Fake videos created using AI are becoming more common. According to Sumsub, a platform for verifying identities, the number of deepfake videos increased four times between 2023 and 2024. These deepfakes now make up 7% of all fraud, including impersonations and social engineering scams.
To tackle this issue, Meta has introduced a new tool called Meta Video Seal. This tool adds invisible watermarks to AI-generated videos to help detect them and protect original content. It is open-source, meaning developers can add it to their own software. Meta Video Seal is part of Meta’s efforts to combat fake content, alongside other tools like Watermark Anything and Audio Seal.
Pierre Fernandez, a researcher at Meta, explained that Meta Video Seal is designed to handle challenges like video compression, which often happens when sharing videos online. Unlike other tools, it works well at scale and is specifically made for videos, not just images.
In addition to watermarks, Meta Video Seal can add hidden messages that reveal a video’s origin. The tool is resistant to common changes like cropping, blurring, and compression. However, heavy editing or extreme compression might affect the watermarks.
A challenge for Meta is convincing developers and companies to adopt this tool, especially those already using their own methods. To encourage adoption, Meta has launched a public leaderboard, Meta Omni Seal Bench, to compare different watermarking tools. They are also hosting a workshop on watermarking at a major AI conference this year.
Meta hopes researchers and developers will use watermarking tools like Video Seal to make it easier to detect and manage AI-generated content.
Source: techcrunch
