Safe-Sora: Safe Text-to-Video Generation
via Graphical Watermarking

Zihan Su¹, Xuerui Qiu², Hongbin Xu³, Tangyu Jiang¹, Junhao Zhuang¹,
Chun Yuan^1†, Ming Li^4†, Shengfeng He⁵, Fei Richard Yu⁴

¹ Tsinghua University ² Institute of Automation, Chinese Academy of Sciences
³ South China University of Technology
⁴ Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ)
⁵ Singapore Management University
^†Corresponding Author

 

Release

• [09/19] 🚀 🚀 Code Released!
• [09/18] 🎉 🎉 Safa-Sora is accepted by NeurIPS 2025!
• [05/23] Initial Preview Release 🔥 Coming Soon!

🔆 Introduction

Safe-Sora is the first framework that integrates graphical watermarks directly into the video generation process.



The following results show the original video, the watermarked video, the difference between them (×5), the original watermark, the recovered watermark, and the difference between them (×5).

📋 File Preparation

Download the files and place them in the root directory.

• checkpoints contains the pretrained weights for Safe-Sora, VideoCrafter2, the VAE, and the 3D-CNN (simulating H.264 compression).
• dataset contains the Logo-2K dataset and the Panda-70M dataset.
• mamba is provided for setting up the environment.
• causal-conv1d is provided for setting up the environment.

💻 Requirements

To install requirements:

conda create -n safe-sora python=3.9
conda activate safe-sora
conda install pytorch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 pytorch-cuda=11.8 -c pytorch -c nvidia

pip install packaging ninja==1.11.1.1
cd causal-conv1d
python setup.py install
cd ../mamba
python setup.py install

cd ..
pip install -r requirements.txt

🐶 Training

To train Safe-Sora, run this command:

bash train.sh 

We use DDP to train Safe-Sora. You can modify the parameters in train.sh to select which GPUs to use.

🚀 Evaluation

To evaluate our models, run:

bash test.sh 

🙌🏻 Acknowledgement

Our code is based on these awesome repos:

• Mamba
• RainMamba
• FreqMamba

📖 BibTeX

If you find our repo helpful, please consider leaving a star or cite our paper :)

@article{su2025safe,
  title={Safe-Sora: Safe Text-to-Video Generation via Graphical Watermarking},
  author={Su, Zihan and Qiu, Xuerui and Xu, Hongbin and Jiang, Tangyu and Zhuang, Junhao and Yuan, Chun and Li, Ming and He, Shengfeng and Yu, Fei Richard},
  journal={arXiv preprint arXiv:2505.12667},
  year={2025}
}