Overview
- Google opened the SynthID Detector to the public, Tuesday, with a web portal that lets users upload images, video, or audio for invisible watermark checks.
- The detector recognizes SynthID marks embedded by Google and partner models including OpenAI, NVIDIA, and Kakao, with Apple support announced as coming soon but not yet available.
- Users must sign in with a Google, OpenAI, or Apple account and face a modest daily quota of about 10 checks to reduce the risk that the tool is used to train watermark-removal methods.
- A detected SynthID watermark is strong evidence a supported AI tool created or edited the media, but a negative result does not prove media is authentic because many models do not use SynthID.
- SynthID began at Google DeepMind in 2023 and is already built into products like Gemini, Search/Lens, and Chrome, and Google says those systems together handle roughly 1 million verification requests per day.