SynthID Detector Explained: Check If an Image, Video or Audio File Is AI

If you have ever looked at a photo online and wondered whether it was real, Google has made its own answer to that question public. On October 7, 2026 Google opened SynthID Detector to everyone, having previously limited it to media professionals. You give it an image, a video or an audio file; it tells you whether it can find SynthID, the invisible watermark Google and a handful of other AI companies stamp into what their tools produce. It is out globally in English. It does not check text, and that limit matters more than it sounds.

Key Takeaways

• It is open to everyone now, globally, in English. Google, verbatim: "Now, we're expanding access to everyone -- with the tool available globally in English starting today." That is one language and no country list, which is unusually simple for a Google launch.
• Three file types: images, video, audio. Google: "anyone can easily check if an image, video, or audio file was made with AI from Google or our partners."
• It is not only Google's AI it recognises. Google names its partners: "including OpenAI, NVIDIA, Kakao, and soon, Apple." That makes it the broadest consumer provenance checker any of these companies has shipped.
• Text is not covered, and that is the biggest gap for most people. The thing people most often want checked is writing, and SynthID Detector does not do it. OpenAI and Anthropic each watermark their own text through separate systems with separate checkers.
• Google published no price, no plan and no sign-in requirement. It said "everyone" and "anyone", and named nothing else. We could not confirm the sign-in question from any Google surface, for the reason in the gaps section below.
• Google's own scale numbers, for context on how much is watermarked: "we've watermarked over 180 billion images and videos, along with 240,000 years of audio content."
• You may already be using the same check without knowing. Google says the tool "joins our built-in verification features in Search, the Gemini app, and Chrome, which now regularly handle over 1 million requests daily."

What SynthID Detector Actually Checks

It checks for a watermark, not for AI-ness. SynthID is an imperceptible pattern that Google and its partners embed into files their AI tools generate. The detector looks for that pattern. So the question it answers precisely is "does this file carry a SynthID watermark", and Google has been shipping the technology since 2023.

That table is the whole scope. Google published no further breakdown: no formats, no size caps, no length caps.

Whose AI It Recognises

Four companies today, with a fifth named as coming. This is the part that makes the tool worth knowing about, because a Google-only checker would be far less useful on content you find in the wild.

Two cautions on reading that table. Google wrote "including", so the four named companies are examples rather than a closed list, and it named no specific products: "AI from OpenAI" does not tell you whether a particular ChatGPT image from a particular month is covered. And Apple has its own dated promise here: its iOS 27 feature list says SynthID labels on AI-edited images arrive in "a software update later this year", which is the other half of the same handshake. See iOS 27 Apple Intelligence features.

Why a Clean Result Does Not Mean a Human Made It

Because the tool can only find a watermark that is there. If an image was generated by an AI tool that does not use SynthID -- and most do not -- there is nothing in the file for the detector to find, and you get the same clean result you would get from a photograph.

So read the two outcomes differently:

This asymmetry is true of every watermark-based checker, including OpenAI's, and it is the single most misunderstood thing about them. Treat a positive as an answer and a negative as "no information".

What Google Has Not Published

Four things, and the reason we cannot fill them in is itself worth stating, because it tells you how much more you will learn by poking at the tool yourself.

• Whether you have to sign in to a Google Account. Not stated anywhere in the announcement.
• Which file formats it accepts. Not stated.
• Any file-size or duration cap. Not stated.
• Whether results are logged, and what happens to a file you upload. Not stated in the announcement.

Why those gaps stayed open. synthid.com answers 200 but it is a JavaScript application: 8,093 bytes of HTML that reduce to the 18 characters "SynthID Detector", with no limits, no terms and no sign-in text in the page source. And SynthID Detector has no help centre: support.google.com/synthid/ redirects to the generic Google Help page, which contains the string "SynthID" zero times. Both checked October 10, 2026. So the usual second surface does not exist yet, and the announcement is the only document. When Google publishes a help article, the limits are the first thing it will answer.

Should You Care?

Use it when you have a specific file and a specific doubt. A photo in a group chat, a voice message that sounds slightly wrong, a clip that looks too clean. It costs you nothing that Google has disclosed, takes seconds, and a positive result is genuinely conclusive.

Do not use it to vet text. It will not help, and reaching for the wrong tool is worse than knowing you have none. For writing, the honest answer in October 2026 is still that no reliable public checker exists for arbitrary text; the vendor-specific systems only catch their own output.

Do not treat a clean result as proof. This is the mistake that will cause real arguments. "SynthID Detector says it is fine" means "Google's watermark is not in this file", nothing more.

Nothing to buy and nothing to install. Google named no plan and no price, and the tool runs in a browser. If you want the wider picture of which AI companies mark their output and how, start with ChatGPT's text watermark and Claude's, which cover the text half this tool leaves out.

The bottom line: SynthID Detector is a free-to-use, globally available web check for AI-made images, video and audio from Google, OpenAI, NVIDIA and Kakao. It is the most useful provenance tool a normal person can reach today, and it is useless on text. Check a file when you doubt it, believe a positive, and do not read anything into a negative.

Sources

• Google, "We're making it easier to identify AI-generated content globally", article:published_time 2026-10-07, read October 10, 2026 -- the only document Google has published about the public launch, and the source of every quote and figure above
• synthid.com and synthid.com/verify, both read October 10, 2026: HTTP 200, 8,093 and 8,054 bytes, stripping to 18 characters each. No limits, terms or sign-in text in the page source
• support.google.com/synthid/, read October 10, 2026: redirects to support.google.com/, the generic Google Help page, with zero occurrences of "SynthID". There is no SynthID help centre