SDNext Update for 2024-05-28 - eviltoast
GitHub - vladmandic/automatic: SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models
github.com
SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models - vladmandic/automatic

Highlights for 2024-05-28

New SD.Next release has been baking in dev for a longer than usual, but changes are massive - about 350 commits for core and 300 for UI…

Starting with the new UI - yup, this version ships with a preview of the new ModernUI
For details on how to enable and use it, see Home and WiKi
ModernUI is still in early development and not all features are available yet, please report issues and feedback
Thanks to @BinaryQuantumSoul for his hard work on this project!

Screenshot-ModernUI Screenshot-ModernUI-Img Screenshot-ModernUI-Control

What else?

New built-in features

  • PWA SD.Next is now installable as a web-app
  • Gallery: extremely fast built-in gallery viewer
    List, preview, search through all your images and videos!
  • HiDiffusion allows generating very-high resolution images out-of-the-box using standard models
  • Perturbed-Attention Guidance (PAG) enhances sample quality in addition to standard CFG scale
  • LayerDiffuse simply create transparent (foreground-only) images
  • IP adapter masking allows to use multiple input images for each segment of the input image
  • IP adapter InstantStyle implementation
  • Token Downsampling (ToDo) provides significant speedups with minimal-to-none quality loss
  • Samplers optimizations that allow normal samplers to complete work in 1/3 of the steps!
    Yup, even popular DPM++2M can now run in 10 steps with quality equaling 30 steps using AYS presets
  • Native wildcards support
  • Improved built-in Face HiRes
  • Better outpainting
  • And much more…
    For details of above features and full list, see Changelog

New models

While still waiting for Stable Diffusion 3.0, there have been some significant models released in the meantime:

  • PixArt-Σ, high end diffusion transformer model (DiT) capable of directly generating images at 4K resolution
  • SDXS, extremely fast 1-step generation consistency model
  • Hyper-SD, 1-step, 2-step, 4-step and 8-step optimized models

Note
SD.Next is no longer marked as a fork of A1111 and github project has been fully detached
Given huge number of changes with +3443/-3342 commits diff (at the time of fork detach) over the past year,
a completely different backend/engine and a change of focus, it is time to give credit to original author, and move on!