- Filter non-model files in scanner when folder_paths category extensions are empty
- Skip candidates with known non-model extensions in find_matches
- Add .engine, .patch, and .tflite to MODEL_EXTENSIONS
- Add unit tests for empty extension directory scanning and matcher candidate filtering
- Recover completed .part files on HTTP 416 without deleting and redownloading
- Add stream stall detection and automatic reconnect for interrupted socket reads
- Implement persistent download state store and pause/resume lifecycle controls
- Add RateLimitCoordinator and ConnectivityGuard circuit breaker for requests
- Support proxy_url configuration and validate aria2 proxy settings
- Add unit tests for 416 recovery, stall reconnect, state store, and network guard
- Keep failed or interrupted downloads in active downloads queue instead of dismissing immediately
- Present interrupted state with error tooltip, distinct status chip, and progress percentage
- Add in-place Retry download play action that resumes downloading from existing partial files
- Support dismissing interrupted downloads via cancel button
- Update CSS styling and unit tests for download action handlers and progress states
- Drain aria2 daemon stderr in a background thread to prevent Windows pipe buffer saturation and deadlock
- Promote critical disk write and permission errors from aria2 stderr to warnings
- Automatically refresh expired signed CDN URLs and resume downloads via aria2 continue=true
- Shield addUri against mid-registration cancellations to prevent orphaned downloads
- Retry Range requests directly on target URL when redirects drop the Range header in Python downloads
- Add auto-reconnect stream loop on transient socket disruptions with seamless partial-hash resumption
- Increase default download chunk size to 4MB for improved I/O throughput
- Parse HTTP 429 Retry-After headers for informative rate limit reporting
- Add unit test suite covering new aria2 and Python downloader resilience behaviors
- Omit blocking checksum and check-integrity options from aria2 transfers to prevent long pauses at 100% on large models
- Run SHA-256 integrity verification asynchronously in a background thread via Python hashlib
- Update download progress and metadata sidecars upon background verification completion or mismatch
- Add unit tests for aria2 option omission and background verification
Add custom node adapter for ComfyUI_SmartModelLoader_Nova suite:
- Support 'Smart Model Loader [SML2]', 'CLIP Loader [SML2]', and 'LoRA Stack [SML2]'
- Parse and serialize canonical JSON configuration widget
- Extract diffusion models, VAE, CLIP text encoders, and LoRAs with strengths
- Enable model resolution by hash using per-resource SHA256 stored in the node
- Update model paths and SHA256 checksums in canonical JSON format during relinking
- Add unit and end-to-end integration tests
- Add paste handler (Ctrl+V) in Metadata tab for workflow JSON, images, and Civitai image URLs/HTML
- Resolve authentic .safetensors filenames and local ComfyUI model paths via Civitai modelVersionId
- Normalize prompt <lora:...> tags to authentic filenames and enrich transfer/loaded chip labels
- Enable multi-selection on model chips with click, Ctrl+click (toggle/add), Shift+click (range from anchor), and Ctrl+Shift+click.
- Add keyboard shortcuts: Ctrl+C to copy selected model tokens to clipboard with feedback, Ctrl+A to select all visible chips, and Escape to clear selection.
- Add a subtle green outline and background highlight for selected chips (.is-selected).
- Add a 'Selected' mode button to the copy toolbar that dynamically appears when chips are selected and syncs token count and preview text.
- Stop click event propagation on chips and clean up prior selection instances during tab switching and re-rendering to prevent ghost listeners and bubbling conflicts between tabs.
- Add comprehensive Node test suite covering selection operations, keyboard shortcuts, DOM rendering, and tab transitions.
Use order-insensitive token comparison in calculate_filename_confidence
so variant names with reordered tokens (e.g., int8_convrot vs convrot_int8)
score higher. Replace first-match partial matching in HuggingFace source with
best-match selection based on filename confidence, and surface confidence in
search results.
Refactor searchMissingBatch to schedule each source independently via a
per-source promise chain (scheduleSource) instead of processing all targets
in a single serial loop. This keeps source ordering intact while allowing
different sources to advance in parallel across models.
Also update searchOnline to accept scheduleSource and emit a pending UI
state for queued searches, split pending/running source colors in CSS, and
add tests covering independent batch source progression plus error/cancel
survival.
Replace inline node_id/widget_index DOM key construction with the
centralized getMissingModelDomKey helper across search panel and
missing browser methods. This keeps search results and button IDs
stable when the same model appears in multiple workflows/widgets.
Also adds a regression test verifying that search progress stays
with its model when switching between shared widgets.
Reuse model metadata across _find_civitai_file_in_model calls within a single search, avoid resolve_urn when modelVersions already contains usable file lists, and add tests for request-count behavior.
Dariusz L