Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Throw a rock and you’re likely to hit deep fraud. The commercialization of generative artificial intelligence has led to an absolute explosion of fake content online: according to the identity verification platform Sumsub, the number of deepfakes worldwide has quadrupled from 2023 to 2024. In 2024, according to Sumsub, deep fraud accounted for 7% of all fraud. , from impersonations and account hijacking to sophisticated social engineering campaigns.
In what it hopes will be a meaningful contribution to the fight against deepfakes, Meta is releasing a tool to apply invisible watermarks to AI-generated video clips. The tool, called Meta Video Seal, announced Thursday is open source and designed to be integrated into existing software. The tool joins Meta’s other watermarking tools, Watermark Anything (re-released today under a permissive license) and Audio Seal.
“We developed Video Seal to provide a more effective video watermarking solution, especially for detecting and preserving authenticity of AI-generated videos,” Pierre Fernandez, AI research scientist at Meta, told TechCrunch.
Video Seal is not the first technology of its kind. DeepMind’s SynthID can watermark videos and Microsoft have their own video watermarking methodologies.
But Fernandez argues that many existing approaches fail.
“While other watermarking tools exist, they do not offer sufficient robustness to video compression, which is common when sharing content via social platforms; they were not efficient enough to work at scale; open or not repeated; or derived from image watermarking that is not optimal for videos,” Fernandez said.
In addition to watermarking, Video Seal can add a hidden message to videos that can later be revealed to identify their origin. Meta claims that Video Seal is resistant to common edits such as blurring and cropping, as well as popular compression algorithms.
Fernandez acknowledges that Video Seal has some limitations, mainly a trade-off between how perceptible the tool’s watermarks are and their overall resistance to manipulation. Heavy compression and significant editing can change watermarks or make them unrecoverable.
Of course, the bigger problem Video Seal faces is that developers and the industry, especially those already using proprietary solutions, won’t have much reason to adopt it. To address this, Meta is launching the Meta Omni Seal Bench, a public leaderboard dedicated to comparing the performance of different watermarking methods, and is hosting a watermarking workshop at ICLR, a major AI conference this year.
“We hope that more and more AI researchers and developers will incorporate some form of watermarking into their work,” Fernandez said. “We want to collaborate with industry and the academic community to make faster progress in this area.”
