Google's invisible watermarks were already embedded in billions of files. Now there's a public tool that lets you check for them yourself.
Google Opens the SynthID Detector to the Public
Google opened its SynthID Detector to the general public on October 7, giving anyone a way to check whether an image, video, or audio file carries the company's invisible AI watermark. The tool launches in English first, through a portal at synthid.com.
Until now, the detector was a limited verification tool. Opening it to everyone turns a behind-the-scenes safety feature into something you can use on a file a friend sends you. That's a real shift.
The stakes are easy to see. AI-generated content keeps getting harder to spot by eye, and a watermark you can verify is one of the few checks that doesn't rely on gut instinct. That's a big deal.
How to Check a File for a SynthID Watermark
Using the tool is simple. Upload a picture, a video, or a clip of audio, and the detector reports whether it finds a SynthID watermark. The watermark is invisible, woven into the file itself rather than stamped on the corner.
Why bother with all this? Because spotting AI content by eye keeps getting harder.
Simple idea, real value.
Here's what to know. Google says the detector supports JPG, PNG, MP4, and MP3, and that it recognizes watermarks from Google's own models like Gemini, Veo, and Lyria, plus marks added by partners including OpenAI, NVIDIA, and Kakao. That cross-company reach is the part that makes it more than a Google-only utility.
There's a second path that doesn't involve uploading anything. SynthID checks also run inside Google Search, through Lens, AI Mode, and Circle to Search, and through a right-click menu in the Chrome browser via Gemini in Chrome. If you're already looking at a suspect file in your browser, you may be able to check it without leaving the page.
What the 100 Billion Watermarked Files Tell Us
Google says more than 100 billion images and videos generated by Gemini alone already carry the invisible watermark, and it puts the total across all partners even higher. That volume is why a detector starts to matter. If a large share of AI media carries a mark, checking becomes worth the few seconds it takes.
It also sets a limit on the value. Watermarks only help when AI tools add them, and not every tool does. A model that skips watermarking leaves nothing to find. So the detector can confirm a mark, but it can never rule out AI entirely.
Where SynthID Fits in the Fight Against AI Fakes
SynthID is a provenance tool, which means it's built to trace where content came from. It's part of a broader push that includes C2PA metadata, which records signing and editing history in a file's header, and tools like Adobe's Content Credentials. Watermarks and metadata tackle the same problem from different angles.
The honest framing: no single check solves AI fakery. Watermarks can be stripped, metadata can be scrubbed, and clever editing can slip past both. Stacking a few of these methods, and staying skeptical of anything that matters, does more than trusting any one.
What SynthID Can and Can't Prove
The detector is a real step forward for anyone who has to verify media for a living, or for anyone who's been fooled by a convincing fake. It gives a concrete answer where people used to guess.
Just don't treat a clean result as a verdict. No watermark found means the file wasn't marked by a supported tool, nothing more. That's a narrow answer, not a broad one. An AI image built by an unlisted model, or one where the mark was removed, looks identical to a genuine photo to this tool.
SynthID Detector is live now at synthid.com, in English. Start by running it on a file you're suspicious of, then run it again on a real photo to see how the two results compare. You'll learn the tool's limits fast that way, and the limits are the part that keeps you from over-trusting it.
