Video Seal: Meta’s New Weapon Against Deepfakes | WhatsApp Meta AI | Meta AI app | Meta WhatsApp | Turtles AI

The uncontrolled spread of fake AI-generated content, particularly deepfakes, is a growing challenge. Meta introduces Video Seal, an innovative, open source tool for adding imperceptible watermarks to videos, making them identifiable and secure against manipulation.

Key points:

• Exponential increase in deepfakes: From 2023 to 2024, deepfakes quadrupled, accounting for 7 percent of global fraud.
• Meta Video Seal: An advanced tool that integrates invisible watermarks into AI-generated videos.
• Robust technology: Resists compression, blurring, cropping and other common manipulations.
• Collaboration and transparency: Video Seal is open source and comes with a public ranking to compare existing technologies.

The explosion of automated content generation, aided by the increasing accessibility of AI tools, has created fertile ground for the rise of deepfakes, extremely sophisticated forgeries that now pose a threat not only to digital security but also to the authenticity of information. According to data from Sumsub, an identity verification platform, from 2023 to 2024, deepfakes grew fourfold to account for 7 percent of global fraud, including account takeovers and sophisticated social scams.

To counter this drift, Meta announced a major breakthrough: Video Seal, an open source tool designed to integrate invisible watermarks into AI-generated videos. The system is designed with the goal of making original content traceable and providing a barrier against digital manipulation. Compared to other existing solutions, such as DeepMind’s SynthID and Microsoft’s technologies, Video Seal aims to address obvious shortcomings: poor resistance to compression, unsuitability for large-scale use, and lack of transparency.

Pierre Fernandez, an AI researcher at Meta, emphasized the effectiveness of Video Seal in protecting videos from alteration. Among its main features, the system is able to include not only a strong watermark, but also a hidden message that allows the origin of the content to be traced. In addition, the technology is designed to be robust against blurring, cropping, and compression algorithms widely used on social platforms. However, Fernandez admitted that technical challenges remain, such as the trade-off between watermark visibility and resistance to heavier manipulations. Drastic compressions or significant modifications can indeed reduce the effectiveness of the system.

To boost Video Seal adoption in the industry, Meta has launched Meta Omni Seal Bench, a public ranking that compares the performance of different watermarking solutions available, promoting virtuous competition. In addition, the company will organize a workshop on watermarking at ICLR, one of the most prestigious conferences in the field of AI, with the aim of stimulating collaboration between academics and developers.

With initiatives like this, Meta aims not only to consolidate the reliability of technologies against fraudulent content, but also to foster a more cohesive and transparent community to more effectively address the challenges of an era dominated by AI-generated media.