EraseAnything (ICML 2025)

EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers is an open-source project for Concept Erasing in Rectified Flow models: e.g. Flux, SD 3.

Features

• ✅ Supports [diffusers]
• ✅ Easy to extend and integrate

Installation & Environment Setup

1. Install Rust (if required):

   curl https://sh.rustup.rs -sSf | sh
   export PATH="$HOME/.cargo/bin:$PATH"

2. Install Python dependencies:

   pip install transformers sentencepiece einops omegaconf
   pip install tokenizers==0.20.0
   pip install nltk wandb openai

3. Install local packages:

   cd peft
   pip install -e .[torch]
   
   or
   
   python setup.py install
   
   cd ../diffusers
   pip install -e .[torch]
   
   or
   
   python setup.py install

Quick Start

0. Set AI Secret Key (Strongly recommend OpenRouter!)

   API_KEY='xxx'
   END_POINT='https://research-01-02.openai.azure.com/'
   API_VERSION = "2024-08-01-preview"

1. Train the model:

   bash train.sh

2. Run inference:

   python inference.py

Training & Inference

• Training script: train.sh
• Inference script: inference.py
• Results will be saved in the results/ directory (if applicable).

Acknowledgments

This project is inspired by and builds upon the work of Erasing Concepts from Diffusion Models and other open-source projects. We thank the community for their valuable contributions.

Citation

If you use this project in your research, please cite:

@article{gao2024eraseanything,
  title={EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers},
  author={Gao, Daiheng and Lu, Shilin and Walters, Shaw and Zhou, Wenbo and Chu, Jiaming and Zhang, Jie and Zhang, Bang and Jia, Mengxi and Zhao, Jian and Fan, Zhaoxin and others},
  journal={ICML 2025},
  year={2024}
}

Contact

For technical questions, please contact samuel.gao023@gmail.com.