlineMask:"Reading source image and mask",AdachiWebUIInpaint:"Preparing inpaint sampling",
+    KSampler:"KSampler",KSamplerAdvanced:"KSampler",SamplerCustom:"Sampler",
+    VAEDecode:"VAE decode",VAEEncode:"VAE encode",AdachiImageCompare:"Preparing comparison images",
+};
+
+// Percentages describe the current node's reported steps, not invented overall
+// job progress. Nodes without numeric progress use an indeterminate bar.
+export class WorkspaceProgress {
+    constructor(onChange=()=>{}) {this.onChange=onChange;this.state='idle';this.buffer=[];}
+    begin(nodes={},promptId=null) {
+        this.nodes=nodes;this.promptId=promptId;this.state='queued';this.label=T("Queued…");
+        this.nodeId=null;this.value=null;this.max=null;this.buffer=[];this.emit();
+    }
+    attach(promptId) {
+        this.promptId=promptId;const events=this.buffer;this.buffer=[];
+        for(const [name,data] of events)if(data?.prompt_id===promptId)this.handle(name,data);
+    }
+    emit(){this.onChange(this);}
+    stage(id) {
+        const node=this.nodes?.[String(id)],type=node?.class_type;
+        return stageNames[type]?T(stageNames[type]):node?._meta?.title||type||T("Executing workflow");
+    }
+    finish(success,message) {
+        this.state=success?'complete':'error';this.label=message||(success?T("Generation complete"):T("Generation stopped"));
+        this.value=success?1:null;this.max=success?1:null;this.emit();
+    }
+    connection(connected) {
+        if(!['queued','running'].includes(this.state))return;
+        this.label=connected?T("Reconnected; waiting for execution progress…"):T("Connection lost; recovering execution progress…");this.value=this.max=null;this.emit();
+    }
+    handle(name,data={}) {
+        // Older frontends reduce executing payloads to a scalar node ID and
+        // dispatch null at completion, including before the submit HTTP reply.
+        if(data==null)return false;
+        if(name==='executing' && ['string','number'].includes(typeof data))data={node:String(data)};
+        if(typeof data!=='object')return false;
+        if(!['queued','running'].includes(this.state))return false;
+        if(!this.promptId){this.buffer.push([name,data]);if(this.buffer.length>64)this.buffer.shift();return false;}
+        if(data.prompt_id && data.prompt_id!==this.promptId)return false;
+        if(name==='execution_success'){this.finish(true);return true;}
+        if(['execution_error','execution_interrupted'].includes(name)){this.finish(false,data.exception_message||T("Generation stopped"));return true;}
+        if(name==='execution_start'){this.state='running';this.label=T("Starting execution…");this.emit();return true;}
+        if(name==='execution_cached'){
+            if(this.state==='queued'){this.label=T("Reading cache · {count} nodes",{count:data.nodes?.length||0});this.emit();}return true;
+        }
+        if(name==='progress_state') {
+            const running=Object.entries(data.nodes||{}).filter(([,n])=>n.state==='running');
+            if(!running.length)return false;
+            const [id,node]=running.at(-1);
+            return this.handle('progress',{...node,prompt_id:this.promptId,node:id,
+                max:Number(node.max)>1?node.max:0});
+        }
+        if(name==='executing' && data.node==null && data.node_id==null)return false;
+        const id=data.node ?? data.node_id ?? this.nodeId;
+        if(id==null || !this.nodes?.[String(id)])return false;
+        const changed=String(id)!==this.nodeId;this.nodeId=String(id);this.state='running';
+        if(changed || name==='executing'){this.value=this.max=null;}
+        const max=Number(data.max),value=Number(data.value);
+        if(name==='progress' && Number.isFinite(max) && max>0 && Number.isFinite(value)){
+            this.max=max;this.value=Math.max(0,Math.min(max,value));
+            const type=this.nodes[this.nodeId]?.class_type;
+            this.label=`${type==='AdachiWebUIInpaint'?T("Inpaint sampling (KSampler)"):this.stage(id)} · ${this.value} / ${this.max}`;
+        } else this.label=this.stage(id);
+        this.emit();return true;
+    }
+}

web/workspace_routes.js

@@ -0,0 +1,35 @@
+const linkAt = (graph, input) => input?.link == null ? null
+    : graph.getLink?.(input.link) ?? graph.links?.get?.(input.link) ?? graph.links?.[input.link];
+// Follow the actual sampler conditioning inputs, including KJ Get/Set routes.
+// Ambiguous processing graphs are left under the workflow's control.
