Allow the number of CivitAI search candidates to be configured (default 5, bounds 1–20) and thread this limit through the frontend and backend. Backend: introduce DEFAULT/MAX constants, validate candidate_limit, include it in cache keys and logs, collect and deduplicate candidates from tRPC and HTML searches up to the limit, prefer exact 100% matches and otherwise pick the best-confidence result. Frontend: persist civitaiCandidateLimit in localStorage, add an input to options UI, clamp/save the value, include civitai_candidate_limit in search payloads, and apply suggested subfolder info from returned civitai_search_result. Misc: UI copy updated and minor log/message tweaks.
Add UI and behavior for showing/hiding API tokens: new CSS (.ml-options-input-row, .ml-options-visibility-btn), visibility buttons next to CivitAI, CivitAI session and HuggingFace inputs, and JS helpers (getVisibilityIcon, syncVisibilityToggle, bindVisibilityToggle) to toggle input type and update button state/aria attributes. Introduce an inline help button and expandable help panel with step-by-step instructions for obtaining the __Secure-civitai-token and an "Open CivitAI" action. Remove the Clear Tokens button from the actions and its previous handler. Overall improves UX for managing and understanding token usage while keeping tokens stored locally.
Add support for an optional CivitAI session token and cache-management for search results. UI: new input for __Secure-civitai-token saved to localStorage (modelLinker.civitaiSessionToken) and cleared via the options dialog; saving/clearing tokens triggers a backend cache clear and updates notifications. Backend: add clear_search_cache() helpers in core.sources.huggingface and core.sources.civitai, register /model_linker/clear-search-cache route to flush both caches, and wire the civitai session token into search flows. CivitAI changes include sending the session token as a Cookie header for tRPC searches, incorporating session and model_type into the search cache key, and passing session_token through search_civitai_for_file and related calls. Purpose: ensure authenticated/session-aware search results (including NSFW/visibility differences) and avoid stale cached search results after token/settings changes.
Add formatSearchResultMeta and renderOnlineSearchResultCard helper methods to web/linker.js and replace repeated HTML construction for online search results (Popular, Model Database, HuggingFace, CivitAI) with calls to these helpers. This centralizes formatting for filename/confidence/size/secondary text and download action rendering, simplifies per-source code (including a top-line model link for CivitAI), and improves readability/consistency without changing user-facing behavior.
Pass category to CivitAI searches and overhaul civitai source logic: prefer civitai.red tRPC search, extract candidates from tRPC/HTML, add type mapping, richer filename normalization and similarity scoring, per-version file matching, and confidence thresholds; use caching and fallbacks when needed. Also update web UI to display match confidence and file size. These changes aim to produce more accurate model/file matches and surface confidence to users.
Add extensive observability/logging across ModelLinker and source search modules. __init__.py: introduce summarize_result and log_search_result helpers, replace ad-hoc logger calls with log_info/log_exception, and add start/result/summary logs for local, popular, model_list, HuggingFace and CivitAI searches (including extra metadata like confidence, files_count, results_count). core/sources/civitai.py: log cache hits, API call start, response counts, model/version/file checks, no-result info and improved exception logging. core/sources/huggingface.py: log cache hits, API call start, response counts, per-repo debug info, repo tree sizes, no-result info and improved exception logging. Changes are focused on diagnostics and do not alter core search logic.
Add real-time analysis progress reporting and client-side polling during workflow analysis. __init__.py: import asyncio, track per-analysis progress in self.analysis_progress, pass an update callback to analyze endpoint (using asyncio.to_thread), set progress states on success/error, and expose a new GET /model_linker/analyze-progress/{analysis_id} endpoint. core/linker.py: add optional progress_callback parameter to analyze_and_find_matches and emit staged progress updates (extracting, scanning, analyzing, identifying, matching, per-model progress and completed). web/linker.js: introduce analysis progress UI rendering, polling loop (pollAnalysisProgress) keyed by an analysis token, send analysis_id with analyze request, show progress bar/message, and replace search-source buttons with a compact select picker. These changes improve UX by providing live feedback during potentially long analysis runs.
Adjust visual styling and spacing in LinkerManagerDialog: remove the hard .ml-card:hover background, add subtle themed styles for download and footer buttons (hover and default states), soften .ml-tab hover colors/gradient, and reduce padding in several header/content areas for a more compact layout. Also set a dialog background variable (--ml-bg) for the main panel. These tweaks improve visual consistency and hover feedback across the dialog.
Tighten up LinkerManagerDialog UI: reduce paddings, margins and gaps across cards, columns, rows and inputs to create a more compact layout. Wrap download controls in a new .ml-download-actions container and introduce compact button variants (.ml-btn-sm, .ml-btn-download-main, .ml-btn-download-side, .ml-btn-link-compact) and adjusted icon-only sizes for consistent button sizing. Minor HTML tweaks to download block and match buttons (added compact class on link buttons). Adjust tab active background and shadowing to use theme variable and softer shadows. Remove the inline CivitAI API-key note and add a subtle header bottom border.
Overhauled the LinkerManagerDialog appearance and title rendering. Introduces new CSS variables (darker background, surface elevation, text-soft, refined accents), updates spacing, radii, shadows, fonts and component styles (cards, chips, badges, buttons, tabs, match rows, combo inputs and footer). Restructures card title HTML into meta/eyebrow/primary/secondary elements and improves URN resolution UX by showing a "Resolving CivitAI model..." placeholder and updating the async resolver output to use a classed link and smarter label fallback. Overall changes improve visual consistency, readability and UX for resolved/unresolved URNs.
Major UI refactor for the LinkerManagerDialog: replace dark gradients with translucent theme variables, unify card/match/button styling using transparent borders and inset shadows, and add new utility/button variants (ml-btn-utility, ml-btn-icon-only). Replace manual confidence badge markup with getConfidenceBadge and add getStatusBadge helper. Convert the Locate button into a clickable locatable node chip with icon/title, change combo refresh button to an icon-only control, and update the "other matches" entry to use an inline action class. Small layout/interaction tweaks include hover transforms, adjusted paddings/radii, divider variable, and presenting "Not needed" download state when a perfect local match exists. These changes improve visual consistency, compactness, and UX for locating and selecting matches.
Adjust visual styling in web/linker.js to create a more compact, consistent UI. Reduces paddings, margins, gaps and font sizes across cards, rows, tabs and controls; tweaks border radii, shadows and combo/list dimensions. Introduces a new .ml-btn-ghost variant (and applies it to Link/Locate/Refresh buttons) for a subtle button appearance, and updates confidence badge colors/borders for clearer visual hierarchy. Overall changes aim to tighten spacing and harmonize button/badge styling for improved readability and consistency.
Enhance the LinkerManager dialog UI and add safe HTML handling for model descriptions. This adds tooltip '?' helpers with explanatory titles for multiple info rows, wraps descriptions with a collapsible "Show more" widget, and implements sanitizeDescriptionHtml to allow a safe subset of tags/styles/links (and enforce target/rel for external links). Styling updates polish tabs, footer buttons and the description area, remove inline button sizing, and introduce a tab content animation. Also adjust click handling for description toggle and trained-words copy behavior and standardize footer button classes.
Enhance the model info dialog UI and behavior: add selectable trained-words with copy button, normalize/escape/truncate helpers, and persist dialog size using ResizeObserver/localStorage. Update image rendering to safer escaped URLs and richer captions, improve layout and responsive styles, allow resizing with min dimensions, and clean up observers/timers on close. Also bind interaction handlers once per dialog and update summary/count state for selected words.
Performance and correctness improvements across model scanning, matching, workflow analysis, and CivitAI URL handling:
- Add short-lived in-memory cache for get_model_files (with force_rescan and invalidate_model_files_cache) to reduce frequent directory scans.
- Pass scanned available_models into workflow analysis to avoid rescanning and build category-indexed lists plus an ordered-candidates cache for faster candidate ordering.
- Improve analyze_and_find_matches to special-case urn:air: URNs (use filename suffix when present) and skip fuzzy matching when appropriate.
- Rewrite find_matches to maintain only the top N results via heapq, cache normalized candidate path/filename values on candidate dicts, and early-exit on max_results<=0 to reduce allocations and CPU.
- Add helper functions in CivitAI source to extract/build stable image page URLs (_extract_civitai_image_id, _build_civitai_image_url) and use them when populating image metadata.
