SynthID Detector: Use the Official Google Route

You have an image in a Slack thread, a short clip in a deck, or an audio sample someone swears came from a Google AI tool. The question is simple: “Can I run it through a SynthID Detector and know what it is?” The honest answer has two parts. For SynthID itself, use Google’s own detector. For the vendor-neutral evidence around the file, check metadata and Content Credentials with tools that say exactly what they found.

That split matters. SynthID is Google’s watermarking technology. It is designed to place a signal inside generated or edited media in ways that are not meant to be obvious to a viewer. A normal metadata reader cannot see that watermark. A Content Credentials verifier cannot decode it either. If a third-party page presents itself as a Google SynthID Detector, the first thing to ask is whether it is using an official Google mechanism or simply looking for other clues and using a stronger name than the evidence deserves.

What Google’s detector is meant to answer

The official SynthID Detector is the place to go when your specific question is whether supported Google-generated content carries a SynthID signal. Google has described SynthID across several kinds of AI output, including images made with Gemini, Nano Banana, and Imagen, video from Veo, audio associated with Lyria and NotebookLM, and text from Gemini. That does not mean every file in the world is covered, and it does not mean every copy of a supported file will remain checkable after being exported, recompressed, cropped, copied, pasted, or otherwise changed.

Plainly: a SynthID Detector is not a general “AI or human” switch. It is a way to look for a particular Google provenance signal when the file and format are in scope. If the tool reports a signal, that is useful evidence about a Google AI system being involved. If it does not report one, you have not proven the file is human-made. You have only learned that this route did not return a readable SynthID signal for that item.

How to reach the official portal without falling for lookalikes

The safest route is to begin with Google’s own public SynthID information and follow the links Google provides for detector access. If you are reviewing media for work, bookmark the official source rather than searching fresh each time and clicking the first upload form that appears. Search pages change. Ads move. Copycat pages can use convincing words. The official route should live on a Google property or be explicitly linked from one.

Before uploading sensitive material anywhere, pause. Ask whether you are allowed to share the file with that service. Ask whether the tool explains retention, supported file types, and its relationship to Google. An honest page will not pretend that the absence of a watermark means the file is safe, authentic, or human-made. It will keep the result narrow. That narrowness is not a weakness; it is what makes the result credible.

Where our free checks fit in

Check AI Media does not claim to read SynthID. Our free checks are the vendor-agnostic half of the workflow: they look for signals that are designed to be readable outside one company’s private detector. The metadata viewer can surface camera fields, editing software tags, and embedded descriptive data when those fields are still present. The Content Credentials checker can verify C2PA manifests when a file carries signed provenance.

Those checks are useful before and after an official SynthID pass. They can show that a file carries a signed editing history, that a tool name appears in metadata, or that a credential validates. They can also show nothing. Nothing is still information, but it is not a verdict. Files travel through messengers, social networks, editors, and screenshot workflows that strip context. A missing credential or empty metadata panel does not tell you the content was made by a person.

A practical review order

When someone sends you questionable Google-looking media, start by preserving the original file if you can. Do not rely on a screenshot of a screenshot. Run the file through open checks first if you want to see whether it carries standard metadata or C2PA Content Credentials. If the file is plausibly from a Google AI system and the question is specifically SynthID, continue to the official Google detector. If you are not sure what kind of evidence a tool can find, read what can be detected before you interpret the result.

The goal is not to collect as many badges as possible. The goal is to avoid saying more than the evidence allows. “This file carries a valid credential” is a real statement. “Google’s detector found SynthID” is a real statement when it comes from the official tool. “No signal was found, therefore it is not AI” is not a real statement. That last leap is where credibility disappears.

What to write in a report

If you need to document a finding, use careful language. Say which tool you used, what file you tested, and what the tool returned. Keep a copy of the original file and a note about where it came from. If the official detector was part of your workflow, describe it as an official SynthID check, not as a general authorship test. If our site found C2PA or metadata signals, describe those signals by name.

The most useful provenance notes are boring in the right way. They separate evidence from interpretation. They let a colleague repeat the check. They do not pretend a clean result settles the whole story. For Google-specific watermarking, start with the official SynthID Detector. For the open signals that can travel with many files, use independent checks and read the result literally.