+export function conditioningSource(graph, sampler, inputName, modes = new Map()) {
+    const mode = node => modes.get(node) ?? node.mode ?? 0;
+    const visit = (node, slot, seen = new Set()) => {
+        if (!node || seen.has(node) || mode(node) === 2) return null;
+        seen = new Set(seen); seen.add(node);
+        const follow = input => {
+            const link = linkAt(node.graph || graph, input);
+            return link ? visit((node.graph || graph).getNodeById(link.origin_id), link.origin_slot, seen) : null;
+        };
+        if (node.type === 'AdachiExclusiveSwitch') {
+            const sources = (node.inputs || []).filter(i=>['input_a','input_b'].includes(i.name)).map(follow).filter(Boolean);
+            return sources.length === 1 ? sources[0] : null;
+        }
+        if (node.type === 'GetNode') {
+            if (mode(node) !== 0) return null;
+            const name = node.widgets?.[0]?.value;
+            const setter = node.findSetter?.(node.graph || graph) || Array.from(graph._nodes || []).find(n=>n.type === 'SetNode' && n.widgets?.[0]?.value === name);
+            return setter ? visit(setter,slot,seen) : null;
+        }
+        if (node.type === 'SetNode' || node.type === 'Reroute') {
+            if (mode(node) !== 0) return null;
+            return follow(node.inputs?.[0]);
+        }
+        if (node.type === 'CLIPTextEncode') return mode(node) === 0 ? node : null;
+        const resolved = node.resolveVirtualOutput?.(slot);
+        if (resolved?.node) return visit(resolved.node,resolved.slot ?? 0,seen);
+        return null;
+    };
+    const link = linkAt(graph, sampler.inputs?.find(i=>i.name===inputName));
+    return link ? visit(graph.getNodeById(link.origin_id),link.origin_slot) : null;
+}

web/workspace_splitter.js

@@ -0,0 +1,48 @@
+import {T} from './i18n.js';
+export function installWorkspaceSplitter(root,main,divider) {
+    const key='adachi-workspace-sidebar-width',minimum=280,canvasMinimum=320;
+    let requested=360,drag=null;
+    try {const saved=Number(localStorage.getItem(key));if(Number.isFinite(saved) && saved>=minimum)requested=saved;}catch{}
+    const bounds=()=>{
+        const style=getComputedStyle(main),width=main.clientWidth-parseFloat(style.paddingLeft)-parseFloat(style.paddingRight);
+        return {width,max:Math.max(minimum,Math.min(760,width-canvasMinimum-10))};
+    };
+    const render=()=>{
+        const {width,max}=bounds(),actual=Math.min(max,Math.max(minimum,requested));
+        root.style.setProperty('--aw-sidebar-width',`${actual}px`);
+        divider.setAttribute('aria-valuemin',String(Math.min(canvasMinimum,width-minimum-10)));
+        divider.setAttribute('aria-valuemax',String(Math.max(0,width-minimum-10)));
+        divider.setAttribute('aria-valuenow',String(Math.max(0,width-actual-10)));
+        divider.setAttribute('aria-valuetext',T("Settings panel {width} pixels wide",{width:Math.round(actual)}));
+        return actual;
+    };
+    const save=()=>{try{localStorage.setItem(key,String(requested));}catch{}};
+    const change=value=>{requested=Math.min(bounds().max,Math.max(minimum,value));render();};
+    divider.setAttribute('role','separator');divider.setAttribute('aria-orientation','vertical');
+    divider.setAttribute('aria-label',T("Resize canvas and settings"));divider.tabIndex=0;
+    divider.title=T("Drag to resize the canvas and settings; double-click to reset, or use arrow keys for fine adjustments");
+    divider.onpointerdown=e=>{
+        if(e.button!==0)return;e.preventDefault();e.stopPropagation();
+        drag={id:e.pointerId,x:e.clientX,width:render()};divider.setPointerCapture(e.pointerId);
+        root.dataset.resizing='true';divider.focus({preventScroll:true});
+    };
+    divider.onpointermove=e=>{if(drag?.id!==e.pointerId)return;e.preventDefault();change(drag.width+drag.x-e.clientX);};
+    const finish=e=>{
+        if(drag?.id!==e.pointerId)return;drag=null;delete root.dataset.resizing;
+        if(divider.hasPointerCapture(e.pointerId))divider.releasePointerCapture(e.pointerId);save();
+    };
+    divider.onpointerup=finish;divider.onpointercancel=finish;divider.onlostpointercapture=finish;
+    divider.ondblclick=e=>{e.preventDefault();requested=360;render();save();};
+    divider.onkeydown=e=>{
+        const step=e.shiftKey?48:16;
+        if(e.key==='ArrowLeft')change(render()+step);
+        else if(e.key==='ArrowRight')change(render()-step);
+        else if(e.key==='Home')change(bounds().max);
+        else if(e.key==='End')change(minimum);
+        else if(e.key==='Enter'){requested=360;render();}
+        else return;
+        e.preventDefault();e.stopPropagation();save();
+    };
+    const observer=new ResizeObserver(render);observer.observe(main);render();
+    return ()=>{observer.disconnect();delete root.dataset.resizing;};
+}

web/workspace_transport.js

@@ -0,0 +1,44 @@
+import {T} from './i18n.js';
+// Check protocol support before POSTing: old ComfyUI processes may serve the
+// updated JavaScript while still running the previous Python node package.
