Meta launches a watermarking tool to combat deepfakes

The rise of generative AI has led to a surge in fake content online, with deepfakes becoming more common. Meta has introduced a new tool, Meta Video Seal, to help combat deepfakes by adding imperceptible watermarks to AI-generated videos. The tool is open-source and can be integrated into existing software. It joins Meta's other watermarking tools, Watermark Anything and Audio Seal.

According to reports, Pierre Fernandez, an AI research scientist at Meta, explained that the goal of Video Seal is to offer a more effective solution for detecting AI-generated videos and protecting originality. While other technologies, like DeepMind's SynthID and Microsoft's methods, exist, Meta believes theirs is more robust.

"Many existing watermarking tools fail to work well with video compression, are not scalable, or are not open or reproducible," Fernandez said in the report.

In addition to watermarks, Video Seal can add hidden messages to videos that reveal their origins. Meta claims the tool can withstand common edits such as blurring and cropping, as well as popular compression algorithms.

However, Fernandez acknowledged that the tool has limitations. The balance between how visible the watermark is and its resistance to manipulation can be tricky. Heavy compression or significant edits may distort or remove the watermark.

The bigger challenge, Fernandez notes, is convincing developers and the industry to adopt the tool, especially if they are already using proprietary solutions. To address this, Meta is launching a public leaderboard called Meta Omni Seal Bench to compare different watermarking methods. They are also hosting a workshop at the ICLR AI conference this year.

With this, the company hopes that more AI researchers and developers will integrate watermarking into their work.