With the proliferation of deepfake technology, the amount of false content online has surged. According to data from ID verification platform Sumsub, the number of deepfakes quadrupled from 2023 to 2024, accounting for 7% of all fraud in 2024. This includes various forms such as identity impersonation, account takeover, and complex social engineering attacks.
To address this challenge, Meta recently launched a new watermarking tool — Meta Video Seal, designed to add nearly imperceptible watermarks to AI-generated video clips. This tool has been open-sourced and is designed to integrate seamlessly with existing software. Meta Video Seal joins Meta's other watermarking tools, such as Watermark Anything and Audio Seal.
Meta's AI research scientist Pierre Fernandez stated in an interview with TechCrunch, “We developed Video Seal to provide a more effective video watermarking solution, especially for detecting AI-generated videos and protecting originality.”
Although Video Seal is not the first technology of its kind, with similar video watermarking technologies from DeepMind and Microsoft, Fernandez believes that some existing solutions have shortcomings. He pointed out, “Many watermarking tools lack robustness during video compression, making them unsuitable for widespread use on social platforms, and they lack openness and reproducibility.”
In addition to watermarking, Video Seal can also embed hidden information into videos for later identification of their source. Meta claims that this tool can withstand common editing operations such as blurring and cropping, as well as popular compression algorithms. However, Fernandez also acknowledged that there is a trade-off between the perceptibility of the watermark and the robustness against manipulation. Excessive compression and heavy editing could affect the watermark, potentially rendering it unrecoverable.
Of course, a bigger challenge for Video Seal is that developers and the industry may not be highly willing to adopt it, especially for companies already using proprietary solutions. To address this issue, Meta plans to launch a public leaderboard — Meta Omni Seal Bench, specifically designed to compare the performance of different watermarking methods, and will host a watermarking workshop at this year's ICLR conference.
Fernandez hopes that more AI researchers and developers will integrate some form of watermarking into their work, collaborating with industry and academia to advance this field.
Key Points:
🔍 The Meta Video Seal tool launched by Meta can add watermarks to AI-generated videos, resisting editing and compression.
📊 The tool is open-sourced, designed for integration with existing software, and aims to advance watermarking technology in the industry.
🏆 Meta will also launch a public leaderboard to compare different watermarking methods, promoting industry collaboration and communication.