+const unavailable = () => new Error(T("The workbench backend is not loaded. Stop and restart ComfyUI, then check connections again; refreshing the browser is not enough."));
+export async function workspaceRequest(api, path, prepared, {signal}={}) {
+    const controller = new AbortController();
+    const abort=()=>controller.abort();
+    if(signal?.aborted)controller.abort();else signal?.addEventListener('abort',abort,{once:true});
+    const timer = setTimeout(() => controller.abort(), path === 'submit' ? 60000 : 20000);
+    try {
+        controller.signal.throwIfAborted();
+        let base;
+        for (const candidate of ['/adachi-inpaint/workspace', '/api/adachi-inpaint/workspace']) {
+            const probe = await api.fetchApi(`${candidate}/capabilities`, {signal:controller.signal});
+            if (!probe.ok) continue;
+            let info; try { info = await probe.json(); } catch { continue; }
+            if (info.api_version === 1 && info.can_validate) { base = candidate; break; }
+        }
+        if (!base) throw unavailable();
+        const response = await api.fetchApi(`${base}/${path}`, {
+            method:'POST', headers:{'Content-Type':'application/json'},
+            body:JSON.stringify(prepared), signal:controller.signal,
+        });
+        if ([404,405].includes(response.status)) throw unavailable();
+        let data;
+        try { data = await response.json(); } catch { throw new Error(T("Invalid workbench response ({status}); check your proxy settings or restart ComfyUI",{status:response.status})); }
+        if (!response.ok || !data.valid) {
+            const errors = Object.entries(data.node_errors || {}).flatMap(([id,node]) =>
+                (node.errors || []).map(e => {
+                    const labels={ckpt_name:T("Main model"),unet_name:T("Main model"),clip_name:T("Text encoder"),vae_name:'VAE',lora_name:'LoRA'};
+                    const field=e.extra_info?.input_name,value=e.extra_info?.received_value;
+                    if(e.type==='value_not_in_list' && labels[field] && typeof value==='string')return T("{label} not found: {value}",{label:labels[field],value});
+                    return `#${id} ${e.message || ''}: ${e.details || ''}`;
+                }));
+            const error = new Error(errors.join('\n') || data.error?.message || T("Workflow validation failed"));
+            error.nodeErrors = data.node_errors || {}; throw error;
+        }
+        return data;
+    } catch (error) {
+        if(signal?.aborted)throw error;
+        if (error.name === 'AbortError') throw new Error(path === 'submit' ? T("Submission timed out. Check the task queue before retrying to avoid duplicate generations.") : T("Connection check timed out; check your network and retry"));
+        throw error;
+    } finally {clearTimeout(timer);signal?.removeEventListener('abort',abort);controller.abort();}
+}

web/workspace_ui.js

@@ -0,0 +1,205 @@
+import {T} from './i18n.js';
+import { app } from '../../scripts/app.js';
+import { api } from '../../scripts/api.js';
+import {selectWorkbenchPipeline} from './workbench_analysis.js';
+import {executionId} from './workspace_flow.js';
+import { WorkspaceBindings } from './workspace_bindings.js';
+import { WorkspaceHistory } from './workspace_history.js';
+import { WorkspaceProgress } from './workspace_progress.js';
+import { installWorkspaceSplitter } from './workspace_splitter.js';
+import { prepareWorkspacePrompt, validateWorkspace, submitWorkspace } from './workspace_execution.js';
+import { syncCurrentDimensions } from './parameter_bridge.js?v=0.8.2';
+const el=(tag,text,cls)=>{const e=document.createElement(tag);if(text)e.textContent=text;if(cls)e.className=cls;return e;};
+const btn=(text,fn)=>{const b=el('button',text);b.type='button';b.onclick=fn;return b;};
+const icons={edit:'<path d="m14 3 7 7-11 11H3v-7L14 3Z"/><path d="m11 6 7 7"/>',compare:'<rect x="3" y="4" width="18" height="16" rx="2"/><path d="M12 2v20M3 8h9M3 13h9M3 18h9"/>',side:'<rect x="3" y="4" width="18" height="16" rx="2"/><path d="M12 4v16M6 12h12m-3-3 3 3-3 3"/>',theme:'<path d="M20 15.5A9 9 0 0 1 8.5 4 9 9 0 1 0 20 15.5Z"/>',check:'<circle cx="12" cy="12" r="9"/><path d="m8 12 3 3 5-6"/>',back:'<path d="m9 5-7 7 7 7M3 12h19"/>'};
+const iconButton=(label,kind,fn)=>{const b=btn('',fn);b.title=label;b.setAttribute('aria-label',label);b.innerHTML=`<svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.7" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">${icons[kind]}</svg>`;b.className='aw-icon';return b;};
+let currentWorkspace=null;
+export class Workspace {
+ constructor(samplerNode=null,options={}){
+  this.preferredSampler=samplerNode;this.pipelineOptions=options;this.mounts=[];this.disposers=[];this.closed=false;this.running=false;this.checkVersion=0;
+  this.sessionController=new AbortController();
+  this.root=el('div','','adachi-workspace');this.root.adachiWorkspace=this;this.root.setAttribute('role','dialog');this.root.setAttribute('aria-label','Adachi Workbench');
+  const head=el('header'),title=el('h1','Adachi Workbench');head.append(title);
+  this.status=el('span',T("Reading the current workflow…"),'aw-status');this.themeButton=iconButton(T("Switch theme"),'theme',()=>this.toggleTheme());head.append(this.status,this.themeButton,iconButton(T("Check connections"),'check',()=>this.check()),iconButton(T("Return to workflow"),'back',()=>this.close()));this.root.append(head);
+  const main=el('main'),canvas=el('section','','aw-canvas'),toolbar=el('div','','aw-toolbar');
+  this.tabs={edit:iconButton(T("Edit mask"),'edit',()=>this.show('edit')),compare:iconButton(T("Slide to compare"),'compare',()=>this.show('compare'))};
+  this.sideButton=iconButton(T("Show before"),'side',()=>this.show(this.view==='before'?'after':'before'));
+  toolbar.append(this.tabs.edit,this.tabs.compare,this.sideButton);
+  toolbar.append(el('span','','spacer'));this.adoptButton=btn(T("Adopt"),()=>this.adopt());this.adoptButton.title=T("Use this result as the next source image and keep the mask; restore earlier images below");this.adoptButton.className='primary';toolbar.append(this.adoptButton);
+  this.surface=el('div','','aw-surface');this.still=el('img','','aw-result-image');this.still.alt=T("Selected source image or generated result");this.still.hidden=true;this.surface.append(this.still);
+  const foot=el('div','','aw-footer');this.undoButton=btn(T("↶ Undo"),()=>this.restoreHistory('undo'));this.originalButton=btn(T("Initial source"),()=>this.restoreHistory('original'));this.historySelect=el('select');this.historySelect.setAttribute('aria-label',T("Choose an adopted source image"));this.historySelect.onchange=()=>this.restoreHistory(Number(this.historySelect.value));foot.append(this.undoButton,this.originalButton,this.historySelect);
+  this.footer=foot;this.undoButton.title=T("Restore the previous source image, mask and crop");this.originalButton.title=T("Restore the initial source image of this session");this.historySelect.title=T("Choose an earlier adopted or uploaded source image");this.note=el('div','','aw-note');canvas.append(toolbar,this.surface,foot,this.note);
+  this.side=el('aside');this.generationPanel=el('div','','aw-generation-progress');this.generationPanel.hidden=true;this.generationLabel=el('span');this.generationBar=el('progress');this.generationBar.max=1;this.generationBar.setAttribute('aria-label',T("Current stage progress"));this.generationPanel.title=T("Actual progress from the current node; stages without step counts show an indeterminate bar");this.generationPanel.append(this.generationLabel,this.generationBar);this.side.append(this.generationPanel,el('h2',T("Workflow settings")));this.settings=el('div');this.params=el('div','','aw-parameters');this.runButton=btn(T("Generate"),()=>this.run());this.runButton.disabled=true;this.runButton.className='primary';this.runButton.style.width='100%';this.side.append(this.settings,this.runButton,this.params);this.divider=el('div','','aw-splitter');main.append(canvas,this.divider,this.side);this.root.append(main);
+  this.root.addEventListener('keydown',e=>{if(e.key==='Escape'&&e.target===this.root)this.close();});
+  this.root.addEventListener('pointerdown',e=>e.stopPropagation());document.body.append(this.root);document.body.classList.add('adachi-workspace-open');this.stopSplitter=installWorkspaceSplitter(this.root,main,this.divider);
+  // Render first. Discovery/catalog and actual server validation run afterward.
