Video Seal introduces a comprehensive framework for neural video watermarking, jointly training an embedder and an extractor to ensure watermark robustness. The model applies transformations such as video codecs during training to enhance robustness against distortions like flipping and blurring. The approach includes temporal watermark propagation, converting image watermarking models to efficient video watermarking models without watermarking every high-resolution frame. Experimental results demonstrate the model's effectiveness in terms of speed, imperceptibility, and robustness, outperforming strong baselines under challenging distortions. The codebase, models, and a public demo are open-sourced under permissive licenses to foster further research and development.