Google SynthID Detector Is Now Open to Everyone: Upload an Image, Video, or Audio File to Check for Invisible AI Watermarks

Picture a night-vision clip of a raccoon sitting quietly and eating a sandwich. It is so cute you want to send it to a friend right away. Then, halfway through, a thought pops up: was this actually filmed? Funny enough, the demo screenshot in Google's official announcement is exactly that kind of raccoon video.

Google announced on October 7 that SynthID Detector is now open to everyone worldwide at synthid.com. According to the Google official site, it is a tool for identifying SynthID watermarks in AI-generated media. Google's official blog puts it more plainly: anyone can check whether an image, video, or audio file was made with AI from Google or its partners. In this article I will walk through what it checks, how to use it, and why “not detected” is the result people misread most often.

SynthID Detector is a web-based checking tool from Google that scans an image, video, or audio file for SynthID, a hard-to-notice watermark placed in AI-generated content. According to Google's official blog, SynthID launched in 2023 and has since been applied to more than 180 billion images and videos, along with 240,000 years of audio. That is a staggering number, and it means plenty of AI content already carries a mark that we simply cannot see.

The biggest change this time is who gets to use it. Google says an early version came out last year, but only for media professionals verifying AI-generated content. Now it is open to everyone, available globally, though for the moment only in English. For ordinary readers, a tool that used to feel far away is now just a web page you can open.

How is it different from a typical AI detector?

SynthID Detector looks for a watermark that an AI tool embedded when it generated the content; it does not guess whether a picture “looks like” AI. This matters because many people see the words “AI detector” and expect a verdict of real or fake the moment they upload a file. In practice, going by Google's own description, it can answer only one question: does this file carry a SynthID mark?

My own analogy is the tag on a piece of clothing. If there is a tag, you know which brand made it; if there is no tag, it could be from another brand, or your grandmother may have knitted it by hand. So a missing tag, by itself, tells you nothing about where the garment came from.

Which AI tools can it detect, and when does Apple join?

Right now SynthID Detector can read content made with Google's AI and with partners OpenAI, NVIDIA, and Kakao, while Google lists Apple as coming soon. The interesting part of that list is that this is no longer a one-company effort. OpenAI is on it, which means content made with Gemini or with OpenAI tools can be recognized by the same site.

As for Apple, Google only said “soon” and gave no timetable. With so many people carrying iPhones, Apple joining would make the site far more useful. Until then, anything from a tool that has not joined will probably just come back as “not detected.”

How does it compare with the built-in checks in Search, Gemini, and Chrome?

SynthID Detector is a standalone checking website, while the verification features in Search, the Gemini app, and Chrome put the same kind of check inside products you already use. Google's blog says these built-in features now handle more than 1 million requests a day. That is a useful hint that far more people want to know whether something is AI-made than we might assume.

My take: if you just want to double-check one image while browsing, the built-in feature is enough. For a suspicious file you want to examine properly, I would guess the dedicated site feels more natural. The two do not conflict, since Google is simply putting the same check in more places. For people who often share or quote images, having one more dedicated page to use is a small comfort.

How do you use SynthID Detector?

Checking a file with SynthID Detector comes down to three things: open the site, upload the file, and read the result. The official announcement does not walk through the screens, so what follows is a general flow I put together, and the actual interface on the site takes priority. One thing up front: for now the site is English only.

Step 1: open synthid.com in your browser, which is the official entry point.

Step 2: if the site asks you to sign in, follow the on-screen prompts, and confirm the sign-in requirements on the live site.

Step 3: upload the image, video, or audio file you want to check.

Step 4: wait for the analysis, then see whether it reports a SynthID watermark.

My own tip is to use the original file rather than a screenshot or a copy that has been forwarded many times. Once the quality is compressed or the frame is cropped, less information survives and the result is harder to read. This is a habit of mine, not an official recommendation, so if a result looks odd, retrying with the original is the quickest fix.

Back to the raccoon clip: the official demo screenshot shows a video player playing that night-vision video. For a first try, I would test one image recently made with Gemini or ChatGPT, then one photo you took on your phone. The first has a chance of being recognized, and the second will most likely come back as “not detected.” Seeing the two results side by side teaches you more than reading ten explainers.

If no watermark is found, does that mean the content is real?

If no watermark is found, all it means is that no SynthID mark was detected in the file; it does not prove the footage is real, and it certainly does not prove AI was never involved. Say a friend drops a very convincing news clip into the group chat, and you upload it and get “not detected.” What you can conclude is limited: it may not have been made with a tool that uses SynthID, or the mark may no longer be there.

There are at least three explanations behind that result: the clip was truly filmed; it was made with an AI tool that does not use SynthID; or it was heavily processed and the mark may be gone. All three look identical on screen, and the site cannot tell you which one applies. So please do not treat “not detected” as a guarantee.

The reverse needs care too. My understanding is that a detection means AI was involved, but whether the whole thing was generated or only touched up by AI may not be clear from the result page. For anything important, combine it with the original source, where it was first published, and other ways of verifying. That is why I see it as supporting evidence rather than a verdict.

What does this global opening mean for ordinary people?

The significance of this opening is that AI content verification, once available only to media professionals, is now a public website anyone can open. Until now, if we wanted to check a picture, we could only go by instinct: look at the fingers, the text in the background, the lighting. Now there is at least one reference with a clear basis. But how useful it is depends entirely on how many AI tools agree to take part.

Looking at the list, Google, OpenAI, NVIDIA, and Kakao are in, and Apple is lined up next, so more players are clearly joining. But anything outside the list is invisible to the site, and that gap can only be closed as more companies sign on. That is why I treat it as the first step of a verification process, not the last.

SynthID Detector FAQ: what do people ask most?

Here are the 6 questions people ask most often, answered from the official announcement, and I say so plainly where the official announcement is silent.

How is SynthID Detector different from a regular AI detector?

A typical detector analyzes visual features to estimate the probability that something is AI-made. SynthID Detector instead checks for the SynthID watermark that an AI tool adds while generating content. That gives it a firm basis for participating tools and little to offer for tools that do not participate.

Is it available in my country, and is there a non-English interface?

Google says the tool is available globally starting from launch day and is offered in English, so readers anywhere can use it, with the interface in English for now. The official announcement does not mention other languages.

Does it cost money or require an account?

The official announcement does not mention pricing or sign-in requirements, and I did not see any fee described there. I suggest opening the site and going by what the live page shows.

What kinds of files can it check?

The official announcement mentions images, video, and audio, and it does not mention text. It also does not list the supported file formats one by one, so go by the notes on the site's upload screen.

If no watermark is found, can I assume the content is real?

No. A miss only means no SynthID mark was detected. The content may be genuine footage, or it may have been made with an AI tool that does not participate, and the site cannot tell you which.

Can it detect content made with Apple AI tools yet?

Google only says Apple is coming soon and gives no date, so do not expect it to work yet. Once Apple is officially announced, test it with real content to be safe.

Is SynthID Detector worth bookmarking?

SynthID Detector can show that a file does carry an AI watermark, but it cannot show that a file has never touched AI. So I treat it as a first step: a hit is a meaningful lead, and a miss means you keep asking where the file came from. In an age of ever more AI content, the habit of checking one more thing matters more than trusting any single tool's answer.