+  let theme='light';try{theme=localStorage.getItem('adachi-workspace-theme')||theme;}catch{}this.setTheme(theme);
+  setTimeout(()=>this.initialize().catch(e=>this.setStatus(e.message,'error')),0);
+ }
+ setTheme(theme){this.root.dataset.theme=theme==='dark'?'dark':'light';const label=this.root.dataset.theme==='dark'?T("Switch to light theme"):T("Switch to dark theme");this.themeButton.title=label;this.themeButton.setAttribute('aria-label',label);try{localStorage.setItem('adachi-workspace-theme',this.root.dataset.theme);}catch{}}
+ toggleTheme(){this.setTheme(this.root.dataset.theme==='dark'?'light':'dark');}
+ setStatus(text,state='pending'){if(this.closed)return;this.status.textContent=text;this.status.title=text;this.status.dataset.state=state;}
+ renderTaskProgress(){
+  const p=this.taskProgress;if(!p || this.closed)return;
+  this.generationPanel.hidden=p.state==='idle';this.generationPanel.dataset.state=p.state;
+  const known=p.max>0 && p.value!=null;
+  this.generationLabel.textContent=p.label+(known && p.max>1?` · ${Math.round(p.value/p.max*100)}%`:'');
+  this.generationBar.setAttribute('aria-valuetext',this.generationLabel.textContent);
+  if(known)this.generationBar.value=p.value/p.max;else this.generationBar.removeAttribute('value');
+  if(this.running && p.state!=='idle')this.setStatus(p.state==='complete'?T("Generation complete"):p.state==='error'?p.label:T("Generating"),p.state==='error'?'error':p.state==='complete'?'ok':'pending');
+ }
+ showGeneratedResult(){if(this.editor?.stroke || this.editor?.drag || this.editor?.pointerId!=null){this.pendingResultView=true;return;}this.pendingResultView=false;this.show('compare');}
+ mount(element,target,canvas=false){
+  if(!element?.parentNode)throw Error(T("Controls are not ready; reopen the workbench shortly"));
+  const marker=document.createComment('workspace');element.before(marker);this.mounts.push({element,marker,style:element.getAttribute('style')});target.append(element);if(canvas)element.classList.add('aw-embedded');
+ }
+ async initialize(){
+  if(this.closed)return;this.graph=app.rootGraph||app.graph;this.nodes=selectWorkbenchPipeline(this.graph,this.preferredSampler,this.pipelineOptions);
+  this.editorNode=this.nodes.editor;this.sampler=this.nodes.sampler;this.compareNode=this.nodes.compare;this.connectionsConfirmed=!this.nodes.ambiguous||!!this.pipelineOptions.confirmed;
+  this.editor=this.editorNode?.adachiEditor;this.compare=this.compareNode?.adachiCompare;
+  if(!this.editor||!this.sampler?.adachiWorkbench||!this.compare)throw Error(T("This workflow needs Inline Mask, Adachi Inpaint Workbench and Image Compare nodes"));
+  this.taskProgress=this.sampler.adachiWorkspaceProgress ||= new WorkspaceProgress();this.taskProgress.onChange=()=>this.renderTaskProgress();this.renderTaskProgress();
+  this.history=this.editorNode.adachiWorkspaceHistory ||= new WorkspaceHistory(this.editorNode);
+  this.history.onChange=()=>this.syncHistory();
+  this.history.onRestore=()=>{syncCurrentDimensions(this.editorNode);app.canvas?.emitBeforeChange?.();app.canvas?.emitAfterChange?.();this.show('edit');this.scheduleCheck();};
+  if(this.editor.ready)await this.history.checkpoint(T("Source when opened"));
+  if(this.closed)return;
+  if((app.rootGraph||app.graph)!==this.graph){this.close();return;}
+  const originalTargets=this.compare.options.getTargets,originalManual=this.compare.manualTarget,originalTarget=this.compare.targetSelect.value;
+  this.compare.options.getTargets=()=>({...originalTargets(),preferred:executionId(this.graph,this.editorNode)});this.compare.manualTarget=false;this.compare.refreshTargets();
+  this.disposers.push(()=>{this.compare.options.getTargets=originalTargets;this.compare.manualTarget=originalManual;this.compare.targetSelect.value=originalTarget;this.compare.refreshTargets();});
+  this.mount(this.editor.root,this.surface,true);this.mount(this.compare.root,this.surface,true);this.mount(this.sampler.adachiWorkbench.root,this.params);
+  // The dock can wrap on narrow screens or with long history labels. Keep
+  // crop/result controls above its measured height instead of a fixed offset.
