With the rise of generative AI, the proliferation of false information has become a serious problem. The rapid development of deep forgery technology has caused network security to face huge challenges. To combat this threat, Meta has launched a new tool called Meta Video Seal, which aims to combat deepfake videos and protect content originality through imperceptible watermarking technology. This tool has been released as open source and combined with Meta’s previously released watermarking tool to form a more comprehensive anti-deepfake solution.
With the commercialization of generative artificial intelligence, the amount of false content online has increased dramatically. According to data from identity verification platform Sumsub, the number of global deepfakes will quadruple from 2023 to 2024. Deepfakes will account for 7% of all fraud in 2024, involving everything from impersonation and account takeovers to sophisticated social engineering campaigns.
To combat this problem, Meta has released a new tool, Meta Video Seal, designed to combat deepfakes by adding imperceptible watermarks to AI-generated videos. The tool, released as open source on Thursday, is designed to be easily integrated into existing software and complement Meta's other previously launched watermarking tools, Watermark Anything and Audio Seal.
“We developed Video Seal to provide a more effective video watermarking solution, specifically for detecting AI-generated videos and protecting originality,” Meta’s artificial intelligence research scientist Pierre Fernandez said in an interview.
Although Video Seal is not the first video watermarking technology proposed, DeepMind's SynthID and Microsoft's watermarking methods have also been used in the market, but Fernandez pointed out that existing watermarking tools generally have some shortcomings. He said: "Although there are other watermarking tools, they do not work well after video compression, which is very common when distributing content on social platforms; some methods also operate less efficiently and cannot be open or replicated; in addition, Many methods are derived from image watermarking, but image watermarking is not suitable for video."
In addition to the watermark function, Video Seal can also embed hidden information in videos, which can later be revealed to trace the source of the video. Meta claims that Video Seal can effectively resist common editing methods such as blurring and cropping, as well as commonly used compression algorithms.
However, Fernandez also acknowledged that Video Seal has certain limitations, particularly the trade-off between the perceptibility of the watermark and its resistance to tampering. He points out that excessive compression and heavy editing can destroy watermarks or render them unrecoverable.
A bigger challenge for Meta is the lack of incentives for developers and industry players to use Video Seal, especially those already using other proprietary solutions. To this end, Meta has launched a public ranking list called Meta Omni Seal Bench to compare the performance of different watermarking technologies, and will hold a seminar on watermarking technology at this year's major AI conference ICLR.
"We hope that more and more AI researchers and developers will incorporate some form of watermarking technology into their work, and we hope to work with industry and academia to drive faster progress in the field," Fernandez said.
The open source release of Meta Video Seal provides a new weapon to fight against deep forgery, but whether it can effectively curb the spread of false information still requires the test of time and the joint efforts of the industry. The continuous advancement and widespread application of technology will be the key to solving this problem.