SKeDA: A Generative Watermarking Framework for Text-to-video Diffusion Models

Researchers propose SKeDA, a new watermarking framework for text-to-video AI models that addresses content authenticity and copyright protection concerns. The system uses shuffle-key-based sampling and differential attention to maintain watermark robustness against video distortions while preserving generation quality.

• →SKeDA addresses growing concerns over AI-generated video content authenticity and copyright protection through advanced watermarking.
• →The framework solves frame alignment issues that plague existing image watermarking methods when applied to videos.
• →Shuffle-Key-based sampling transforms watermark extraction into permutation-tolerant aggregation, improving robustness against frame reordering.
• →Differential Attention enhances watermark reliability against temporal distortions and inter-frame compression.
• →Extensive experiments show SKeDA maintains high video generation quality while strengthening watermark durability.