Google launches SynthID Detector for AI-generated content

Google launched SynthID Detector at its I/O conference, a verification portal that scans uploaded media for the invisible SynthID watermarks the company embeds in output from its Gemini, Imagen, Veo and Lyria models. Users upload an image, audio clip, video or block of text; the tool scans for the watermark and, where found, highlights which portions of the file appear to carry it — flagging specific segments of audio or regions of an image.

Google said more than ten billion pieces of content had already been watermarked with SynthID by launch. Access to the detector began with early testers, with journalists, media professionals and researchers able to join a waitlist for broader use rather than receiving it immediately. Alongside the detector, Google said it was open-sourcing its SynthID text-watermarking method for developers and partnering with Nvidia and GetReal Security to extend detection beyond Google’s own tools.

The launch addressed a limitation critics had raised about invisible watermarking schemes generally: that a watermark is only useful for provenance if there is an accessible way to check for it. SynthID Detector answered that for Google’s own ecosystem, though it could only confirm the presence of Google’s watermark specifically — it offered no way to identify content generated by other companies’ models, and, like any watermarking scheme, did nothing for content that had never been watermarked in the first place.