Google opens SynthID detector to everyone, 1 million daily checks already

A website that accepts a photo, a video or a voice clip from a visitor has had no routine way to check whether the file was made by an AI model. Google said this week that it is opening SynthID Detector, the portal that checks whether a file carries one of its AI watermarks, to anyone, after an earlier period limited to journalists, researchers and media professionals.

TechCrunch, which reported the public launch, lists the file types the site now accepts: JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF and GIF for images; MP4, MOV and WEBM for video; and WAV, MP3, OGG, FLAC, AAC and M4A for audio.

What the detector reads

SynthID is Google's own watermark, introduced in 2023. The DeepMind product page says the watermark is written into content by Google's Nano Banana, Veo and Lyria models, and by any Google tool with generation inside it, including Gemini, Flow, ProducerAI and Vids. OpenAI, NVIDIA and Kakao also watermark with SynthID, and Google says Apple is adding support. The detector reads the watermark back out of a file and reports whether it is present.

Google says SynthID verification is already built into the Gemini app and into Google Chrome. On the DeepMind page the company says 1 million checks a day were being made during the earlier limited phase. The public opening adds to that existing use.

Where the detector stops

Google says the tool reports only on SynthID watermarks. The TechCrunch piece notes that watermark detectors routinely fail even on content created by their own makers' models, so a reading of no watermark from any one of them is weak evidence on its own. OpenAI runs its own site for checking files it generated. Microsoft and Meta use different watermark and provenance standards, and the DeepMind page does not list support for reading those. For a product serving uploads, SynthID Detector covers files produced by the four vendors Google names and no others.

For a team moving user-generated content through moderation, the public opening gives a tool a human reviewer can use in seconds, where before it was behind a request form that let in only journalists and researchers. The TechCrunch piece reports Google tested the tool with that group before the public opening. From today, the same check is a URL anyone can send a file to, which means the result of the check can be stored against the upload the moderator reviewed.

For a search team or a brand working on citations, the use is different but related. An answer engine that quotes a page, an image or a short audio clip is reading from both human and AI-made sources, and a verification route operated by a watermark vendor is one of the few ways to tell apart content produced by Google's or a partner's model from content that is not. A publisher that cares about being quoted correctly now has a way to check what their own AI-generated images look like to Google's own tool.

Google has not published a plan for other vendor watermarks, and the DeepMind page lists only the four providers above. The tool is a verification route for a specific set of outputs, and that is the limit of what it does.

Source

Primary source: SynthID on Google DeepMind. Report: Google's new SynthID website can identify AI-generated media on TechCrunch, by Ivan Mehta.