Meta Debuts Tool for Watermarking AI-Generated Videos
Meta Unveils Open-source Tool to Combat Deepfake Surge
Tech Giant Aims to Tackle Rising Tide of AI-Generated Content with Video Watermarking Technology
The proliferation of deepfakes is reaching alarming levels. A recent report by ID verification platform Sumsub revealed a staggering fourfold increase in deepfakes globally from 2023 to 2024, with these synthetic videos now accounting for 7% of all online fraud.
In response to this growing threat, Meta has announced the release of Video Seal, an open-source tool designed to embed imperceptible watermarks into AI-generated videos. This innovative technology aims to provide a more robust solution for detecting and authenticating video content in an era dominated by synthetic media.
“We developed Video Seal to provide a more effective video watermarking solution, particularly for detecting AI-generated videos and protecting originality,” explained Pierre Fernandez, AI research scientist at meta.
Video Seal joins Meta’s existing suite of watermarking tools, Watermark Anything (re-released today under a permissive license) and Audio Seal, further demonstrating the company’s commitment to combating the spread of misinformation.
While other watermarking technologies exist,Fernandez emphasized that Video Seal addresses key shortcomings of previous approaches.”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,” he said.
Video Seal not only embeds a watermark but also conceals a hidden message within the video,allowing for later verification of its origin. Meta claims the technology is resilient against common video edits like blurring and cropping, as well as popular compression algorithms.
However, Fernandez acknowledged that Video seal, like any watermarking technology, faces limitations.
“We admit that Video Seal has certain limitations, mainly the trade-off between how perceptible the tool’s watermarks are and their overall resilience to manipulation,” he said. “Heavy compression and significant edits may alter the watermarks or render them unrecoverable.”
To encourage wider adoption of watermarking technology, Meta is launching a public leaderboard, Meta Omni Seal Bench, to compare the performance of various watermarking methods. The company is also organizing a workshop on watermarking at the prestigious ICLR AI conference later this year.
“We hope that more and more AI researchers and developers will integrate some form of watermarking into their work,” Fernandez said. “We want to collaborate with the industry and the academic community to progress faster in the field.”
Fighting the Fake News Flood: An Interview with Meta’s Pierre Fernandez on Video Seal
The rise of deepfakes is a growing concern, with synthetic videos now accounting for 7% of online fraud globally. In an effort to combat this trend, Meta has launched Video Seal, an open-source tool designed to embed imperceptible watermarks into AI-generated videos. we spoke with Pierre Fernandez, AI research scientist at Meta, to learn more about this innovative technology.
NewsDirectory3: What drove Meta to develop Video Seal?
Pierre fernandez: We developed Video Seal to provide a more effective video watermarking solution,especially for detecting AI-generated videos adn protecting originality.
NewsDirectory3: How does Video Seal address limitations of existing watermarking technologies?
Pierre Fernandez: 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.
NewsDirectory3: Can you elaborate on the technology behind Video seal?
Pierre Fernandez: Video Seal not onyl embeds a watermark but also conceals a hidden message within the video, allowing for later verification of its origin. We designed it to be resilient against common video edits like blurring and cropping, as well as popular compression algorithms.
NewsDirectory3: What are the limitations of Video Seal?
Pierre Fernandez: We admit that Video Seal has certain limitations, mainly the trade-off between how perceptible the tool’s watermarks are and their overall resilience to manipulation. Heavy compression and critically important edits may alter the watermarks or render them unrecoverable.
NewsDirectory3: What steps is Meta taking to encourage wider adoption of Video Seal?
Pierre Fernandez: We are launching a public leaderboard,Meta Omni Seal Bench,to compare the performance of various watermarking methods. we are also organizing a workshop on watermarking at the ICLR AI conference later this year. We hope that more and more AI researchers and developers will integrate some form of watermarking into their work. We want to collaborate with the industry and the academic community to progress faster in this field.