SD GUI 1.8.0


Changelog:

  • New: Stable Diffusion DirectML implementation, enables image generation on AMD GPUs
  • New: Inpainting models are now supported, providing much better quality than the old method
  • New: Models are cached in RAM when switching. Switching back loads a model in ~2 seconds
  • New: Text-based masking - Describe what you want to mask instead of drawing the mask
  • New: Initialization image can be previewed by Shift+Hovering over the filename in the GUI
  • New: Generation resolution is automatically set to init image size if it doesn't match
  • New: Pop-Up Image Viewer now has an option to enable Slideshow Mode by default
  • New: Model converter (Dev Tools) to convert between Pytorch, Diffusers, ONNX, Safetensors
  • New: Image Import window has a new option to use metadata plus the image itself as init image
  • Improved: Image Viewer now has hotkeys Ctrl+Left/Ctrl+Right for previous/next image
  • Improved: Image Viewer hotkeys (copy, open, etc.) now also work in pop-up viewer
  • Improved: Mask Editor can be resized, is no longer locked to the init image size
  • Improved: Mask Editor now has paste, invert, save, load, and undo functionality
  • Improved: Inpainting mask can now be edited without having to clear it first
  • Improved: Prompt History will not save a new entry if it's identical to the previous one
  • Improved: Prompt History now has a limit of 500 entries to avoid long loading times
  • Improved: Prompt Queue now behaves like a queue, not like a stack (FIFO)
  • Improved: Welcome Screen (MOTD) can now be disabled after it's been shown once per version
  • Improved: Image Import window now shows resolution, import actions are a dropdown to avoid button clutter
  • Improved: Mask Editor performance is better now, blur is only applied after releasing mouse
  • Improved: User can now ignore the Dreambooth VRAM warning, in case it was reported incorrectly etc.
  • Fixed: Post-processing after generation would cause a soft-lock
  • Fixed: Loading settings from prompt history sets incorrect values
  • Fixed: Broken or malicious model files would cause an unhandled exception
  • Fixed: Using a concept with <Name> syntax would show warning about missing placeholder
  • Fixed: InvokeAI CLI wouldn't open if GUI image generation was cancelled before
  • Fixed: Unhandled error when trying to use init images that no longer exist
  • Fixed: GUI would freeze if no model files exist in models folder(s)


Please do a clean re-install (don't overwrite an existing installation) for this update.

If you already use 1.7.x, you can download the NoMdl archive and copy Data/models from your existing installation.

Files

Stable Diffusion GUI 1.8.0 (Including model files) 3 GB
97 days ago
Stable Diffusion GUI 1.8.0 (No models included, provide your own!) 1 GB
97 days ago

Get NMKD Stable Diffusion GUI - AI Image Generator

Download NowName your own price