- Propagate available_models into get_node_model_info and analyze_workflow_models so subroutines reuse the same scanned list.
- Guard frontend and backend folder lookups to return early for unknown categories to avoid unnecessary work.
These changes aim to reduce IO/CPU during frequent matching operations and produce more accurate/efficient matching for URNs and local models.
Include base_model and tags in civitai search/resolve results and model lookup responses. Add a new /model_linker/subfolders API route that uses ComfyUI folder_paths to return known subfolder paths for a given category. Improve downloader.get_download_directory to choose a preferred path (preferring paths outside the ComfyUI install) and add robust path normalization/commonpath checks. On the frontend, add download target controls (folder + optional subfolder) with styling, loading of download directories/subfolders, autocomplete and selection logic, heuristics to suggest civitai subfolders (particularly for loras), and integrate the chosen category/subfolder into the download payload and UI wiring.
Add buildContextMenuModelData to LinkerManagerDialog to normalize/resolve model fields (name, original_path, resolved_path, category) and derive a safe filename fallback. Use this helper when serializing match.model for context menu data in renderLocalMatchesContent and the alternative matches loop, ensuring consistent context payloads and avoiding missing/undefined fields when opening the context menu.
Expose local model matching and folder-opening functionality. Adds search_local_matches to core/linker.py to run local filename/path similarity matching with deduplication and category prioritization. Registers two new HTTP endpoints in __init__.py: /model_linker/local-matches (POST) to search local files and /model_linker/open-containing-folder (POST) to open the OS file explorer at a model's path. Updates web/linker.js to add context-menu item, UI rendering and wiring for local matches (including confidence filtering, expandable alternatives, and button handlers), client-side openContainingFolder call, and triggers to refresh URN local matches. Also surfaces civitai version_name in model results. These changes enable searching and linking local model files from the UI and opening their containing folders.
Introduce an Options tab and UI for storing CivitAI and HuggingFace API tokens locally. Adds styling, a displayOptions() panel with save/clear actions, getStoredTokens() helper, and integrates token persistence via localStorage. The tokens are now included in download requests (hf_token and civitai_key) so gated downloads can authenticate. Also updates tab switching and related UI visibility to accommodate the new tab.
Move direct URN->CivitAI download handling into the frontend and enrich backend metadata. Removed server-side immediate URN download exposure and adjusted comment numbering. Backend now selects a primary file from resolved CivitAI model info, includes model_id, version_id, size, and a modelVersionId query on model URLs.
Frontend: add renderDownloadSourceSection to render download buttons and CivitAI model-card links, display file sizes, and format names. Store missingModels for later updates; render a placeholder URN download section while resolving; when URN resolution returns, inject civitai info into the missing entry, build a download_source object, replace the placeholder with the rendered download UI, and attach the download button handler. Improved error handling and model URL construction to include modelVersionId when available.
Overhaul the Model Linker dialog styling and related markup. Introduces a new dark theme palette, gradients, shadows, and consistent radii via CSS variables; replaces many inline styles with reusable classes (ml-missing-summary, ml-card, ml-card-header, ml-card-title, ml-card-subtitle, ml-card-actions, ml-combo-*, ml-search-source-bar, ml-search-results, etc.). Improves card layout, badges, chips, buttons, combo (autocomplete) dropdown UI, download/search panels, and responsive behavior. Also updates HTML generation and a few JS hooks (e.g. scoped selector for search source buttons, combo list markup and IDs, search-results container, selected-bar styling) to match the new classes. Purpose: improve visual clarity, consistency, responsiveness and maintainability of the linker UI.
Backend: add explicit URN handling to resolve CivitAI model/version into a direct download source; normalize and accept a `sources` parameter (local, huggingface, civitai, all) and drive per-source searching. Reorder and tighten search logic (local/popular, model-list, HuggingFace, CivitAI) and include `searched_sources` in results and improved logging. Frontend: add per-missing-item search result cache and state, UI buttons to pick search sources, merge incremental results instead of overwriting, show selected-source helper text, pass selected source to the server, clear cache on refresh, and improve search status messages and UX. Overall: enables targeted online searches, preserves previous results, and surfaces direct CivitAI downloads for URNs.
