Google Opens Cross-Vendor AI Watermark Detection: ChatGPT Images Can Also Be Checked

动察 Beating AI News Flash: Google is opening SynthID Detector to the global public. This English website can check whether images, videos, and audio contain invisible watermarks left when AI generated them. Previously mainly tested with media professionals, it now also supports identifying watermarks from OpenAI, Nvidia, and Kakao, with Apple also set to join.

OpenAI actually began adding SynthID to generated images as early as May this year, and expanded it to some audio in July. In the past, Google's detection tool could only check its own watermarks, but now multiple vendors' content can be checked through the same entry point.

Google says SynthID has watermarked more than 180 billion images and videos, with total audio duration equivalent to 240,000 years. The new website requires login, and each user has about 10 detection credits per day.

However, what is checked here is the SynthID watermark, and it cannot identify all AI content. Failure to detect a watermark also cannot prove that the content was genuinely captured or manually created. Content generated by AI models that do not use this kind of watermark may also be completely undetectable.

Detection address: https://synthid.com/