+  const dockObserver=new ResizeObserver(()=>{
+   this.root.style.setProperty('--aw-history-height',`${this.footer.getBoundingClientRect().height}px`);
+   const crop=this.editorNode.adachiCrop?.bar;
+   if(crop)this.root.style.setProperty('--aw-crop-height',`${crop.getBoundingClientRect().height}px`);
+  });
+  dockObserver.observe(this.footer);if(this.editorNode.adachiCrop?.bar)dockObserver.observe(this.editorNode.adachiCrop.bar);
+  this.disposers.push(()=>dockObserver.disconnect());
+  const editorWheel=this.editor.options.onGraphWheel,compareWheel=this.compare.options.onGraphWheel;
+  this.editor.options.onGraphWheel=()=>{};this.compare.options.onGraphWheel=()=>{};
+  this.disposers.push(()=>{this.editor.options.onGraphWheel=editorWheel;this.compare.options.onGraphWheel=compareWheel;});
+  this.bindings=new WorkspaceBindings({graph:this.graph,sampler:this.sampler,canvas:app.canvas,onChange:()=>this.scheduleCheck(),markChanged:()=>{app.canvas?.emitBeforeChange?.();app.canvas?.emitAfterChange?.();}});
+  this.settings.append(this.connectionPicker(),this.bindings.createPanel());
+  this.prepare(true).catch(e=>this.setStatus(e.message,'error'));
+  this.bindings.loadCatalog(api.fetchApi.bind(api)).then(()=>this.scheduleCheck()).catch(e=>this.setStatus(e.message,'error'));
+  const oldAdopt=this.compare.options.onAdopt;
+  this.compare.options.onAdopt=async(...args)=>{const result=await this.history.transact(T("Adopted result"),()=>oldAdopt(...args));if(!this.closed && this.editorNode.adachiCrop?.mode==='crop')this.show('edit');return result;};
+  this.disposers.push(()=>{this.compare.options.onAdopt=oldAdopt;});
+  const oldLoad=this.compare.loadSelected;
+  this.compare.loadSelected=async(...args)=>{const result=await oldLoad.apply(this.compare,args);if(!this.closed){this.show(this.view||'edit');this.updateButtons();}return result;};
+  this.disposers.push(()=>{this.compare.loadSelected=oldLoad;});
+  const oldUpload=this.editor.upload;
+  this.editor.upload=async(...args)=>{await oldUpload(...args);if(this.editor.ready&&!this.closed){await this.history.checkpoint(T("Uploaded source"));this.syncHistory();this.scheduleCheck();}};
+  this.disposers.push(()=>{this.editor.upload=oldUpload;});
+  const onExecuted=this.compareNode.onExecuted;
+  this.compareNode.onExecuted=(output)=>{const ret=onExecuted?.call(this.compareNode,output);if(output?.adachi_results?.length){this.showGeneratedResult();this.note.textContent='';this.updateButtons();}return ret;};
+  this.disposers.push(()=>{this.compareNode.onExecuted=onExecuted;});
+  const event=(name,fn)=>{api.addEventListener(name,fn);this.disposers.push(()=>api.removeEventListener(name,fn));};
+  for(const name of ['execution_start','executing','execution_cached','progress','progress_state','execution_success','execution_error','execution_interrupted'])event(name,e=>this.taskProgress.handle(name,e.detail));
+  event('reconnecting',()=>this.taskProgress.connection(false));event('reconnected',()=>this.taskProgress.connection(true));
+  const queued=e=>{if(e.detail.sampler!==this.sampler)return;this.promptId=e.detail.promptId;this.running=!['complete','error'].includes(this.taskProgress.state);this.runButton.textContent=this.running?T("Generating…"):T("Generate");this.renderTaskProgress();this.updateButtons();};
+  document.addEventListener('adachi-workspace-queued',queued);this.disposers.push(()=>document.removeEventListener('adachi-workspace-queued',queued));
+  const recovered=e=>{if(e.detail.node!==this.compareNode || e.detail.promptId!==this.promptId)return;if(e.detail.entry.outputs?.[executionId(this.graph,this.compareNode)]?.adachi_results?.length)this.showGeneratedResult();this.taskProgress.finish(e.detail.entry.status?.status_str==='success');this.running=false;this.sampler.adachiWorkspaceTask=null;this.runButton.disabled=false;this.runButton.textContent=T("Generate");this.setStatus(e.detail.entry.status?.status_str==='success'?T("Generation complete"):T("Generation failed; check the ComfyUI task history"),e.detail.entry.status?.status_str==='success'?'ok':'error');this.updateButtons();};
+  document.addEventListener('adachi-result-recovered',recovered);this.disposers.push(()=>document.removeEventListener('adachi-result-recovered',recovered));
+  for(const name of ['execution_success','execution_error','execution_interrupted'])event(name,e=>{if(e.detail.prompt_id===this.promptId){this.running=false;this.sampler.adachiWorkspaceTask=null;this.runButton.disabled=false;this.runButton.textContent=T("Generate");this.setStatus(name==='execution_success'?T("Generation complete"):e.detail.exception_message||T("Generation stopped"),name==='execution_success'?'ok':'error');}});
+  this.timer=setInterval(()=>{if((app.rootGraph||app.graph)!==this.graph||this.editor.destroyed)this.close();else{this.updateButtons();if(this.pendingResultView)this.showGeneratedResult();}},500);
+  // The comparison's exact-task recovery also serves the workspace. A second
+  // history poller doubled weak-network traffic and could replay old results.