Stop performing automatic online searches during scanning and matching: HuggingFace/CivitAI lookups are no longer done synchronously for each missing file. Core linker now marks URN entries for async resolution instead of calling CivitAI during scan, avoiding blocking network calls and reducing noise. Server-side /model_linker/search remains the on-demand endpoint; __init__.py adds resolve_urn import and tightens validation for URN requests (require model_id+version_id for URN-only requests). Frontend (web/linker.js) updated to show a loading indicator for URN entries and to fetch CivitAI info asynchronously via a new resolveUrnAsync() routine, then update the UI when results arrive. Also includes minor formatting/import grouping and UI label tweaks.
Bump logging version to 1.4.1 and improve module name detection by adding _find_project_root in log_funcs to locate the repository root (searches for a .git directory) and use its basename when LOG_MODULE_NAME is not set, with a fallback to the previous behavior. Also update the same version string in logger.py. Additionally, comment out two noisy debug log lines in core/scanner.py to reduce verbose output when directories are missing or after scanning.
Refactor AzLogsLogger instance methods (debug/info/warn/error/exception) and the module-level helper functions to call the central logger.log(...) with the appropriate LogLevel, and ensure exception() passes exc_info=True. Also add a standard library logger in core/log_system/log_funcs.py and emit an error if importing the custom logger fails to aid diagnosis.
Replace direct uses of the logging module and module-level loggers with the centralized log_system.log_funcs (log_debug/log_info/log_warn/log_error/log_exception) across multiple modules. Updated imports and removed unused logging.getLogger() instances in downloader, scanner, sources (civitai, huggingface, model_list, popular), workflow_analyzer, workflow_updater, and __init__.py. Minor message/text tweaks and import formatting applied; behavior unchanged aside from routing logs through the shared logging helpers.
Introduce a new logging subsystem under core/log_system (logger.py, config.py, log_funcs.py) implementing AzLogs (LogLevel enum, colored console output, optional file rotation) and a lightweight wrapper (log_debug/info/warn/error/exception) with safe fallback when AzLogs is unavailable. Integrate these log functions across many core modules (downloader, linker, matcher, scanner, workflow_analyzer, workflow_updater, sources and related files) and export logger and helper log functions from core.__init__. Replace several direct logging calls in linker with the new log_* helpers and update core.__init__ exports. Also include small code style and robustness tweaks (path/filename normalization, JSON/file handling, minor refactors and formatting fixes) to improve matching and model-list/popular lookups.
Enhance model linking with CivitAI integration and UI improvements.
- __init__.py: Expand model path resolution by building filename→path lookup (including no-ext keys) and using folder_paths to map common categories; add a /model_linker/civitai-search POST route that attempts metadata/hash-based lookup (or filename fallback) and returns CivitAI model/version info for the frontend.
- core/sources/civitai.py: Add hashing and metadata handling utilities: _get_sha256_hash, metadata discovery/reading, helpers to extract images/trained-words, get_model_info_by_hash and get_model_info_for_file (which prefers local metadata and otherwise queries CivitAI by hash). Also tidy/normalize existing search/resolve functions and add caching for hash lookups.
- web/linker.js: Add a right-click context menu for model chips with actions (Show Info, Open in CivitAI), implement dialog UI to display model info, and wire calls to the new civitai-search endpoint. Includes positioning/closing behavior and dialog population from returned data.
These changes enable accurate model identification via local metadata or sha256 lookups on CivitAI and provide convenient frontend actions to view model details or open CivitAI pages.
Add the "Lora Stacker (LoraManager)" node type to NODE_TYPE_TO_CATEGORY_HINTS and to the list of text-based Lora loader types so the analyzer treats the LoraManager Stacker as a lora loader and applies the same model-info extraction logic.
Add attachDragHandleIfNeeded() and call it when opening the dialog. The new helper binds a single mousedown listener on #model-linker-drag-handle (and a one-time document mouseup) to start dragging, set cursor state, and avoid attaching multiple times. It also skips drag behavior in fullscreen and safely no-ops if the handle element is missing.
Make MLFilterSwitch more robust by guarding against a missing filter button and caching the element reference before toggling active state. Use getAttribute('data-...') instead of dataset for consistent data access, and adjust child DIV indexing to account for a category header so active/inactive sections are selected correctly. Add comments and clearer variable names for maintainability.
