The SynthID Detector is now open to everyone. As of October 7, 2026, anyone can upload an image, video, or audio clip at synthid.com and see whether it carries Google’s invisible AI watermark. It’s free, it’s global, and for now it works in English only.
Here’s the catch: it only finds watermarks that were put there in the first place. A clean result doesn’t prove something is real.
Quick facts before the details:
- Launched globally on October 7, 2026, per Google’s announcement.
- Checks images, video, and audio. Google’s post doesn’t mention text.
- Works at synthid.com; the interface is English-only for now.
- Partners include OpenAI, NVIDIA, and Kakao. Apple is listed as “coming soon.”
Why it matters: if you’ve ever squinted at a viral photo and wondered whether it’s fake, this is the first official place to ask.
How does the SynthID Detector actually work?
SynthID adds an invisible watermark while an AI model generates the content. You can’t see or hear it, and it’s designed to survive things like cropping or compression.
The detector simply looks for that hidden signal. If it finds one, it tells you the file came from a tool that uses SynthID. Google says it has watermarked over 180 billion images and videos and 240,000 years of audio since launching the system in 2023.
An earlier version was offered last year to media professionals only. What changed this week is that you no longer need access or an invitation.
Which AI tools can it catch?
Anything made with Google’s own models, plus partners that agreed to use the watermark. The table below shows where things stand today.
| Source | Watermarked with SynthID? |
|---|---|
| Google (Gemini and other Google models) | Yes |
| OpenAI, NVIDIA, Kakao | Yes, as named partners |
| Apple Intelligence images | Not yet; listed as coming soon |
| Other generators | Not unless the maker adds a watermark |
That last row is the big one. Plenty of image tools sit outside the scheme, so the detector can’t speak for them.
If you follow how other companies handle labeling, our piece on the ChatGPT watermark coming to EU text covers a similar approach and its limits.
How to check if an image is AI-generated with SynthID
The process is short. Open synthid.com in your browser, upload the file, and wait for the result. Google’s post doesn’t list file-size limits, so a very large video may need trimming.
Google also says built-in verification already exists in Search, the Gemini app, and Chrome. Those handle more than a million requests a day, so you may not even need the standalone site for a quick check.
What the SynthID Detector can’t tell you
A “no watermark found” result is not a certificate of authenticity. The picture could come from a tool that never adds a watermark, or the mark may have been damaged.
It also can’t tell you who made something or why. Think of it as one clue, not a verdict. For anything important, such as news photos or a suspicious message, check the original source as well.
Apple’s absence matters too. Our look at Apple Intelligence on macOS 27 shows how widely those tools will be used once images carry the watermark. Google hasn’t given a timeline for Apple’s support.
Three habits that make the check more useful
A watermark scan works best alongside ordinary common sense. These three habits take about a minute and catch most of what the tool misses.
- Run the original file, not a screenshot. Re-saving or screenshotting can degrade the signal.
- Look for the source. Reverse-search the image and see who posted it first.
- Check the details. Odd hands, garbled text, and warped backgrounds still give AI images away.
Why this matters: the detector answers one narrow question, and these habits cover the rest.
Frequently Asked Questions
Is the SynthID Detector free?
Google describes it as open to anyone, and its announcement mentions no fee. You just visit synthid.com.
Can it detect AI-written text?
Google’s announcement covers images, video, and audio only. It doesn’t mention text.
Does it work in languages other than English?
Not yet. English is the only supported language at launch, and Google hasn’t shared a date for more.
If it finds nothing, is the image real?
No. It only means no SynthID watermark was found. The image could still be AI-made by a tool that doesn’t use one.
My take: bookmark it and use it as a fast first check, but don’t treat it as a lie detector. It becomes much more useful once Apple and more image tools join, and until then a “clean” result deserves a healthy dose of doubt.
Leave a Reply