FreeDiff: Progressive Frequency Truncation for Image Editing with Diffusion Models

This is the official implementation of the ECCV 24 paper "FreeDiff: Progressive Frequency Truncation for Image Editing with Diffusion Models".

A more detailed introduction of this project(This readme file), more examples of editing of different types of editing in a ipynb file, our results and the datasets will be released gradually.

Usage

Requirements

We implement our method with a similar code structure to Prompt-to-Prompt. The code runs on Python 3.10.12 with Pytorch 2.1.0 and Diffusers 0.27.2. We believe that mild version alterations of Pytorch and Diffusers will not affect the code much.

Checkpoints

We mainly examine our method on public available pretrained stable diffusion models SD v1-4("CompVis/stable-diffusion-v1-4") and SD v1-5("runwayml/stable-diffusion-v1-5").

Acknowledgements

We thank the awesome research work Prompt-to-Prompt