Google opens public SynthID detector for AI watermarks

Google’s public SynthID site checks partner watermarks in images, video, and audio, with login limits and over 180 billion items marked.

Google has opened a public SynthID detector, previously limited to journalists, media professionals, and researchers, so people can check whether an image, video, or audio file carries a watermark from Google or its partners. TechCrunch says the site accepts JPG, PNG, MP4, MOV, WAV, and MP3, while The Decoder lists only JPG, PNG, MP4, and MP3. Ars Technica says the improved worldwide tool requires a Google, OpenAI, or Apple login and limits users to about 10 checks a day so the service is harder to use for building SynthID-removal tools. Google says Nano Banana, Veo, Lyria, Gemini, Flow, ProducerAI, and Vids watermark output with SynthID, and that OpenAI, NVIDIA, and Kakao also support it, while Microsoft and Meta use their own approaches; The Decoder adds that Google expects Apple to join. Sources put daily verification volume at about one million—TechCrunch describes overall requests, and The Decoder attributes that figure to Search, the Gemini app, and Chrome—while The Decoder also reports more than 180 billion watermarked images and videos. Detection remains incomplete: Google says the tools can miss content from its own models, and Ars and The Decoder say unmarked media and many open-weight models will not be flagged; Ars also notes C2PA cryptographic labeling of authentic media on Pixel phones.

• Google opened a public SynthID detector after a pilot limited to journalists, media professionals, and researchers.
• TechCrunch says accepted formats include JPG, PNG, MP4, MOV, WAV, and MP3; The Decoder lists JPG, PNG, MP4, and MP3.
• Ars Technica says the worldwide tool requires a Google, OpenAI, or Apple login and about 10 checks a day, a cap meant to hinder SynthID-removal tools.
• Google says Nano Banana, Veo, Lyria, Gemini, Flow, ProducerAI, and Vids embed SynthID; OpenAI, NVIDIA, and Kakao also support it, while Microsoft and Meta use their own approaches.
• The Decoder reports more than 180 billion watermarked images and videos and says Google expects Apple to join.
• About one million verification requests a day; The Decoder attributes that volume to Search, the Gemini app, and Chrome.
• Detection is incomplete: Google says the tools can miss its own models, and unmarked media and many open-weight models will not be flagged.
• Ars Technica says Google also points to C2PA cryptographic labels for authentic media, already used on Pixel phones.