This is a Stable-Diffusion Web Application written using Streamlit, Pytorch/Torchvision and HuggingFace Diffusers With proper hardware, it can generate images locally from pre-trained Stable Diffusion Models.
To download model checkpoints, have a look at HuggingFace.
The application expects to find model checkpoints files in the models/stable-diffusion folder, and loras in models/loras folder.
$ uv run streamlit run main.py
It tries to use any discovered GPU that is supported by pytorch, but it can run on CPU (very slowly)
Tested:
- Apple Metal Performance Shaders
- NVIDIA CUDA
- AMD CPU
- SD1.5 Inference with optional LoRA support
- SDXL Inference with optional LoRA support
- SD1.5 Model Merging Capabilities
- Support for custom VAEs in both SDXL and SD1.5 inference phases
- SD1.5 and SDXL Inpainting