diffdiff

Tools for diffing huggingface diffuser models

Usage

Fast example

Generate images with different models

python -m diffdiff generate \
    --versions "CompVis/stable-diffusion-v1-4 stabilityai/stable-diffusion-2" \
    --prompts "Xenomorph birthday party & Escher style pencil drawing of octopus bicycle"
    --save_dir weird_images

Run Gradio UI to compare images

python -m diffdiff ui --save_dir weird_images

Generation

python -m diffdiff generate
    --prompts <PROMPTS> \
    --versions <VERSIONS> \
    --config <CONFIG> \
    --save_dir <SAVE_DIR> \
    --sep <SEP>

• Supported <PROMPTS>

  • a list of prompts separated by <SEP>
  • path to text file, each line is a prompt

• <VERSIONS>="CompVis/stable-diffusion-v1-4"

Model versions (separated with space) to run, at minimum one version.

• <CONFIG>

  • a string that can be interpreted as Python dictionary
  • a path to config file

• <SAVE_DIR>=images Directory where we will save images

• <SEP>="&"

Separator to delineate prompts.

Configuration

See stable diffusion pipeline documentation