Meta recently launched a new neural watermarking system called Video Seal, which aims to help identify AI-generated videos and effectively trace their origins even after editing. This watermark is invisible to viewers but can be detected to verify the video's authenticity.
Meta has released the entire system as open-source software under the MIT license, including the Video Seal watermark model, research papers, and training and inference code. Additionally, Meta has provided an interactive demo that allows users to test the technology themselves. Besides Video Seal, Meta also introduced two supporting tools: Omni Seal Bench, for comparing the performance of neural watermarks across different media types; and Meta Watermark Anything, which is also open-source under the MIT license.
Image Source Note: The image is AI-generated, provided by the image licensing service Midjourney.
Meta's initiative aims to address the growing issue of deepfake videos. With the advancement of AI technology, the barriers to creating deepfake videos have lowered, and their realism has increased, making it difficult for the public to distinguish between real and fake, posing serious threats to social order and personal privacy.
The advantage of Video Seal technology lies in its use of an advanced neural watermarking method that embeds watermark information directly into the video content itself, rather than simply adding a visual marker. This means that even if the video is cropped, resized, or compressed, the watermark can still be detected, effectively preventing forgers from tampering with the video's source.
The open-source strategy is another highlight of Meta's initiative. By making the technology open-source, Meta hopes to attract more developers to participate in the research and improvement of watermarking technology, collectively tackling the challenges posed by deepfake videos. It also aims to promote the adoption and application of this technology, contributing to the authenticity and security of the online environment.
These series of actions by Meta reflect the company's determination and responsibility in combating deepfake technology. However, whether this technology can truly be effective ultimately depends on its performance in real-world applications and its ability to maintain a competitive edge against constantly evolving forgery technologies. We look forward to Meta's Video Seal technology making a contribution to purifying the online space.
