Train DreamBooth with Lightning

Training a diffusion model with the DreamBooth method using Pytorch Lightning.

Setup

We recommend miniforge to set up your python environment. In case VSCode does not detect your conda environments, install nb_conda in the base environment.

conda env create -n $YOUR_ENV_NAME -f environment.yml
conda activate $YOUR_ENV_NAME
pip install -r requirements.txt
pre-commit install

Props

• jsonargparse
• diffusers
• transformers
• lightning

Intention

This repository is meant to simplify the train_dreambooth_lora.py script found in the Huggingface Dreambooth instructions (https://huggingface.co/docs/diffusers/training/dreambooth).

The training can be executed by running

python train_lightning.py (--config configs/train_dreambooth_lightning.yaml)