Comfy UI

ComfyUI v0.1

ComfyUI v0.1

Hi, all the Comfy users! Dev Team is happy to announce the release of ComfyUI v0.1, a larger-than-normal update that brings improvements and new features.

Improved Search Capabilities

The updated search box now includes fuzzy search and node source information, making it easier to locate the nodes you need, even if you’re unsure of the exact name.

Beta Menu UI Enhancements

  • Experimental UI: You can now enable a more experimental version of our UI, which introduces several new features.
  • Node Library: The beta menu UI now includes a node library, simplifying the process of exploring and utilizing ComfyUI’s extensive features.
  • Queue / Image Feed: The updated beta menu UI also features a queue and image feed, providing enhanced control and visibility over your generation tasks.

Litegraph Quality-of-Life Enhancements

Thanks to our dedicated contributors, Litegraph has received several quality-of-life improvements in this release.

Model Enhancements

  • FP8 LORA Improvements: A major update in this release is the enhancement of FP8 LORA application. This update is a breaking change, as LORAs now produce results nearly identical to those from 16-bit weights when applied to FP8 weights. While existing workflows may need some adjustments, the boost in quality and consistency is a significant upgrade.
  • Experimental FP8 Matrix Multiplication Support: For those seeking to maximize performance, we’ve introduced experimental support for FP8 matrix multiplication. This feature, activated with the new --fast argument, greatly accelerates float8_e4m3fn inference on NVIDIA 4000/ADA GPUs, though with a slight trade-off in quality. To try this feature (NVIDIA only):
  1. Install PyTorch 2.4.
  2. Run Comfy with python main.py --fast (if using the experimental standalone package, edit the .bat file to include the --fast argument).
  • Embracing LLM Quantization Techniques: We also want to spotlight the ComfyUI-GGUF project by city96, which showcases the application of Large Language Model (LLM) quantization techniques to image models.

Conclusion

Dev Team encourage all users to update to the latest version and explore these new features. Your feedback is invaluable, and we can’t wait to see the incredible creations you’ll produce with ComfyUI v0.1.0!

Read related articles: