v0.12.3
📦 comfyuiView on GitHub →
✨ 4 features🐛 4 fixes🔧 4 symbols
Summary
This release focuses on stability and improvements for the ACE (Audio Creative Engine) workflow, including fixes for model loading, checkpoint saving, and addressing NaN issues during ACE steps. New features include enhanced audio decoding and LLM sampling options.
Migration Steps
- Disable sage attention if encountering issues on ACE step 1.5, as it was disabled by default in this release.
✨ New Features
- Added File3DAny output to Load3D node.
- Extended SaveGLB to accept File3DAny as input.
- Added LLM sampling options and made reference audio work on ACE step 1.5.
- Added VAE tiled decode node for audio.
🐛 Bug Fixes
- Fixed crash on safetensors load for some 3rd party Models.
- Fixed checkpoint saving in mp module.
- Fixed ACE step NaN issue on some hardware/PyTorch configurations.
- Attempted to fix ACE text encoder slowness on some configurations.