Google opens its SynthID Detector for AI-generated media to the public, but an account is required

Google is widening access to its SynthID Detector, a tool that checks whether an image, video or audio file carries the invisible SynthID watermark. In a post on Google's blog, Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind, says the Detector is now open to everyone and available worldwide in English from the day of the announcement. Until now it was aimed at media professionals: Google says it introduced an early version last year, and heise reports that registered media representatives have been able to use it for some months.

According to Google, the Detector recognises content made with its own AI tools and with those of partners that also use SynthID. The company names OpenAI, NVIDIA and Kakao, and says Apple will be added soon. SynthID itself has existed since 2023: Google embeds the marks when its models generate media, and heise explains that they are meant to stay invisible to the human eye while remaining readable by machines. By Google's count, more than 180 billion images and videos and 240,000 years of audio have carried the watermark so far.

heise qualifies the claim that the tool is open to everyone. To use the portal, people need a Google, Apple or ChatGPT account, so anyone who does not want to sign up, whether for privacy or other reasons, still cannot run a check. heise says it asked Google whether this is set to change and had not received an answer at the time of its report.

The standalone Detector adds to verification functions that Google has already built into Search, the Gemini app and Chrome. Google says these features now regularly handle more than 1 million requests per day. heise adds that a number of third-party services also claim to be able to read the SynthID watermark, and that they differ in their terms of use and costs.

heise writes that watermarks are considered among the best options currently available for spotting AI content, but that they are not fully protected against manipulation or later removal. Other AI companies use similar techniques: according to heise, Anthropic gives its Claude models a text watermark, and OpenAI has recently started to add watermarks to text as well, mostly on an optional basis. In heise's view, the advantage of SynthID is that it is relatively widespread in the market.