The Chosen One: Consistent Characters in Text-to-Image Diffusion Models (Unofficial implementation)

[Dec 2024] Please note that this Repo is not actively maintained.

This repository contains the unofficial PyTorch implementation of the paper The Chosen One: Consistent Characters in Text-to-Image Diffusion Models, using the Diffuser framework.

(Note that I didn't carefully adjust the hyperparameters for generating the results above and they are still good enough.)

Getting Started

Installation and Prerequisites

Install diffuser 0.24.0.dev0:

git clone https://github.com/huggingface/diffusers
cd diffusers
pip install .

Clone the repository and install the required packages:

git clone git@github.com:ZichengDuan/TheChosenOne.git
cd TheChosenOne
pip install -r requirements.txt

You also need to modify your configuration file in config/theChosenOne.yaml to fit your local environment.

Data backup folder preparation

You need to create a backup data folder to store the initial images generated in the first loop for faster training start up next time if you want to train on the same character again. This is set up in the configuration file as follows:

backup_data_dir_root: Your absolute path to the data folder

Run the codes

Training

python main.py

Inference

Simply run:

python inference.py

The script will load the model you designated in the inference.py and your config file.

Citing the paper

Please always remember to respect the authors and cite their work properly. 🫡

@article{avrahami2023chosen,
  title={The Chosen One: Consistent Characters in Text-to-Image Diffusion Models},
  author={Avrahami, Omri and Hertz, Amir and Vinker, Yael and Arar, Moab and Fruchter, Shlomi and Fried, Ohad and Cohen-Or, Daniel and Lischinski, Dani},
  journal={arXiv preprint arXiv:2311.10093},
  year={2023}
}