Authors: Jaehong Yoon*, Shoubin Yu*, Vaidehi Patil, Huaxiu Yao, Mohit Bansal

University of North Carolina at Chapel Hill

Setup

Install Dependencies

1. (Optional) Creating conda environment

conda create -n SAFREE python=3.10.13
conda activate SAFREE

2. build from source

pip install -r requirements.txt

Safe generation using SD-v1.4 by removing nudity concepts

bash scripts/run_nudity.sh

Safe generation using SD-XL by removing nudity concepts

bash scripts/run_nudity_sdxl.sh

COCO image generation using SD-v1.4 while removing nudity concepts

bash scripts/run_coco.sh

COCO image generation using SD-XL while removing nudity concepts

bash scripts/run_coco_sdxl.sh

Artist style image generation using SD-v1.4 while removing "Van Gogh" and "Kelly McKernan"

bash scripts/run_artist.sh

Acknowledgments

The code is built upon Prompting4Debugging, Diffusers, CogVideoX, and ZeroScopeT2V.

Reference

Please cite our paper if you use our models in your works:

@article{yoon2024safree,
  title={SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation},
  author={Yoon, Jaehong and Yu, Shoubin and Patil, Vaidehi and Yao, Huaxiu and Bansal, Mohit},
  journal={ICLR},
  year={2025}
}