A library for training custom Stable Diffusion models (fine-tuning, LoRA training, textual inversion, etc.) that can be used in InvokeAI.

https://invoke-ai.github.io/invoke-training/

• Stable Diffusion
  • LoRA
  • DreamBooth LoRA
  • Textual Inversion
• Stable Diffusion XL
  • LoRA
  • DreamBooth LoRA
  • Textual Inversion
  • LoRA and Textual Inversion

More training modes coming soon!

For more installation details, see the Installation section of the documentation.

# A recent version of pip is required, so first upgrade pip:
python -m pip install --upgrade pip

# Editable install:
pip install -e ".[test]" --extra-index-url https://download.pytorch.org/whl/cu121

invoke-training pipelines can be configured and launched from either the CLI or the GUI.

Run training via the CLI with type-checked YAML configuration files for maximum control:

invoke-train --cfg-file src/invoke_training/sample_configs/sdxl_textual_inversion_gnome_1x24gb.yaml

Run training via the GUI for a simpler starting point.

invoke-train-ui

# Or, you can optionally override the default host and port:
invoke-train-ui --host 0.0.0.0 --port 1234

Training progress can be monitored with Tensorboard: Validation images in the Tensorboard UI.

All trained models are compatible with InvokeAI:

Example image generated with the prompt "A cute yoda pokemon creature." and a trained Pokemon LoRA.

Contributors are welcome. For developer guidance, see the Contributing section of the documentation.