●Change8

v0.25.0

📦 comfyuiView on GitHub →
✨ 22 features🐛 15 fixes🔧 25 symbols

Summary

This release introduces numerous new partner nodes, model supports (like SCAIL-2 and Bernini-R), and 10-bit video capability. It also includes several bug fixes related to Ideogram stability, 3D preview nodes, and various partner node issues.

Migration Steps

  1. If using audio nodes that were consolidated, update workflows to use the new SaveAudioAdvanced node (CORE-202).
  2. If using external source of pinned memory, check for potential interoperation issues.
  3. If using Seedance 2.0, check for frame stretching issues at 1080p.
  4. If using Ideogram, ensure model dtype is not being incorrectly set to fp8 or causing crashes related to cast buffers.
  5. If using 3D nodes, note the reordering of inputs/outputs for Preview3DAdvanced / PreviewGaussianSplat / PreviewPointCloud (feat(3d)).
  6. If using Ideogram 4, be aware of potential dtype issues.
  7. If using SCAIL-2, review reference mask background convention.
  8. If using Manager, note that --enable-manager-legacy-ui now implies --enable-manager.

✨ New Features

  • Allow SaveWEBM node to save with alpha channel.
  • Add BriaTransparentVideoBackground node.
  • Add Bria Green Background node.
  • Add Krea 2 Medium Turbo model.
  • Add seed input to Flux Erase node.
  • Consolidate audio nodes into SaveAudioAdvanced node.
  • Add PreviewGaussianSplat node.
  • Add PreviewPointCloud nodes.
  • Add SeedVR2 support (Note: This was reverted later in the notes, but was introduced as a feature).
  • Add Color primitive.
  • Improve ResolutionSelector functionality.
  • Extract image dimensions at ingest and emit on asset responses.
  • Add model support for SCAIL-2.
  • Add Bernini-R model support (Wan video).
  • Add Depth anything 3 support.
  • Add cursor-based pagination on GET /api/assets.
  • Add job_ids filter to GET /api/assets.
  • Include asset id in executed WebSocket message.
  • Add Comfy-Usage-Source pass-through for API node requests.
  • Enable Bria Replace Background node.
  • Add Runway Aleph2 node.
  • Add 10-bit video support.

🐛 Bug Fixes

  • Fix image grid bug (CORE-215).
  • Fix Seedance 2.0: prevent 1080p first/last-frame stretch jump.
  • Fix ideogram if model dtype gets set to fp8.
  • Fix interoperation with external source of pinned memory pressure.
  • Fix 3D nodes: let Load3D accept output/temp paths, save Preview3DAdvanced to temp/ (BE-1172).
  • Fix ideogram crash: mm: dont reset cast buffers in cleanup_models_gc().
  • Avoid conditions become trainable in training node ([Trainer/bug]).
  • Ensure model is not inference mode during training ([Trainer/bug] CORE-72).
  • Fix: Add back apply_rotary_emb for Qwen Image.
  • Fix ideogram crash: mm: dont reset cast buffers in cleanup_models_gc().
  • Fix SCAIL-2 reference mask background convention.
  • [Partner Nodes] Fix GPT Image: handle mismatched image sizes returned when size="auto".
  • [Partner Nodes] Fix KlingTextToVideoNode: validation error for "kling-v2-master" model.
  • Fix potential dtype issue with ideogram 4.
  • Use comfy kitchen apply rope in omnigen2 model.

Affected Symbols

SaveWEBM nodeBriaTransparentVideoBackground nodeSeedance 2.0Bria Green Background nodeKrea 2 Medium Turbo modelFlux Erase nodeSaveAudioAdvanced nodePreviewGaussianSplat nodePreviewPointCloud nodesLoad3DPreview3DAdvancedSeedVR2Color primitiveResolutionSelectorIdeogramQwen ImageLTXV/LTXV modelsSCAIL-2Bernini-R modelDepth anything 3Gemini text nodeNanoBanan nodeGPT Image nodeKlingTextToVideoNodeRunway Aleph2 node