Gwanhyeong Koo,
Sunjae Yoon,
Ji Woo Hong,
Chang D. Yoo
KAIST
- [07/03] Initial Preview Release 🔥 Coming Soon!
- [09/07] Official Release of Code 🔥 Available Now!
Current image editing methods using DDIM Inversion face challenges with non-rigid edits due to high-frequency components in DDIM latent, which hinder structural changes. We introduce FlexiEdit, which improves fidelity to input text prompts by refining these components in targeted areas. FlexiEdit features Latent Refinement for layout adjustments and Edit Fidelity Enhancement for accurate text prompt reflection, demonstrating superior performance in complex edits through comparative experiments.
It is recommended to run our code on a Nvidia GPU with a linux system. We have not yet tested on other configurations. Currently, it requires around 20 GB GPU memory to run our method.
To install the required libraries, simply run the following command:
git clone https://github.com/kookie12/FlexiEdit.git
cd FlexiEdit
conda env create -f environment.yml
conda activate flexiedit
To start using FlexiEdit, execute the following commands:
python run_flexiedit.py
Note: Within run_flexiedit.py, you can set parameters such as model_path, out_dir, and source_image_path.
To use a user-defined mask for edits, obtain coordinates using draw_bbox.py and input them into bbx_start_point and bbx_end_point. If you prefer an automatic mask, simply set these values to None.
Compatibility: FlexiEdit has been primarily tested on Stable Diffusion versions v1-4, but is also compatible with newer versions like v1-5 and v2-1. You can download these checkpoints from their official repository.
The repo is still being under construction, thanks for your patience.
- Release of code.
Our code is based on these awesome repos:
If you find our repo helpful, please consider leaving a star or cite our paper :)
@article{koo2024flexiedit,
title={FlexiEdit: Frequency-Aware Latent Refinement for Enhanced Non-Rigid Editing},
author={Koo, Gwanhyeong and Yoon, Sunjae and Hong, Ji Woo and Yoo, Chang D},
journal={arXiv preprint arXiv:2407.17850},
year={2024}
}