Skip to content

LightDiffusionV4.0

Choose a tag to compare

@Aatricks Aatricks released this 01 Oct 16:56
· 27 commits to main since this release
262246d

LightDiffusion

The purpose of this repository and project is to remake the famous stable-diffusion in only one python script, with the
least number of lines and in the least complex way. It's made by retro-engineering Stable-Diffusion, ComfyUI and
SDWebUI.

Features

  • Original Txt2Img, Img2Img (R-ERSGAN4x+ UltimateSDUpscaling DPM++ 2M)
  • Attention syntax
  • Hires-Fix (euler ancestral normal)
  • GPU only
  • Xformers and Pytorch optimization
  • Stable-Fast implementation offering a 70% speedup at the cost of pre inference model optimization windup time
  • FP16 and FP32 precision support
  • Saved state in between starts
  • GUI
  • DPM Adaptive Karras
  • Clip Skip
  • LoRa and textual inversion (embeddings) support
  • Automatic Prompt-Enhancing with llama3.1 (ollama)
  • Discord bot integration with the installation of Boubou and the usage of LightDiffusion's pipeline
  • Automatic Face and body Detailer (based on Impact pack)

Screenshot 2024-06-10 140130

Installation

To install, please place your SD1.5 model in _internal/checkpoints and execute the run.bat file and you should be good to go.

What's Changed

  • Auto detailer by @Aatrick in #4

Full Changelog: V3.0...V4.0