Google opens its SynthID Detector so anyone can check if media is AI-made

It's a standalone website. You upload a file, and it checks for SynthID, an invisible watermark. Google introduced SynthID in 2023. The mark is built into the pixels or the sound of AI-made media, so you can't see or hear it, but software can find it.

Google says the tool is for anyone who wants to confirm "whether content was created using Google AI or tools from our industry partners." It was announced by a vice president of science and strategic initiatives at Google DeepMind.

The tool isn't brand new. Google says it introduced an early version last year for media professionals. TechCrunch adds that it was offered to select journalists, media professionals and researchers for testing at Google I/O last year. Now the doors are open to the public.

The first public step came in November 2025. Google then let people upload an image to the Gemini app and ask whether it was created with Google AI. Gemini checked for the SynthID watermark and answered in plain words.

Google said at the time that video and audio checks would follow, and that it planned to support C2PA content credentials too. CNET reports that video and audio detection reached Chrome and Search at Google I/O in May.

All three of the big media types. TechCrunch lists the supported formats:

• Images: JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF and GIF.
• Video: MP4, MOV and WEBM.
• Audio: WAV, MP3, OGG, FLAC, AAC and M4A.

CNET says you need a Google, Apple or ChatGPT account to upload media.

In real life Your uncle sends the family chat a video of a bear riding a bicycle through a supermarket. You save it, drop it into the detector, and a few seconds later you know whether SynthID shows up.

This is the part to read twice. The detector doesn't spot "AI" in general. It spots SynthID tags, so it only works on media made by companies that add those tags.

Google names the current list: itself, plus OpenAI, NVIDIA and Kakao. Apple is listed as coming "soon."

TechCrunch says Google's own Nano Banana, Veo and Lyria models, and tools with generation features such as Gemini, Flow, ProducerAI and Vids, all use SynthID to watermark what they make.

The catch: a result of "not detected" means no supported tag was found. It does not mean a human made the file. An image from an AI tool that doesn't use SynthID will come back clean.

Big. Google says that since 2023 it has watermarked over 180 billion images and videos, along with 240,000 years of audio. That is a lot of listening.

For comparison, in November 2025, when Google added image checks to the Gemini app, it said over 20 billion pieces of content had been watermarked. The number has grown roughly ninefold in under a year.

The detector joins checks that already exist in Search, the Gemini app and Chrome. Google says these now regularly handle over 1 million requests a day.

CNET's reviewer ran the tool on several kinds of images, and the results show both its strengths and its edges:

• ChatGPT image: an image made from a prompt in ChatGPT was correctly flagged.
• Pixel edit: a photo edited with Gemini in Google Photos on a Pixel 11 Pro XL, asking for a night version with an aurora, was correctly flagged.
• Apple Reframe: an iPhone 18 Pro photo edited with Apple's Reframe tool was detected, while an untouched iPhone photo correctly showed no tag.
• Apple Playground: an image made in Apple's Playground app with ChatGPT's image model was not flagged. CNET thinks the app doesn't have SynthID tagging wired in yet.

CNET also checked the same AI samples on OpenAI's own verification site, and they came up negative there. That site only looks for OpenAI's own watermarks.

Apple is the interesting guest. CNET reports that Apple said it would support SynthID at its recent fall event, where it announced the iPhone 18 Pro and iPhone Duo. A Google spokesperson told CNET there is no specific date, but that support will be available soon.

CNET notes that Google's tool already seems to recognize tags in some Apple-edited photos, as the Reframe test showed.

Apple has not signed on to C2PA, the Coalition for Content Provenance and Authenticity standard, according to CNET. Instead it uses its own Reference Image technology, which stores a verified original of a photo. Those shots have to be taken in Reference mode in the Camera app.

Google, by contrast, takes a two-part approach on its own phones. CNET says it puts C2PA metadata in images to show the original capture, and adds a SynthID tag when AI edits the image.

CNET read the terms of service you accept on your first visit. Here is what they say, according to the outlet:

• Processing: the files you upload are handled in real time.
• Deletion: the files are deleted after the result is shown.
• Signature: a digital signature of the media is linked to your user ID, then deleted after 24 hours.

CNET called that a sensible way to handle the media. We haven't seen independent testing of those claims, so treat this as what Google says it does.

Because there's no single standard. TechCrunch notes that Microsoft and Meta have their own standards for watermarking and verifying content. It adds that these tools are "not infallible," and often fail to identify content created by their makers' own models.

A Google spokesperson told CNET that working with other companies is crucial, because no single company can solve the problem alone.

It's also work for the viewer. CNET points out that you have to download the media, open the tool and upload it. Most of us won't do that for every picture in a feed. The tool helps when something looks suspicious and you care enough to check.

Our own coverage shows the same pattern from other angles. OpenAI is adding invisible text watermarks in the EU, as we reported in OpenAI will start invisibly watermarking ChatGPT text in the EU, and DeepMind has extended the idea to proteins in SynthID Bio. The watermark is spreading. The shared checker is catching up.

• Use it when you doubt something. If a viral image or clip seems too perfect, upload it before you share it.
• Read "not detected" carefully. It means no supported tag was found, not that the file is real.
• Check more than one way. Gemini, Search and Chrome have built-in checks, and OpenAI runs its own site for its own marks.
• Mind the account. CNET says you sign in with a Google, Apple or ChatGPT account, so decide whether you're comfortable uploading a file under it.