stable-diffusion-configurator

A utility for automatically downloading stable diffusion model weights from civit.ai and the web

Install

To install the utility, run the following commands:

wget https://raw.githubusercontent.com/SaladTechnologies/stable-diffusion-configurator/main/configure
chmod +x configure

Use

To use the utility, specify the models and the output paths for the models you want to download. This can be done with environment variables or command line arguments, or a mix. If both are present, the command line arguments take precedence. The utility will download the models from civit.ai and the web and save them to the specified paths. The utility will also download the extensions from the specified git urls and save them to the specified paths. The utility logs a JSON manifest of the downloaded models and extensions to the console.

Usage: ./configure [--options] [--help]
  --ckpt-path <path>: path where checkpoints should be saved. Env: CKPT_PATH
  --vae-path <path>: path where vae weights should be saved. Env: VAE_PATH
  --controlnet-model-path <path>: path where controlnet models should be saved. Env: CONTROLNET_MODEL_PATH
  --lora-path <path>: path where lora models should be saved. Env: LORA_PATH
  --extension-path <path>: path where extensions should be saved. Env: EXTENSION_PATH
  --civitai-version-ids <ids>: comma separated list of civitai version ids. Env: CIVITAI_MODEL_VERSION_IDS
  --ckpt-urls <urls>: comma separated list of urls to download checkpoints from. Env: CKPT_URLS
  --vae-urls <urls>: comma separated list of urls to download vae weights from. Env: VAE_URLS
  --controlnet-model-urls <urls>: comma separated list of urls to download controlnet models from. Env: CONTROLNET_URLS
  --lora-urls <urls>: comma separated list of urls to download lora models from. Env: LORA_URLS
  --extension-urls <urls>: comma separated list of git urls to download extensions from. Env: EXTENSION_URLS
  --sdxl-base: download SDXL base model. 1 for yes, 0 for no. Default is 0. Env: LOAD_SDXL_BASE
  --sdxl-refiner: download SDXL refiner model. 1 for yes, 0 for no. Default is 0. Env: LOAD_REFINER
  --help: print this message

Example, capturing the manifest

manifest=$(./configure --ckpt-path ./data/models/ckpt --civitai-version-ids 128713)
echo $manifest
# { "models": { "checkpoints": [ "dreamshaper_8.safetensors" ], "vae": [], "controlnet": [], "lora": [] }, "extensions": [], "sdxl": false }

Scripts and Dockerfiles

In this repo is a script ./build-baked that provides a consistent build interface for stable diffusion docker images that use this configuration utility. It is intended to paired with Dockerfile.baked to build docker images that include the models they need, and will not have to download them at runtime. These "baked" images are built from our dynamic images, which include the configuration utility and an entrypoint script. The build-baked script will build the docker image and tag it with the current git commit hash. There is an example entrypoint script that shows how to use the configuration utility to download the models and extensions. Your entrypoint script will need to be specific to the stable diffusion backend you are using.

./build-baked

Usage: ./build-baked [OPTIONS]

Options:
  --push                          Push the built image to DockerHub. Default: false
  --civitai-version-ids <v1,v2>   Comma-separated list of CivitAI model version IDs to load. Default: ""
  --load-refiner                  Load the refiner model. Default: false
  --load-sdxl-base                Load the SDXL base model. Default: false
  --controlnet-urls <url1,url2>   Comma-separated list of ControlNet download URLs. Default: ""
  --extension-urls <url1,url2>    Comma-separated list of extension Git URLs. Default: ""
  --ckpt-urls <url1,url2>         Comma-separated list of CKPT download URLs. Default: ""
  --vae-urls <url1,url2>          Comma-separated list of VAE download URLs. Default: ""
  --lora-urls <url1,url2>         Comma-separated list of LoRa download URLs. Default: ""
  --tag <tag>                     Tag to use for the image. Defaults to the CivitAI version IDs with a hyphen between them
  --image <image>                 Image to build. Default: saladtechnologies/sdnext
  --base-tag <tag>                Tag to use for the base image. Default: dynamic
  --base-image <image>            Base image to use. Default: saladtechnologies/sdnext
  --help                          Show this message and exit