Comfyui Model Resolver v1.2.1

v1.2.1

This update expands model discovery for newer Hugging Face and CivitAI naming patterns and current ComfyUI catalogs. It also improves workflow-aware suggestions, search state, downloads, and missing-model handling.

What’s New

• Model Resolver can open directly from ComfyUI’s native missing-model error overlay.
• The built-in catalog adds 24 model entries and six base-model families, including LTXV 2.5, Muse Image, Wan Video 3.0, MiniMax Music 3, Pixal3D, and Trellis.2. The update also refreshes the bundled Hugging Face fallback index.
• Hugging Face downloads support repository and revision identifiers such as owner/repository and main. Filename matching recognizes convrot, mixed-precision, and scaled-precision variants.

Improvements

• CivitAI search can find models whose workflow filenames differ from provider titles, reject short generic partial matches, and avoid repeated requests when complete file data already exists.
• Automatic model-family filters and download-folder suggestions follow the active checkpoint or diffusion model. LoRA suggestions ignore CLIP and VAE dependencies, upscalers use an independent filter, and manual “Suggest Subfolder” follows the configured folder rule.
• Batch search now runs sequentially within each provider while different providers run independently. Queued items, partial failures, and refresh actions show clearer status feedback.
• The Downloads panel keeps its Active and History tabs available while queue views refresh. The Missing Models browser restores its saved split before appearing.
• Retained SHA-256-mismatch downloads include both expected and calculated hashes, allowing the calculated hash to be reused if the file is restored.

Fixed Issues

• Downloads resolve model destinations across ComfyUI folder aliases.
• Search and download state stays attached to the correct missing-model card. Local matches survive post-download analysis, stale responses from another workflow are ignored, and shared node/widget coordinates no longer mix progress.
• The Apply action remains disabled until successful linking removes its queued card and becomes available again when linking fails.
• Loaded-model detection prefers explicit model fields over stale embedded metadata.
• Saved workflow SHA-256 records follow model paths and names instead of node/widget positions.

Do You Need to Do Anything?

Reload or restart ComfyUI after updating. No workflow migration, setting changes, or credential reconnection is required.

Short Version

• Open Model Resolver directly from native ComfyUI missing-model errors.
• Find more Hugging Face and CivitAI models with improved workflow-aware matching.
• Get more reliable batch searches, downloads, and workflow state handling.

Changes since v1.2.0

• Update pyproject.toml (b5ab69e)
• Update README.md (d6b5b30)
• Update huggingface-author-index.json (2d687a2)
• fix(search): exclude CLIP and VAE from LoRA workflow fallback base model detection (62c04ab)
• Make batch source search sequential per source and independent across sources (cf5c73c)
• Use getMissingModelDomKey for search UI element IDs (9f9b3da)
• Optimize CivitAI lookups by caching model data and skipping unnecessary version requests (9f537ea)
• fix: match workflow hashes by model path instead of widget position (6305081)
• fix: keep Apply button disabled during queued resolution and re-enable on failure (b93f63e)
• frontend: standardize options status styling with CSS classes (e9bc636)
• Add refresh animations for busy option/search controls (d26b0a7)
• Save metadata on SHA256-mismatched downloads and verify reused hash behavior (248ccea)
• frontend: ignore stale local-match refresh responses after workflow switch (2d1bc22)
• data update (df90db5)
• perf: optimize hash cache building with in-memory dir entries and thread pool (27ce316)
• frontend: prefer workflow base model over incompatible archive metadata in auto mode (ea3b987)
• Treat upscale categories as Upscaler for auto base model resolution (f358df4)
• feat: prefer active workflow widgets over model metadata for base model detection (c5371e7)
• Ignore stale embedded model metadata when a node exposes an explicit model widget (1eb3af3)
• frontend: manual subfolder suggestion respects configured folder rule (cd626de)
• feat: add CivitAI title-based search fallback for filename mismatches (ddde0ac)
• Fix CivitAI filename matching false positives for short generic prefixes (7627c61)
• Update huggingface-author-index.json (854ce6c)
• feat: support HuggingFace string identifiers, convrot variants, and mixed/scaled precision tokens (9c24d92)
• refactor: unify model data contracts (ddc5333)
• Add native ComfyUI error overlay detection and auto-open integration (6b96e94)
• frontend: restore missing-browser split width synchronously on initial paint (65fd1ac)
• Model Database Update (8f9ecb6)
• Add persistent queue downloads subtabs container (a0f370e)
• Fix model download directory alias resolution (026ba97)

Full Changelog: v1.2.0...v1.2.1