Web-based UI for Stable Diffusion by Zuiho

Forked from stable-diffusion-webui

There are mainly changed points:

1. Use virtual environment to replace Anaconda. The default CUDA version of PyTorch is 11.6 (compatible with 11.3 to 11.7) and you could modify it in the line 117 of webui.sh.
2. Fix gpu switching bugs. Now you could choose other gpus in your machine freedomly by change the environment variable of CUDA_VISIBLE_DEVICES in the line 123 of webui.sh.
3. Only optimize for streamlit. The default listening address is set to 0.0.0.0 and default port is set to 18888, you could modify them in the line 123 of webui.sh.
4. Default configurations are optimized for max efficiency (12GB to 40GB VRAM usage for the model without pruned).

BibTeX

@misc{rombach2021highresolution,
      title={High-Resolution Image Synthesis with Latent Diffusion Models}, 
      author={Robin Rombach and Andreas Blattmann and Dominik Lorenz and Patrick Esser and Björn Ommer},
      year={2021},
      eprint={2112.10752},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

@software{Li_Web-based_UI_for_2022,
      author = {Li, Guanghui},
      doi = {10.5281/zenodo.7228428},
      month = {10},
      title = {{Web-based UI for Stable Diffusion with GPU selector}},
      url = {https://github.com/sxlllslgh/stable-diffusion-webui-Zuiho},
      version = {1.0.0},
      year = {2022}
}