Google SynthID Doubles Down on AI Watermarking with Video and Text Support

Google SynthID Watermark Video and Text
In Short
  • Google has expanded the capabilities of its SynthID watermarking system.
  • Having already supported image and audio, SynthID now embeds watermarks into videos and AI-generated text.
  • SynthID will watermark texts generated through Gemini apps, and videos created by Google's new AI video generator, Veo.

The Google I/O 2024 event saw a bunch of new AI features coming to Google’s suite of apps. Among the additions is an upgrade to SynthID, the company’s existing AI watermarking system. Adding to its detection of AI-generated images and audio, Google is expanding the use of SynthID to more media formats.

SynthID can now embed watermarks to videos and texts generated by AI. Google says the digital watermark is undetectable to the human eye and doesn’t compromise the quality of videos. The technique involves adding a watermark to into the pixels of every frame of a video.

The company says the watermarks will stay in place even after modifications to videos. This means, someone can crop a video, add filters, modify frame rates, and adjust color tones and the digital watermark will still remain detectable. With AI video generation slowly picking up pace, the purpose of such a system is to help people understand whether a video they’re viewing is an AI generation.

Google SynthID Watermark Text
Image Courtesy: Google DeepMind

As for text-based content, SynthID will add a watermark while generating texts in real-time. The system will start by predicting the next most likely set of text to generate for a sequence and then assign probability scores to a possible character, word or part of a phrase. This action is repeated throughout an entire text generation, and its combined probability score is considered as the watermark.

Google SynthID Availability

Google confirms that SynthID will start watermarking texts generated on the Gemini mobile app and web app. It’s also embedding digital watermarks to videos generated on Google’s new Veo AI Video Generator. The company shares that SynthID can also be combined with other AI detection techniques for wider coverage. To help prevent misuse of AI tools, Google plans to open-source SynthID to make it useable across a wider AI ecosystem.

What do you think about Google’s upgraded SynthID system? Do you think this tool can stop spreading misinformation and help people make informed decisions about AI-generated content? Let us know in the comments below.

comment Comments 0
Leave a Reply