← All posts
This post is written in English.
Google opened its SynthID Detector to everyone in English, after an early version for media professionals. It checks whether an image, video or audio file carries the imperceptible SynthID watermark, from Google's AI or from partners. The post names OpenAI, NVIDIA, Kakao and Apple.
Google says built-in verification across Search, the Gemini app and Chrome now handles over 1 million requests a day.
What it can't tell you
The post doesn't give accuracy rates or failure cases. And there's a basic limit: it finds content from Google and its partners. Open-source models, or any tool that doesn't take part, won't carry the mark.
So there are two different answers here. "Watermark found" tells you something useful. "No watermark" doesn't tell you a person made it.
Why it still helps
This is part of a broader shift toward marking AI content, pushed along by rules like Article 50 of the EU AI Act. A shared watermark across big companies is more useful than each company having its own.
My read
I can already picture someone screenshotting "no watermark detected" as proof that a fake image is real. That would be the worst way to use this tool. Treat it as one small clue, alongside the old questions: who posted this, where did it first appear, and does anyone credible confirm it?
A published false-positive and miss rate would make it far more useful, and that's what I'd ask Google for next.