Introduce queuing of model resolutions and a compact combo-style all-models picker. Adds updateSelectedBarForMissing and removeQueuedByKey to show/remove queued selections per missing slot and rebuild pending index after changes. Change individual match buttons (including alternative matches) to queue selections instead of immediately resolving, and wire remove button to key-based removal. Replace the old multi-select search with a debounced dropdown list that shows folder info, supports click-to-queue, focus/blur behavior and a refresh button to reload all models. Also insert a visible selected-bar element into the slot UI and adjust various handlers and logging to reflect the new workflow.
Introduce locateNodeInGraph in LinkerManagerDialog to center, select, and highlight nodes in the ComfyUI canvas with compatibility fallbacks for different app versions and error notifications. Add a 'Locate' button to node headers (top-level nodes only) and wire its click handler so the UI will focus the corresponding node. Improves UX for finding and inspecting missing or referenced nodes.
Add support for the gguf model format across the codebase and improve handling of nested/dict-type widget model references. Changes include:
- Add ".gguf" to model extension lists in core/linker.py, core/scanner.py, core/sources/huggingface.py, and core/workflow_analyzer.py so gguf files are detected and suggested.
- Update HuggingFace search/repo listing to consider gguf files and improve some request/response handling and logging.
- Improve scanner implementation: formatting/PEP8 cleanup, robust extension checks, normalized MODEL_EXTENSIONS set, safer path handling, and clearer logging.
- Add NESTED_MODEL_KEYS and logic in workflow_analyzer to detect model filenames stored inside dict-type widget values (e.g. rgthree Power Lora Loader) and record a nested_key for resolution.
- Propagate nested_key through apply_resolution mapping and update update_model_path to write resolved paths back into nested dict keys when present (instead of only replacing the top-level widget value).
- Update web UI (web/linker.js) regexes to strip/display gguf extensions in model name formatting.
These changes allow gguf models to be discovered, resolved and updated in workflows, and enable support for nodes that store model references inside nested dict fields.
Redesign the Model Linker dialog to improve UX: add a draggable window, fullscreen toggle, resizable two-column layout with a splitter, and a persistent queued selection panel for batching model resolutions. Implements queue management (queueResolution, applyPendingResolutions, remove/clear queued items), a searchable "all models" dropdown (debounced filtering, loaded via /model_linker/models), and batch apply via /model_linker/resolve. Persists UI state (position, size, fullscreen, splitter width, collapsed state) to localStorage and updates header/footer controls and styles. Also includes helper utilities (debounce, ensureAllModelsLoaded) and various UI wiring to keep selection bars and buttons in sync.
Add a visual category header to each model section in the Linker dialog and ensure the dialog defaults to the "Missing Models" tab when opened. Also skip rendering the missing-models list when the active tab isn't "missing" to avoid unnecessary DOM work. Minor UI tweaks include toggling tab classes and showing download/auto-resolve buttons when switching to the missing tab.
Filter out LoraManager (LoraLoaderV2) entries that already exist locally so they don't appear as missing, and skip matching for those entries. Integrate scanner data to detect local lora files (full_path/exists) in workflow analysis. Add special handling in update_model_path/update_workflow_nodes to update LoraManager lora lists and the formatted lora text widget (widgets_values[2] and [1]) when a new lora is resolved, using mapping fields is_lora_v2 and original_lora_name. Improve convert_to_relative_path and get_base_directory_for_model to prefer folder_paths lookup and preserve exact ComfyUI filename formats. In the web UI, make resolve button IDs unique by including a missingIndex, avoid duplicate listeners, forward is_lora_v2/original_lora_name to the resolve API, and add extra logging for debugging.
Add support for text-based Lora loaders (LoraLoaderV2 / LoraManager) and mark model refs with activity/connection state, plus UI improvements.
- __init__.py: clarify comment and propagate new flags from refs (is_lora_v2, active, connected).
- core/workflow_analyzer.py: reformatting and small style changes; add LoraLoaderV2 and LoraManager node hints; detect node connected/bypassed state; parse text-based Lora widget lists (name/strength/active) and emit model refs with strength, is_lora_v2, active and connected flags; include connected flag for URN and file refs; keep existing URN/file resolution and dedup logic but annotate refs with subgraph/context info.
