Google Opens Global SynthID Detector for Partner AI Watermarks
Google opens SynthID.com worldwide, detecting partner AI watermarks across media, but login quotas and unsupported models leave major gaps in verification.
Summary
Google opened SynthID.com globally on October 7, 2026, letting anyone check images, video, and audio for invisible signals embedded in pixels or waveforms. The expanded detector recognizes distinct SynthID watermarks from Google, OpenAI, Nvidia, Kakao, and soon Apple. Gemini has applied SynthID to more than 180 billion images and videos and 240,000 years of audio. Previously, people outside Google’s trusted tester program had to ask Gemini, which could return confusing answers, while Google’s detector missed partner watermarks, including those in ChatGPT images.
Users must sign in with a Google, OpenAI, or Apple account. The website reports whether supported SynthID technology helped create or edit a file, but unlike Google’s internal tool, it does not highlight marked regions. Each account receives approximately 10 daily image, video, and audio checks, with earlier lockouts possible for repeated similar files to deter watermark removal research. SynthID cannot detect Meta’s different watermarking, systems using no watermark, or locally run open-weight models that omit even basic metadata, and its marks may eventually be stripped. Google’s Pixel phones are among the few devices using C2PA cryptography to authenticate real media at capture, an alternative to labeling synthetic content.
Positives
- SynthID.com combines watermark checks for Google, OpenAI, Nvidia, Kakao, and soon Apple in one globally available tool.
- More than 180 billion Gemini images and videos already carry SynthID, alongside 240,000 years of generated audio.
- Google, OpenAI, and Apple accounts can access checks without joining Google’s trusted tester program.
- C2PA support on Google Pixel phones authenticates real media when captured rather than relying solely on synthetic-content labels.
Risks & concerns
- Approximately 10 checks per day restrict intensive verification, with similar files potentially triggering an earlier lockout.
- Meta’s separate watermarking technology remains outside SynthID detection.
- Unwatermarked services and locally operated open-weight models can produce realistic media that SynthID cannot identify.
- SynthID.com omits the internal tool’s ability to highlight image regions containing watermark signals.
- Possible SynthID removal methods could allow previously marked AI media to evade detection.