Google's SynthID Detector is a free page that tells you whether a photo, a video or an audio file carries an invisible watermark placed there by an AI tool. Until 7 October 2026 only journalists and media professionals could use it. Now anyone can.
The tool is at synthid.com. It is made and run by Google, not by webridgeAI.
What it checks
Since 2023 Google has put an invisible watermark, called SynthID, into the images, video and audio its AI makes. Google says it has marked more than 180 billion images and videos this way. Nvidia, OpenAI and Kakao have adopted the same watermark, and Google says Apple will follow.
The detector looks for that one watermark. It does not study the picture and estimate whether it looks artificial, which is how most "AI detector" sites work. Google's own page puts the difference plainly: those detectors "can sometimes be less reliable than watermarking".
How to run a check
- Open synthid.com and accept the terms.
- Drag in the file, or choose Select file. Images, video and audio all work.
- Use the best copy you have. Google asks for the highest quality file available, because heavy compression and editing weaken the watermark.
- Read the result.
How to read the result
| Result | What it means | What to do |
|---|---|---|
| SynthID was detected | The file was made or edited with AI from one of the companies that use the watermark. The page names which one. | Treat it as AI. Google notes the file may have been edited further since. |
| Not detected | The detector found no confident signal. | Do not treat it as real. See the next section. |
For video and audio, the detector marks which parts of the file carry the watermark, so a clip can be part AI and part not.
What it cannot catch
Google is direct about the limits, and they matter more than the headline.
- It is not a general AI detector. It only recognises media from the companies that have adopted SynthID. An image from a tool that does not use the watermark comes back as not detected.
- "Not detected" does not mean real. In Google's words, the result "does not totally rule out AI generation or manipulation".
- Older files may carry no mark. Anything made before a company adopted the watermark has nothing to find.
- Heavy editing can remove it. The watermark is built to survive filters, colour changes, cropping and compression, but Google says it "is not infallible".
- Rare false alarms exist. Google says that in very rare cases the detector can flag a file that carries no watermark.
- A watermark says how a file was made, not whether it is true. A real event can be illustrated with an AI image, and a real photo can carry a false caption.
A sensible way to use it
A positive result is strong evidence. A negative result is not evidence either way, so pair it with the ordinary checks: where the file first appeared, who posted it, and whether a reverse image search finds an earlier copy.
Google's own advice for a decision that matters is to gather more than one signal before you act on any of them.
Sources
- Google, "We're making it easier to identify AI-generated content globally", 7 October 2026.
- SynthID Detector, the tool and its questions and answers, read on 9 October 2026.
- Google DeepMind, SynthID.