Google DeepMind has made SynthID Detector globally available as a single place to check images, audio and video for invisible AI watermarks. The website supports media from Google and participating providers including OpenAI, Nvidia and Kakao, with Apple support planned.
Previously, users could ask Gemini whether Google-generated media carried a SynthID signal, but checking output from another provider required that provider’s own tool. The new site returns a yes-or-no result across participating systems.
Its coverage remains narrower than a general detector. Content from providers that use other watermarking systems, including Meta and Microsoft, is outside its scope. Heavy editing can also weaken a watermark and produce a false negative, so “no confident signal” does not establish that a file was made by a person.
SynthID embeds signals by changing patterns in pixels, audio frequencies or generated tokens. Provenance metadata such as C2PA offers a complementary record of origin and editing, but metadata can disappear during uploads or screenshots. Neither approach alone settles authenticity in every case.