Official implementation of Robust-Wide: Robust Watermarking Against Instruction-Driven Image Editing.

1. Download the data and put them into the data dir ./data.

2. Configure the train script and then run it.

bash train.sh

1. Put your original image in ./examples.

2. Download the checkpoints and put them in ./checkpoints.

3. Configure the inference script and then run it.

bash inference.sh

To reproduce the results presented in our paper, download the data and place it in the ./eval_data directory. Then, run:

bash evaluate.sh

This code builds on the code from the diffusers library.

If you find this repository useful, please consider giving a star ⭐ and please cite as:

@inproceedings{hu2025robust,
  title={Robust-wide: Robust watermarking against instruction-driven image editing},
  author={Hu, Runyi and Zhang, Jie and Xu, Ting and Li, Jiwei and Zhang, Tianwei},
  booktitle={European Conference on Computer Vision},
  pages={20--37},
  year={2025},
  organization={Springer}
}