●Change8

v0.3.30

📦 comfyuiView on GitHub →
✨ 16 features🐛 10 fixes🔧 7 symbols

Summary

This release introduces significant feature enhancements, including the Unified Weight Adapter system, OFT/BOFT algorithms, API Node support, and updates for HiDream and WAN VACE models. Several bug fixes address frontend issues, video encoding, and specific model implementations.

Migration Steps

  1. If you rely on the old video encoding method, note that aom-av1 has been replaced with svt-av1 for webm saving.
  2. Users utilizing DoRA should benefit from the corrected implementation which maintains backward compatibility.

✨ New Features

  • Added FreSca node.
  • Added support to output camera state.
  • Hidream: Allow loading hidream text encoders in CLIPLoader and DualCLIPLoader.
  • Added CLIPTextEncodeHiDream node.
  • Support for WAN VACE preview model.
  • Added strength control for vace.
  • Introduced Unified Weight Adapter system for better maintainability and future Lora features.
  • Added OFT/BOFT algorithm in weight adapter.
  • Corrected DoRA implementation with backward compatibility.
  • Added option for using fp8_e8m0fnu for model weights.
  • Replaced aom-av1 with svt-av1 for saving webm videos, using preset 6 + yuv420p10le pixel format.
  • Added backbone for supporting API nodes in ComfyUI.
  • Pytorch stable 2.7 is out and supports cu128.
  • fp16 accumulation can now be enabled on the stable package.
  • Implemented multi image prompting for gpt-image-1.
  • Added support for API Nodes in ComfyUI.

🐛 Bug Fixes

  • Updated frontend to version 1.16.9.
  • Slightly better fix for issue #7687.
  • Fixed issue with WAN VACE implementation.
  • Fixed dino lowvram issues.
  • Updated frontend to version 1.17.10.
  • Fixed transparency in outputs.
  • Updated frontend to version 1.17.11.
  • Threw away alpha channel in clip vision preprocessor.
  • Fixed multiple image return from api nodes.
  • Switched LTXVPreprocess to libx264.

🔧 Affected Symbols

graph.get_input_infoCLIPLoaderDualCLIPLoaderCLIPTextEncodeHiDreamMaskPreviewdinoLTXVPreprocess