●Change8

v0.3.51

📦 comfyuiView on GitHub →
✨ 9 features🐛 16 fixes🔧 3 symbols

Summary

This release focuses heavily on compatibility fixes, especially for older PyTorch versions, alongside numerous updates to API nodes and improvements to Qwen model handling. Key additions include context window support and new video/image generation nodes.

Migration Steps

  1. Ensure 'av' dependency is installed as it is now considered essential.

✨ New Features

  • Added context window support to core sampling code.
  • Added FluxKontextMultiReferenceLatentMethod node.
  • Implement wan2.2 camera model.
  • Added record audio node.
  • API nodes added kling-v2-1 and v2-1-master.
  • API nodes added MinimaxHailuoVideoNode node.
  • API nodes added GPT-5 series models.
  • API nodes released version of google's models.
  • API nodes added Vidu Video nodes.

🐛 Bug Fixes

  • Restored frontend version display.
  • Reduced portable size.
  • Fixed XPU iGPU regressions.
  • Fixed crash on old PyTorch versions related to a previous PR.
  • Avoided torch compile graphbreak for older PyTorch versions.
  • Fixed issue where the last commit did not work on older PyTorch.
  • Added warning when using old PyTorch.
  • Fixed MIME type for multipart uploads in OpenAIGPTImage1 node.
  • Fixed invalid nodeid for WanTrackToVideo.
  • Made context step index detection much more robust.
  • Refactored Qwen image model logic.
  • Changed TextEncodeQwenImageEdit node to use logic closer to reference code.
  • Fixed Rope implementation for Qwen VL.
  • Ensured Qwen rotary embeddings match reference code.
  • Disabled prompt weights for Qwen.
  • Fixed bf16 precision issue with Qwen image embeddings.

🔧 Affected Symbols

OpenAIGPTImage1WanTrackToVideoTextEncodeQwenImageEdit