Diffusion Digest: May 2024 ComfyUI Updates

Welcome to the first issue of DiffusionDigest, your go-to source for the latest updates and improvements in the world of stable diffusion and generative AI. In this inaugural newsletter, we'll be focusing on the exciting updates to ComfyUI that were released in May 2024.

πŸ†• New Features and Support

  • Webcam Node: You can now capture live webcam feed directly within ComfyUI! Find the "Webcam Capture" node under the "Image" category. Remember to allow webcam access and refresh the page to deactivate the webcam when you're done.

  • SDXS Model Support: ComfyUI now supports the SDXS model, a real-time one-step latent diffusion model with image conditioning. Download the UNET, text encoder, and VAE files to get started. Learn more about SDXS

  • Additional Model Updates: Several new models have been added, including the Dream Shaper model (base, sketch, and anime versions). Explore SDXS models

🎨 User Experience Enhancements

  • Clear Canvas Shortcut: The shortcut to clear the canvas has been changed from Ctrl+Delete to Ctrl+Backspace on Windows.

  • Reset View Button: A new "Reset View" button has been added to help center and zoom the workflow canvas.

  • Improved Zoom Control: You can now use Alt+Plus and Alt+Minus to zoom in and out of the canvas without affecting the entire browser.

  • Mask Editor Improvements: The default mask editor now includes an opacity slider for better visibility while painting.

  • Node Widget Conversion: When converting node widgets to inputs or vice versa, you'll now see a submenu listing all the available options.

  • LoadLora Node Range: The range for the LoadLora node slider has been expanded from -100 to +100.

πŸš€ Performance Improvements

  • Latent2RGB Preview Speed: The latent2RGB preview method has been optimized for even faster performance.

  • Image Sharpen and Blur Nodes: These default nodes have been improved, resulting in noticeable speedups.

  • Upscale Model Memory Allocation: ComfyUI now automatically allocates all free memory based on available VRAM for upscale models.

  • Low VRAM Detection: ComfyUI will now set the VRAM state to "normal" by default, even on low-end systems. Adjust the settings if needed.

πŸ› οΈ Optimization and Compatibility

  • Intel GPU Support: Significant improvements have been made to Intel GPU support, with full integration and setup instructions available in the GitHub repository. Set up ComfyUI with Intel GPUs

  • Enhanced System Logging: The terminal now displays more useful logs, including missing nodes, broken workflows, and relevant system information.

πŸ› Bug Fixes

  • Black Image Fix for Mac Users: The issue causing black images during generation on Mac has been resolved.

  • ControlNet Memory and Precision Fixes: ControlNet models should now load with the correct memory usage and precision level on low-end systems.

  • NumPy Version Consistency: The download package has been updated to ensure the correct NumPy version is always installed.

  • JPEG and PNG Loading Issues: Problems with loading JPEG and PNG images have been addressed.

That's it for this month's DiffusionDigest! Stay tuned for more exciting updates and insights into the world of stable diffusion and generative AI. If you have any questions, feedback, or suggestions for future topics, feel free to reach out.

Happy generating! The DiffusionDigest Team


or to participate.