+  this.promptId=this.sampler.adachiWorkspaceTask;this.running=!!this.promptId;if(this.running){if(this.taskProgress.state==='idle')this.taskProgress.begin({},this.promptId);this.compareNode.adachiTrackPrompt?.(this.promptId);this.runButton.disabled=true;this.runButton.textContent=T("Generating…");}this.show('edit');this.syncHistory();this.scheduleCheck();
+ }
+ connectionPicker(){
+  const panel=el('details','','aw-connections');panel.open=!this.connectionsConfirmed;panel.append(el('summary',T("Workflow connections")));
+  if(!this.connectionsConfirmed)panel.append(el('p',T("Multiple candidates found; select the nodes to use")));
+  const controls={};
+  for(const [name,label,list,current]of [['sampler',T("Inpaint node"),this.nodes.samplers,this.sampler],['editor',T("Image editor"),this.nodes.editors,this.editorNode],['compare',T("Image comparison"),this.nodes.comparisons,this.compareNode]]){
+   const row=el('label',label),select=el('select');for(const node of list){const id=executionId(this.graph,node),option=el('option',`${node.title||node.type} #${id}`);option.value=id;select.append(option);}select.value=executionId(this.graph,current);row.append(select);panel.append(row);controls[name]=select;
+  }
+  const presetRow=el('label',T("Group presets")),preset=el('select'),none=el('option',T("No preset controller"));none.value='';preset.append(none);
+  for(const node of this.nodes.presets.filter(n=>n.mode===0)){const id=executionId(this.graph,node),option=el('option',`${node.title||node.type} #${id}`);option.value=id;preset.append(option);}preset.value=this.bindings.controller?executionId(this.graph,this.bindings.controller.node):'';presetRow.append(preset);panel.append(presetRow);
+  const apply=btn(T("Use these connections"),()=>{
+   const find=(list,value)=>list.find(node=>executionId(this.graph,node)===value),sampler=find(this.nodes.samplers,controls.sampler.value),editor=find(this.nodes.editors,controls.editor.value),compare=find(this.nodes.comparisons,controls.compare.value);
+   if(!sampler||!editor||!compare)return;sampler.properties.adachi_workspace_connections={editorId:executionId(this.graph,editor),compareId:executionId(this.graph,compare),controllerId:preset.value};
+   this.close();currentWorkspace=new Workspace(sampler,{editorNode:editor,compareNode:compare,confirmed:true});
+  });apply.title=T("Select nodes from this workflow; this inpaint node remembers your choices");panel.append(apply);return panel;
+ }
+ syncHistory(){if(!this.history)return;this.historySelect.replaceChildren();this.history.entries.forEach((item,i)=>{const o=el('option',`${i+1} · ${item.label}`);o.value=i;this.historySelect.append(o);});this.historySelect.value=String(this.history.index);this.undoButton.disabled=!this.history.canUndo;this.originalButton.disabled=this.history.index===0;}
+ updateButtons(){const busy=this.running||this.preparing||this.submitting||this.sampler?.adachiWorkspaceSubmitting;this.runButton.disabled=!!(this.running||this.submitting||this.sampler?.adachiWorkspaceSubmitting);this.adoptButton.disabled=!this.compare?.ready||!!this.compare?.busy||this.compare?.pendingOutputs>0||busy;this.tabs.compare.disabled=!this.compare?.ready;this.sideButton.disabled=!this.compare?.ready;if(!this.preparing)this.bindings?.refreshProfiles();if(this.editor)this.editor.root.inert=(!this.connectionsConfirmed)||!!this.preparing&&!this.discovering;this.settings.inert=!!this.preparing;this.params.inert=!!this.preparing;this.historySelect.disabled=this.preparing||!!this.history?.busy;this.undoButton.disabled=this.preparing||!this.history?.canUndo;this.originalButton.disabled=this.preparing||this.history?.index<=0;if(this.compare?.adoptButton)this.compare.adoptButton.disabled=this.adoptButton.disabled;}
+ show(which){this.view=which;if(this.editor)this.editor.root.hidden=which!=='edit';if(this.compare)this.compare.root.hidden=which!=='compare';this.still.hidden=!['before','after'].includes(which);if(!this.still.hidden){const image=which==='before'?this.compare?.before:this.compare?.after;if(image)this.still.src=image.src;}for(const [id,b]of Object.entries(this.tabs)){b.classList.toggle('active',id===which);b.setAttribute('aria-pressed',String(id===which));}this.sideButton.classList.toggle('active',['before','after'].includes(which));this.sideButton.setAttribute('aria-pressed',String(which==='before'));this.sideButton.title=which==='before'?T("Show after"):T("Show before");this.sideButton.setAttribute('aria-label',this.sideButton.title);this.editor?.schedule();this.compare?.schedule();}
+ async adopt(){if(!this.compare?.ready||this.running)return;await this.compare.adopt();this.syncHistory();this.updateButtons();if(!this.compare.message.classList.contains('dd-error')){this.show(this.editorNode.adachiCrop?.mode==='crop'?'edit':'compare');this.note.textContent='';this.scheduleCheck();}}
+ async restoreHistory(which){try{if(this.preparing||this.history?.busy)return;if(which==='undo')await this.history.undo();else if(which==='original')await this.history.original();else await this.history.restore(which);this.syncHistory();}catch(e){this.setStatus(e.message,'error');}}
+ scheduleCheck(){++this.checkVersion;this.checkController?.abort();clearTimeout(this.checkTimer);if(!this.closed)this.checkTimer=setTimeout(()=>this.check(),200);}
+ prepare(discovery=false){
+  const task=(this.prepareTask||Promise.resolve()).catch(()=>{}).then(async()=>{
+   if(this.closed||(app.rootGraph||app.graph)!==this.graph)throw Error(T("The workflow changed"));
+   if(!this.connectionsConfirmed&&!discovery)throw Error(T("Select the nodes to use in Workflow connections first"));
+   if(!this.editor?.ready&&!discovery)throw Error(T("Upload a source image first"));
+   this.preparing=true;this.discovering=discovery;this.updateButtons();
+   try{return await prepareWorkspacePrompt({editorNode:this.editorNode,samplerNode:this.sampler,compareNode:this.compareNode,allowEmpty:discovery,controller:this.bindings.controller,getBinding:(output,id)=>{this.bindings.updateExecutionGraph(output,id);return this.bindings.getExecutionBinding();}});}
+   finally{this.preparing=false;this.discovering=false;if(!this.closed)this.updateButtons();}
+  });
+  this.prepareTask=task;return task;
+ }
+ async check(){
+  if(this.closed||this.running||this.submitting||this.sampler?.adachiWorkspaceSubmitting)return false;
+  if(this.checking){this.recheck=true;return false;}
+  const version=++this.checkVersion,controller=this.checkController=new AbortController();this.checking=true;
+  this.setStatus(T("Checking in the background…"));
+  try{
+   const prepared=await this.prepare();
+   if(this.closed||this.recheck||version!==this.checkVersion)return false;
+   const result=await validateWorkspace(prepared,{signal:controller.signal});
+   if(this.closed||this.running||this.recheck||version!==this.checkVersion)return false;
+   if(result.valid===false||result.ok===false)throw Error(result.error?.message||result.error||T("Workflow validation failed"));
+   this.setStatus(T("Ready to generate"),'ok');return true;
+  }catch(e){if(version===this.checkVersion&&!this.running)this.setStatus(e.message||String(e),'error');return false;}
+  finally{this.checking=false;if(this.recheck){this.recheck=false;this.scheduleCheck();}}
+ }
+
+ async run() {
+  if(this.closed||this.running||this.submitting||this.sampler?.adachiWorkspaceSubmitting)return;
+  if(!this.sampler||!this.taskProgress){this.setStatus(T("Controls are not ready; reopen the workbench shortly"),'error');return;}
+  this.submitting=true;this.sampler.adachiWorkspaceSubmitting=true;
+  ++this.checkVersion;this.checkController?.abort();clearTimeout(this.checkTimer);
+  this.runButton.disabled=true; this.runButton.textContent=T("Checking…");
+  const seed=this.sampler?.widgets?.find(w=>w.name==='seed'), oldSeed=seed?.value;
+  let queuedSeed=oldSeed;
+  try {
+   for(const widget of this.sampler.widgets || []) widget.beforeQueued?.();
+   queuedSeed=seed?.value;
+   const prepared=structuredClone(await this.prepare());
+   await validateWorkspace(prepared,{signal:this.sessionController.signal});
+   if(this.closed)return;
+   if((app.rootGraph||app.graph)!==this.graph||!this.sampler.graph||!this.compareNode.graph||this.editor.destroyed)throw Error(T("The workflow changed; check connections again"));
+   this.running=true;this.taskProgress.begin(prepared.prompt);this.promptId=null; this.updateButtons();
+   const queued=await submitWorkspace(prepared);
+   this.promptId=queued.prompt_id;this.taskProgress.attach(this.promptId);
+   for(const widget of this.sampler.widgets || []) widget.afterQueued?.();
+   this.sampler.adachiParameters.refresh();
+   this.running=!['complete','error'].includes(this.taskProgress.state);
+   this.sampler.adachiWorkspaceTask=this.running?this.promptId:null;
+   document.dispatchEvent(new CustomEvent('adachi-workspace-queued',{detail:{sampler:this.sampler,promptId:this.promptId}}));
+   this.compareNode.adachiTrackPrompt?.(this.promptId,this.graph);
+   if(this.closed)return;
+   this.runButton.textContent=this.running?T("Generating…"):T("Generate");
+   this.setStatus(this.running?T("Queued for generation"):this.taskProgress.label,this.running?'pending':this.taskProgress.state==='error'?'error':'ok');
+  } catch(e) {
+   this.running=false;
+   if(['queued','running'].includes(this.taskProgress.state))this.taskProgress.finish(false,e.message||String(e));
+   if(seed && seed.value===queuedSeed)seed.value=oldSeed;
+   this.setStatus(e.message||String(e),'error');
+  } finally {
+   this.submitting=false;this.sampler.adachiWorkspaceSubmitting=false;
+   if(!this.running){this.runButton.disabled=false;this.runButton.textContent=T("Generate");}
+   if(!this.closed)this.updateButtons();
+  }
+ }
+
+ close(){if(this.closed)return;this.closed=true;this.sessionController.abort();this.checkController?.abort();++this.checkVersion;clearInterval(this.timer);clearInterval(this.pollTimer);clearTimeout(this.checkTimer);this.stopSplitter?.();for(const dispose of this.disposers.reverse())dispose();if(this.history){this.history.onChange=null;this.history.onRestore=null;}if(this.taskProgress)this.taskProgress.onChange=()=>{};this.bindings?.destroy();for(const {element,marker,style}of this.mounts.reverse()){element.classList.remove('aw-embedded');element.hidden=false;element.inert=false;if(marker.parentNode)marker.replaceWith(element);if(style===null)element.removeAttribute('style');else element.setAttribute('style',style);}this.root.remove();document.body.classList.remove('adachi-workspace-open');this.editor?.schedule();this.compare?.schedule();this.compare?.updateButton();currentWorkspace=null;}
+}
+app.registerExtension({name:'Adachi.InpaintWorkspace',setup(){
+ const css=document.createElement('link');css.rel='stylesheet';css.href=new URL('./workspace.css?v=2.0.0',import.meta.url).href;document.head.append(css);
+ document.addEventListener('adachi-open-workspace',event=>{if(!currentWorkspace)currentWorkspace=new Workspace(event.detail?.sampler||null);});
+}});

workspace_execution.py

@@ -0,0 +1,124 @@
+"""Validate and queue only the selected inpainting output, without a test render."""
