VideoMark: A Distortion-Free Robust Watermarking Framework for Video Diffusion Models

If our project helps you, please give us a star ⭐ on GitHub to support us. 🧙🧙

🔥 News

• 2025-08-10 🎉 We are happy to announce that we release our code.
• 2025-04-23 🌟 We released the paper VideoMark: A Distortion-Free Robust Watermarking Framework for Video Diffusion Models.

The overall framework of VideoMark.

---

🛠️ Requirements and Installation

Install Dependencies

Basic Dependencies:

• Python >= 3.10
• ··· Then run:

pip install -r requirements.txt

Generate a Watermark Key

python PRC_key_gen.py --hight 512 --width 512 --fpr 0.01 --prc_t 3

Watermark Embedding and Extraction

python embedding_and_extraction.py \
        --model_name i2vgen-xl \
        --num_frames 16 \
        --num_bit 512 \
        --num_inference_steps 50 \
        --output_dir <your save dir> \
        --keys_path  <your keys path>\

Robustness Test

python temporal_tamper.py
        --model_name i2vgen-xl \
        --num_bit 512 \
        --num_inference_steps 50 \
        --video_frames_dir <your dir> \
        --keys_path  <your keys path> \

---

📊 Video Quality Evaluation

To evaluate the quality of watermarked videos, you can perform both objective and subjective assessments.

🧪 Objective Evaluation with VBench

We recommend using VBench — VBench: Comprehensive Benchmark Suite for Video Generative Models

👁️ Subjective Evaluation

For subjective assessments, we provide sample videos and guidelines in the following folder: eval_quality:

cd eval_quality

🌟 Star History

📑 Citation

If you find VideoMark useful for your research and applications, please cite using this BibTeX:

@article{hu2025videomark,
  title={VideoMark: A Distortion-Free Robust Watermarking Framework for Video Diffusion Models},
  author={Hu, Xuming and Li, Hanqian and Li, Jungang and Liu, Aiwei},
  journal={arXiv preprint arXiv:2504.16359},
  year={2025}
}