How to Check for SynthID: What You Can Actually Verify Today

"Check for SynthID" sounds like one task. It is at least four, and the honest answer for most of them is that you cannot do it yet. This guide sorts the four by what is actually open to the public on September 14, 2026, and gives the steps for the ones that are.

First, which SynthID?

SynthID is four systems under one name: text, image, audio and video. The route to a check depends on which one you mean and, for images, on who made the image. Work through the sections below in order and stop at the one that matches.

Text from Gemini: waitlist only

There is no public check for SynthID Text. Google's SynthID Detector portal is limited to journalists, media professionals and researchers, and access is by application through the SynthID page on deepmind.google. If you qualify, that is the route; if you do not, there is no paste-and-check page anywhere.

Two things look like alternatives and are not. The open-source SynthID-Text repository publishes the detection code but not Google's production key, so running it against Gemini output tells you nothing. And every website that offers a "SynthID text detector" without Google's involvement is running a stylistic classifier, which returns a likelihood rather than reading a mark.

DetectMark does not pretend otherwise. Paste Gemini text into the SynthID detector and the statistical row reports that verification is unavailable, with the reason. The hidden-character and clipboard rows still run, because those are real checks.

Images from Gemini: Google's own surfaces

This one is open, but only inside Google.

1. Gemini app. Upload the image, or a video or audio clip, and ask whether Google AI created or altered it. The app checks for SynthID and answers.
2. Google Search. Use Lens, AI Mode or Circle to Search on the image and open the About this image panel. SynthID verification runs there.
3. Chrome. The same check is rolling out in the browser as part of the About this image feature.

A positive result means Google's SynthID is present, so a Google model produced or processed the image. A negative means Google's mark was not found, which says nothing about images from other providers. None of these surfaces will tell you about an OpenAI image, an Adobe image or a photograph.

Images from ChatGPT: OpenAI's verifier

Since May 19, 2026, OpenAI images carry SynthID under licence from Google, and OpenAI runs the only public check for them. It is available directly through OpenAI's Verify tool and Content Provenance API, and DetectMark can run it for you.

1. Open the OpenAI image verifier and drop the image. It is read in your browser first.
2. Click Verify with OpenAI and confirm the dialog, which names the file, its exact byte count and the destination, api.openai.com. Nothing is sent until you confirm.
3. Read the two rows that come back: SynthID and Content Credentials, each Found or Clean, with the time OpenAI ran the check.

Does ChatGPT watermark images? walks through what each result proves.

Audio and video

OpenAI audio has carried SynthID since July 31, 2026 and is verifiable through the same OpenAI tool and API. DetectMark does not process audio, so an audio file here reports Unsupported rather than a guess.

For Google's own audio and video, the Gemini app takes a clip the same way it takes an image: upload it and ask whether Google AI created or altered it. The SynthID Detector portal accepts the same files for its approved testers. Google's Search and Chrome surfaces are documented for images only, and the status page tracks the row.

What DetectMark can and cannot do locally

Drop any image into the AI image watermark detector and the file is read in your browser: Content Credentials, EXIF, XMP, GPS and PNG text chunks, each with a verdict. Content Credentials often ship alongside SynthID, so their presence is a useful clue about where a file came from. They are not SynthID, and the result panel keeps the pixel-watermark row at Unknown, because no browser can read the pixel mark. The only way a verdict lands on that row is the OpenAI check above, on consent.

The whole picture in one table

Every row reads the provider's own mark with the provider's own key. There is no cross-provider check, and there is no browser-only check. That is the state of things, and the page linked below will say so until it changes.