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, while being the fastest diffusion algorithm. It's made by retro-engineering Stable-Diffusion, ComfyUI and SDWebUI.
There's also a LightDiffusion version using Flux Dev available in releases and Flux branch.
⚠️ The Repository and all future updates will be made on the LightOrg repo here
For more information, check the documentation
The project grew up with time, leaving the purpose of only being the fastest and simplest SD1.5 image generator, if you seek the original version of about 3000 lines of python, you can still find it at LightDiffusion-og or in the releases here.
Also check out the Refactored version for improved simplicity and assembling SD and Flux at LightDiffusion-Next
- Original Txt2Img, Img2Img (R-ERSGAN4x+ UltimateSDUpscaling DPM++ 2M)
- Attention syntax
- Hires-Fix (euler ancestral normal)
- Nvidia, amd and Intel gpus and cpu
- 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)
- TAESD image previewing at the cost of a 10% performance hit
To install, please download a release or clone this repo and execute the run.bat file in a cmd and you should be good to go.
Else install the python dependencies by writing pip install -r requirements.txt
After doing that, add your SD1/1.5 safetensors model to the checkpoints directory and you should be good to go.
To use the stable-fast optimization refer to this guide to install from the wheel file.
To use the Prompt enhancer refer to this guide to install and run those commands
ollama run llama3.1
pip install ollama
To use LightDiffusion in discord, refer to the installation guide in this repo