⚡️ PostEdit is both inversion- and training-free, necessitating approximately 1.5 seconds and 18 GB of GPU memory to generate high-quality results.

💥 PostEdit is accepted as a poster in International Conference on Learning Representations (ICLR) 2025!

Paper

This code was tested with Python 3.9, Pytorch 2.4.0 using pre-trained models through huggingface / diffusers. Specifically, we implemented our method over LCM. Additional required packages are listed in the requirements file. The code was tested on a single NVIDIA A100 GPU.

Download PIE-Bench dataset, and place it in your PIE_Bench_PATH.

Download the code:

git clone https://github.com/TFNTF/PostEdit.git

Download pre-trained models:

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained("SimianLuo/LCM_Dreamshaper_v7")

pip install -r requirements.txt
python main.py
(Optional) Save a specific image to "all_images" file for single image editing.

@article{DBLP:journals/corr/abs-2410-04844,
  author       = {Feng Tian and
                  Yixuan Li and
                  Yichao Yan and
                  Shanyan Guan and
                  Yanhao Ge and
                  Xiaokang Yang},
  title        = {PostEdit: Posterior Sampling for Efficient Zero-Shot Image Editing},
  journal      = {ICLR},
  year         = {2025},
}

We thank vivo for granting us access to GPUs.

If you have any questions, feel free to contact me through email ([email protected]). Enjoy!