v0.22.0
📦 comfyuiView on GitHub →
✨ 6 features🐛 16 fixes⚡ 2 deprecations🔧 10 symbols
Summary
This release introduces significant feature additions, including support for MoGe and Stable Audio 3, alongside numerous bug fixes across various nodes and models. Several OpenAPI documentation elements were updated, including deprecating an endpoint.
Migration Steps
- When using the OpenAPI specification, use /api/upload/image instead of the deprecated /api/upload/mask endpoint.
- Update usage of Asset schema in API interactions to no longer rely on top-level width/height fields.
✨ New Features
- Added support for MoGe (CORE-168).
- Added downscaled IC-LoRA support to LTXVAddGuide (CORE-102).
- Added optional attention_mask input to LTXVAddGuide (CORE-220).
- Added new StringFormat node.
- Support for Stable Audio 3 model added.
- Added spacial downscale ratio to empty ltxv latent.
🐛 Bug Fixes
- Audio processing nodes now handle None inputs.
- Preserved noise_scale/shift across chained model_sampling patches.
- Allowed negative `batch_index` on `ImageFromBatch` and `LatentFromBatch` (CORE-195).
- Reduced LTX2.3 peak VRAM when guide_mask is in use (CORE-166).
- Fixed Qwen3.5 text generation with multi-image prompts.
- HiDream-O1 now supports area conditioning.
- Partner Nodes: Fixed Opus 4.7 sending deprecated temperature parameter.
- FeatherMask: Corrected negative zero indexing for right/bottom feathering.
- Implemented/fixed dynamic clip saving.
- Partner Nodes: Added ByteDance Seed LLM node.
- Fixed BiRefNet issue.
- Reduced minimum value for Batch Image/Mask/Latent nodes from 2 to 1.
- Initial work to make downscale_ratio_temporal work.
- Used temporal downscale to make empty audio latent nodes more reusable.
- Fixed typo in ops.py.
- Fixed Hunyuan3D 2.1 batch size crashes in attention and forward pass.
Affected Symbols
⚡ Deprecations
- The OpenAPI endpoint /api/upload/mask is deprecated in favor of /api/upload/image.
- Top-level width/height fields on the Asset schema in OpenAPI documentation have been removed.