- web/linker.js: group loaded models into active/inactive per-category, add UI filter buttons (All / Active / Inactive), separate copy buttons for active/inactive/all, update display to indicate active vs inactive models and copy behavior, and add global helper functions for filtering and clipboard operations.
These changes enable accurate detection of LoraV2 loader states and provide a clearer UI for copying model spec strings while preserving existing behavior for standard loaders.
Add a new POST /model_linker/loaded API that analyzes a workflow JSON to extract all referenced models (including URNs, embedded node.properties.models and Lora strength extraction) and returns a structured list of loaded models. Update the LinkerManagerDialog UI: add tabs (Missing Models / Loaded Models), styles for the loaded models view, tab switching logic, and new functions to load and render loaded models grouped by category with copy buttons and a consolidated copy-all string. Also hide download/auto-resolve controls when viewing the Loaded Models tab and adjust header layout to accommodate the tabs. Error handling and logging were added for robust API behavior.
Display the expected filename in the missing-item title and enable wrapping for node/category chips. Remove the duplicated filename row and make the "other matches below 100%" note expandable: clicking it toggles a hidden list of alternative matches. Each alternative now shows a confidence badge, truncated filename/path with tooltip, and a Link button that triggers the existing resolve action. These changes improve readability and UX when many matches are present.
Wrap missing file title in a CivitAI link when a URN is available, and construct a modelUrl from URN info. Make model/version labels clickable and highlight them (orange) when linking to CivitAI; otherwise fall back to muted text. In the found-on-CivitAI section, build a display name that includes the version when present, render it as a styled button that opens the CivitAI page, show the model type in muted text, and use expected_filename or fallbacks for downloads. Removed the previous separate "View" link in favor of the unified clickable model label/button.
Refactor and harden the Model Linker extension: normalize string quoting/formatting and improve logging and error responses across API routes. Make route registration more robust to missing PromptServer and dependencies. Add/expand download-related features: safer request parsing (including is_urn handling), CivitAI URN support & direct download URL resolution, popular/model-list/HuggingFace/CivitAI search fallbacks, improved header handling for HF/CivitAI, and better category/directory mapping. Improve downloader reliability and UX: folder_paths fallback, pre-initialize download progress entries, smoother speed calculation and CLI logging, cancellation cleanup on Windows, stronger exception handling, and consistent JSON responses from endpoints.
Detect ComfyUI-AIR URNs in workflows, resolve them to expected filenames via the CivitAI API, and prefer those filenames for matching.
Key changes:
- core/workflow_analyzer.py: added URN_REGEX, URN_TYPE_MAP, is_model_filename() now recognizes URNs, and get_node_model_info() emits structured URN info (is_urn, urn fields, category, exists=False).
- core/sources/civitai.py: added resolve_urn(model_id, version_id) with caching, API fetch of model/version and primary file, and result shaping (model_name, version_name, expected_filename, files).
- core/linker.py: import resolve_urn and call it for missing URNs, attach civitai_info/expected_filename, and prefer expected_filename as matching target; cleaned up deduplication logic.
- web/linker.js: UI shows URN original plus resolved CivitAI model/version labels where available.
- TODO.md: added plan/notes for URN detection and next steps.
This improves detection and fuzzy matching for models referenced by URNs so missing model resolution is more accurate. Further matcher and updater enhancements remain TODOs.
- Add CSS variables and global styles with custom scrollbar
- Implement two-column layout (Local Matches / Download)
- Add confidence badges with color coding
- Smart filename truncation with hover tooltips
- Consistent button styling with icons
- Sticky footer with backdrop blur
- Status alert containers for errors/success
- Best match highlighting for 95%+ confidence
- Hide Auto-Link button when no 100% matches
- Fix race condition in auto-resolve flow
- Wider dialog (1100px) for full filenames
- Add Download All Missing button to download models without local matches
- Button toggles to Cancel All (N) during active downloads
- Fix notification showing null for info type messages
- Preserve download state when dialog is closed and reopened
- Show downloads in progress message instead of all available during downloads
Dariusz L