v0.3.61
📦 comfyuiView on GitHub →
✨ 7 features🐛 6 fixes🔧 6 symbols
Summary
This release introduces new API nodes for Rodin3D and audio processing, alongside significant internal updates converting many nodes to the V3 schema. Key improvements include memory leak fixes and enhancements to the stable release automation.
Migration Steps
- Update command to install latest nightly pytorch using the new command provided in the release.
- Users relying on specific file naming for stable NVIDIA portables might need to check new configurations.
✨ New Features
- Added Rodin3D Gen-2 generate api-node.
- Added new audio nodes.
- LatentCompositeMasked now works with basic video latents.
- ComfyUI can now be run on the specified Ascend NPU.
- Updated the name of the save file path for Rodin3d api nodes (changed from timestamp to UUID).
- Added WanImageToImageApi node.
- Added workflow templates version tracking to system_stats response.
🐛 Bug Fixes
- Fixed issue with .view() in HuMo.
- Fixed memory leak by explicitly detaching model finalizer.
- Prevented adding template to qwen2.5vl when template is already in prompt.
- Fixed stable workflow creating multiple draft releases.
- Fixed typo in release workflow.
- Fixed stable release test failing intermittently.
🔧 Affected Symbols
HuMoLatentCompositeMaskedsampler_cfg_functionqwen2.5vlRodin3D Gen-2 generate api-nodeWanImageToImageApi