Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Throw a rock and you’ll probably hit deep. The commercialization of AI output has led to an explosion of fake products on the Internet: According to the ID verification platform Sumsub, there will be a 4x increase in deep-pocketed products worldwide from 2023 to 2024. , from the acquisition of viewers and account hijacking to the most sophisticated engineering campaigns.
In what it hopes will help combat deepfakes, Meta is releasing a tool to apply invisible watermarks to AI-generated videos. Announced Thursday, the tool, called Meta Video Seal, is publicly available and designed to integrate with existing software. This tool integrates with Meta’s other watermarking tools, Watermark Anything (also released today under a licensed license) and Audio Seal.
“We created Video Seal to provide a better way to view videos, especially when it comes to identifying AI-generated videos and protecting their content,” Pierre Fernandez, AI research scientist at Meta, told TechCrunch in an interview.
Video Seal is not the first technology of its kind. Images of DeepMind SynthID can watermark videos, and Microsoft it has its own Video watermarking method.
But Fernandez says many of the options available are limited.
“Although other watermarking tools exist, they don’t provide enough power to embed videos, which are very common in sharing content on the web; they weren’t efficient enough to run at scale; they weren’t open source or reproducible; or they were based on watermarking, which is great for videos,” Fernandez said. .
In addition to a watermark, Video Seal can add a hidden message to videos that can later be revealed to identify its source. Meta says that Video Seal combats common edits like blurring and cropping, as well as popular algorithms.
Fernandez acknowledges that Video Seal has limitations, particularly the trade-off between the appearance of the tool’s watermarks and their robustness in editing. Too much pressure and too much change can change watermarks or make them unresponsive, he added.
Of course, the biggest problem Video Seal faces is that devs and industries will have no reason to expand, especially those who are already using its solutions. To address this, Meta is launching a community-led group, the Meta Omni Seal Bench, dedicated to comparing the performance of different watermarking methods, and organizing a watermarking conference this year at ICLR, a major AI conference.
“We hope that more AI researchers and developers will incorporate some watermark feature into their work,” Fernandez said. “We want to collaborate with companies and academics to make rapid progress in the field.”