+
+from copy import deepcopy
+import inspect
+import time
+import uuid
+
+
+def validate_input_mappings(prompt):
+    # Core replacement hooks inspect inputs before validate_prompt runs. Reject
+    # malformed mappings here so a bad request cannot crash those hooks first.
+    for key, node in prompt.items():
+        if not isinstance(node, dict) or not isinstance(node.get("inputs", {}), dict):
+            raise ValueError(f"Invalid input mapping for execution node: {key}")
+
+
+def dependency_prompt(prompt, output_id):
+    """Keep one output and its dependencies, including references into subgraphs."""
+    output_id = str(output_id)
+    if not isinstance(prompt, dict) or not isinstance(prompt.get(output_id), dict):
+        raise ValueError("The selected comparison is not present in the execution graph.")
+    if prompt[output_id].get("class_type") != "AdachiImageCompare":
+        raise ValueError("The workspace output must be an Adachi Image Compare node.")
+    found, pending = set(), [output_id]
+    while pending:
+        key = pending.pop()
+        if key in found:
+            continue
+        found.add(key)
+        node = prompt.get(key)
+        if not isinstance(node, dict):
+            raise ValueError(f"The execution graph references a missing node: {key}")
+        inputs = node.get("inputs", {})
+        if not isinstance(inputs, dict):
+            raise ValueError(f"Invalid input mapping for execution node: {key}")
+        for value in inputs.values():
+            if isinstance(value, list) and len(value) == 2 and isinstance(value[0], str) and isinstance(value[1], int):
+                pending.append(value[0])
+    return {key: value for key, value in prompt.items() if key in found}
+
+
+async def prepare_workspace(server, execution, payload):
+    """Use the same hooks, replacements and validation as ComfyUI /prompt."""
+    if not isinstance(payload, dict) or not isinstance(payload.get("prompt"), dict):
+        raise ValueError("No workflow execution graph was provided.")
+    target = str(payload.get("compare_id", ""))
+    data = deepcopy(payload)
+    validate_input_mappings(data["prompt"])
+    # Hooks may restore Get/Set or subgraph links; prune only after they ran.
+    data = server.trigger_on_prompt(data)
+    if not isinstance(data, dict) or not isinstance(data.get("prompt"), dict):
+        raise ValueError("A workflow extension rejected the execution graph.")
+    validate_input_mappings(data["prompt"])
+    replacements = getattr(server, "node_replace_manager", None)
+    if replacements is not None:
+        replacements.apply_replacements(data["prompt"])
+    data["prompt"] = dependency_prompt(data["prompt"], target)
+    data["partial_execution_targets"] = [target]
+    prompt_id = str(uuid.uuid4())
+    # Older supported ComfyUI builds have validate_prompt(prompt), newer builds
+    # include prompt_id and optional partial execution targets.
+    count = len(inspect.signature(execution.validate_prompt).parameters)
+    if count >= 3:
+        checked = execution.validate_prompt(prompt_id, data["prompt"], [target])
+    elif count == 2:
+        checked = execution.validate_prompt(prompt_id, data["prompt"])
+    else:
+        checked = execution.validate_prompt(data["prompt"])
+    valid = await checked if inspect.isawaitable(checked) else checked
+    return data, prompt_id, valid
+
+
+def register_workspace_routes(server):
+    if getattr(server, "_adachi_workspace_execution_routes", False):
+        return
+    from aiohttp import web
+    import execution
+
+    @server.routes.get("/adachi-inpaint/workspace/capabilities")
+    async def capabilities(request):
+        return web.json_response({"api_version": 1, "can_validate": True})
+
+    async def handle(request, submit):
+        try:
+            payload = await request.json()
+            data, prompt_id, valid = await prepare_workspace(server, execution, payload)
+            response = {"valid": bool(valid[0]), "error": valid[1], "node_errors": valid[3]}
+            if not valid[0]:
+                return web.json_response(response, status=400 if submit else 200)
+            if submit:
+                extra = data.get("extra_data", {})
+                if not isinstance(extra, dict):
+                    raise ValueError("Invalid workflow metadata.")
+                if isinstance(data.get("client_id"), str):
+                    extra["client_id"] = data["client_id"]
+                extra["create_time"] = int(time.time() * 1000)
+                sensitive = {}
+                sensitive_keys = getattr(execution, "SENSITIVE_EXTRA_DATA_KEYS", None)
+                for key in sensitive_keys or ():
+                    if key in extra:
+                        sensitive[key] = extra.pop(key)
+                number = server.number
+                server.number += 1
+                item = (number, prompt_id, data["prompt"], extra, valid[2])
+                if sensitive_keys is not None:
+                    item += (sensitive,)
+                server.prompt_queue.put(item)
+                response.update(prompt_id=prompt_id, number=number)
+            else:
+                response["outputs"] = valid[2]
+                response["node_count"] = len(data["prompt"])
+            return web.json_response(response)
+        except (ValueError, TypeError, KeyError) as exc:
+            return web.json_response({"valid": False, "error": {"message": str(exc)}, "node_errors": {}}, status=400)
+
+    @server.routes.post("/adachi-inpaint/workspace/validate")
+    async def validate(request):
+        return await handle(request, False)
+
+    @server.routes.post("/adachi-inpaint/workspace/submit")
+    async def submit(request):
+        return await handle(request, True)
+
+    server._adachi_workspace_execution_routes = True