Deepfakes have become an increasingly prevalent issue in the digital age. With the commoditization of generative AI, the online landscape is flooded with fake content. According to ID verification platform Sumsub, there has been a remarkable 4x increase in deepfakes worldwide from 2023 to 2024. In 2024 alone, deepfakes accounted for 7% of all fraud, ranging from impersonations and account takeovers to sophisticated social engineering campaigns.
Meta's Contribution to the Fight Against Deepfakes
In an effort to make a meaningful impact, Meta is introducing a tool called Meta Video Seal. This open-source tool is designed to be integrated into existing software and aims to apply imperceptible watermarks to AI-generated video clips. Announced on Thursday, Meta Video Seal joins Meta's other watermarking tools, Watermark Anything (re-released today under a permissive license) and Audio Seal.Providing an Effective Video Watermarking Solution
AI research scientist Pierre Fernandez at Meta explained, "We developed Video Seal to offer a more effective video watermarking solution, especially for detecting AI-generated videos and safeguarding originality." Video Seal not only adds a watermark but also includes a hidden message that can later be uncovered to determine the video's origins. Meta claims that Video Seal is resilient against common edits such as blurring and cropping, as well as popular compression algorithms.However, it's important to note that Video Seal has its limitations. The trade-off between the perceptibility of the watermarks and their overall resilience to manipulation is a key consideration. Heavy compression and significant edits may alter or render the watermarks unrecoverable.
Comparing with Other Technologies
While DeepMind's SynthID can also watermark videos and Microsoft has its own video watermarking methodologies, Fernandez asserts that many existing approaches fall short. "While other watermarking tools exist, they don't offer sufficient robustness to video compression, which is very common when sharing content through social platforms. They weren't efficient enough to run at scale, weren't open or reproducible, or were derived from image watermarking, which is suboptimal for videos," Fernandez said.This highlights the unique features and advantages of Meta Video Seal in the fight against deepfakes.
Addressing Adoption Challenges
The bigger problem facing Video Seal is that developers and the industry may not have much incentive to adopt it, especially those already using proprietary solutions. To address this, Meta is launching a public leaderboard, Meta Omni Seal Bench, dedicated to comparing the performance of various watermarking methods. Additionally, Meta is organizing a workshop on watermarking this year at ICLR, a major AI conference.Fernandez emphasized, "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 make faster progress in this field."
