Imagine a photograph of a giant pumpkin hanging above a busy street. It could show a real decoration, a generated scene or a real photograph changed with AI. This is an invented example, but it captures a familiar problem: the picture alone may leave its origin unclear.
A tool that checks for an invisible watermark can supply a particular kind of evidence. Understanding what it looks for helps you interpret the answer.
Google’s October 7 announcement, updated October 29, says its standalone SynthID Detector is now available globally in English. It covers images, video and audio from Google and participating partners.
A marker woven into the content
DeepMind describes SynthID as embedding a digital watermark directly into generated media. People are not meant to see or hear it, but a detector can look for it. For images, the useful question becomes: did a supported AI tool leave a recognizable marker in this file?
That question is narrower than deciding whether everything shown in the picture happened. A marker can tell you something about how the file was made or changed. It does not independently establish where a photograph was taken, when an event occurred or whether its caption is accurate.
The checker matters. Google’s standalone Detector covers participating partners; Gemini’s SynthID check currently recognizes Google AI content, according to its help page.
Read the result precisely
Google says a SynthID detection in Gemini means an image or video was wholly or partly generated or edited by its AI. For our pumpkin example, that supports AI involvement without establishing whether the entire street was invented.
An absent mark leaves more possibilities open. The image might come from a generator outside the checker’s coverage. Google also documents uncertain cases involving very simple content or small edits, and warns that many alterations can make a watermark undetectable. A “no watermark” result is therefore insufficient evidence that a picture is an untouched camera photograph. See the documented limits.
Then check the story around the picture
Suppose the pumpkin photograph really was taken with a camera, but someone reposted last year’s decoration as something happening tonight. Establishing that no AI was involved would still leave the date wrong.
A practical next step is to find the earliest available source, read its caption and look for evidence about the claimed place and time. If a detector reports AI involvement, retain that finding accurately: avoid turning “edited with AI” into “every part is fabricated.” If it cannot settle the question, keep that uncertainty visible.
Watermark checks are useful when they add a specific piece of evidence. Pair that evidence with the source of the image and the claim being made about it.
