Google opens SynthID AI detector to public worldwide

Google Alphabet Inc. ↘ 0.36% made its SynthID Detector available to the public on Wednesday. The free website, synthid.com, works worldwide in English and lets anyone upload an image, video or audio file to check it for the invisible watermarks Google and several partner companies put into AI-generated media. Before this, the tool was open only to journalists, researchers and other media professionals. The detector can read SynthID marks from Google's own models and from OpenAI, NVIDIA ↘ 0.80% and Kakao, and Google says support for Apple ↗ 0.57% is coming "soon".blog+1

"Now, we're expanding access to everyone – with the tool available globally in English starting today," Pushmeet Kohli, a vice president at Google DeepMind, wrote in the company's blog post.unite+1

How it works

SynthID adds a signal to content at the moment it is created. In images and video the mark sits in the pixels, and in audio it sits in the waveform, so it isn't lost the way metadata can be. Google says it has watermarked more than 180 billion images and videos and 240,000 years of audio since SynthID launched in 2023. It also says the verification features already built into Search, the Gemini app and Chrome now handle more than 1 million requests a day.arstechnica+2

According to Ars Technica, the new site fixes a problem with Google's earlier detector, which checked only for Google's own watermark. ChatGPT images carried SynthID but were not flagged. Users have to sign in with a Google, OpenAI or Apple account. A Google representative told Android Authority there are no other ways to log in. The public version gives a simple result saying whether a watermark was detected. Unlike the internal tool, it does not highlight which parts of an image carry the mark.androidauthority+1

Limits

The site's FAQ says plainly: "This is not a general AI detector". It finds content only from companies that have adopted SynthID. A "not detected" result does not rule out AI, because the file may have come from a model without the watermark, or heavy editing may have weakened the signal. The FAQ also notes rare cases where the detector falsely flags content that has no watermark.unite

Engadget reported that the tool cannot tell content made entirely by AI apart from content that was only edited with AI, and that watermarks can sometimes be stripped out. Detection still depends on a patchwork of standards. Meta ↘ 0.36% runs its own tool, which identifies images from its latest Muse models but not those from other companies.engadget

Google says it is "continuously working on extending our partnerships to allow detecting more generated media." ElevenLabs had already committed to supporting the watermarks. Google has not said when languages other than English will be added.goosed+1