Google opens its AI-content checker to everyone, with limits
Google has opened its SynthID Detector website to anyone, letting people upload images, video or audio to check for hidden AI watermarks. That gives regular users a way to test suspicious media before they share it. Each person gets about 10 checks a day, and the tool now reads watermarks from OpenAI, Nvidia and Kakao as well as Google.
Anyone can now check whether a photo, video or audio clip was made by some of the biggest AI tools. Google opened its SynthID Detector to the public at a new website, SynthID.com, Ars Technica reported on Wednesday.
SynthID is an invisible signal hidden in the pixels of AI images and video, and in the sound waves of AI audio. People cannot see or hear it, but software can detect it. Google introduced it in 2023 and puts it on media from its Gemini, Veo, Nano Banana and Lyria models, TechCrunch reported. Google says Gemini alone has marked more than 180 billion images and videos.
The big change is reach. Until now, outsiders could only check for SynthID by asking Gemini, and the old detector only spotted Google's own marks. The new site also reads watermarks from Google's partners, including OpenAI, Nvidia and Kakao, with Apple coming soon, Ars reported. Google says people already make 1 million checks a day through tools like the Gemini app and Chrome.
There are limits. Users must sign in with a Google, OpenAI or Apple account. Each person gets "approximately 10" checks a day. Google engineers told Ars the cap helps stop people from using the detector to build tools that strip the watermarks out.
The answer is also simple. The site tells you whether SynthID was used to make or edit the file. It does not highlight which parts were changed, as Google's internal tool did.
The bigger problem is what the tool cannot see. Meta and Microsoft use their own watermark systems. Some AI makers use none at all. And open models that people run on their own computers can create realistic images with no label. So a clean result does not prove a picture is real, and the site carries a disclaimer saying so, Ars noted.
One idea gaining ground works the other way: labeling real photos when they are taken, using a standard called C2PA. Google's Pixel phones already do this.
The site accepts common formats, including JPG, PNG and HEIC images, MP4 and MOV video, and MP3 and WAV audio.
What we know: SynthID.com is open to all, with about 10 checks a day.
What we don't: How much AI content in the wild carries no watermark the tool can find.
Where outlets agree Ars Technica and TechCrunch both report the public SynthID site checks images, video and audio and supports watermarks from OpenAI, Nvidia and Kakao, with Apple to follow.
Where they differ Google calls it a one-stop checker, but Ars notes Meta, Microsoft and many open models use other marks or none, so a clean result does not prove media is real.