Date:
December 13, 2024
While this is not a first-of-its-kind watermarking technology, it fills gaps in aspects like scaling where other tools are suboptimal.
AI-generated deepfakes have taken the world by storm. According to an ID verification platform, Sumsub, AI deepfakes multiplied fourfold from 2023 to 2024. The sophistication of AI-generated content and hyper-real output has evolved to escape the eyes of general users, accounting for nearly 7% of all online frauds. To strike a strong blow against rapidly growing AI Deepfakes, Meta launched a new AI tool that automatically watermarks AI-generated videos.
The technologies that AI deepfakes are increasingly being misused for nonconsensual impersonations, account takeovers, and social engineering campaigns of mass manipulation. With the launch of the Video Seal watermarking tool, Meta aims to bolster the fight against such entities while helping AI-generated content creators possessing genuine intent to thrive unrestrictedly.
The latest tool joins Meta’s existing watermarking inventory of tools, including Watermark Anything and Audio Seal. Watermark Anything was re-released yesterday alongside the Video Seal launch under a permissible license. Video Seal is available in open source and can be easily integrated with existing software.
“We developed Video Seal to provide a more effective video watermarking solution, particularly for detecting AI-generated videos and protecting originality,” said Meta’s AI Research Scientist, Pierre Fernandez, in an interview, “While other watermarking tools exist, they don’t offer sufficient robustness to video compression, which is very prevalent when sharing content through social platforms; weren’t efficient enough to run at scale; weren’t open or reproducible; or were derived from image watermarking, which is suboptimal for videos.”
Along with the watermarks, Video Seal can also embed a hidden message, which can help authorities uncover it later while determining their origins. The tool is resilient against surface edits like blurring, cropping, and compression algorithms from popular platforms. However, heavy compression and significant editing efforts may alter the video enough to make the watermarks ineffective on mainstream media publishing platforms.
As many proprietary AI watermarking tools from top tech giants already exist, adopting Meta’s freshly introduced open-source solution may face roadblocks. To address them, Meta is launching Omni Seal Bench, a leaderboard dedicated to comparing the performance and effectiveness of various watermarking methods. Meta says that it will also propagate the adoption through workshops that help users learn the best use of Video Seal along with its differentiators from existing watermarking solutions compatible with the top AI video generators.
