Tuesday, 17 November 2026

SynthID AI detector lets anyone verify AI-made images and audio

Google's free checker now spots hidden watermarks from Google, OpenAI, Nvidia, Kakao and, since iOS 27.2, Apple's image tools. A clean result still proves nothing.

Hands holding laptop with printed photo and voice recorder on desk

The short version

  • Google has opened its SynthID Detector to everyone worldwide in English, after a limited test with journalists and researchers.
  • It looks for invisible watermarks left by AI tools from Google, OpenAI, Nvidia and Kakao, and Apple now adds them in iOS 27.2.
  • You must sign in with a Google, Apple or ChatGPT account, and checks are capped at roughly 10 a day.
  • A clean result does not mean a file is real, because many AI tools, including Meta's, never add a SynthID mark.
Quick read · 2 min

Google has opened its SynthID Detector to everyone in the world in English. You upload a photo, video or audio clip and it tells you whether it carries the invisible watermark that AI tools from Google, OpenAI, Nvidia and Kakao hide inside what they create. Since iOS 27.2 in late October, Apple's Image Playground and some Photos edits add the same mark.

It is handy for checking viral or suspicious media, but the limits matter:

  • A clean result proves nothing, because many AI tools, including Meta's, never add SynthID.
  • You must sign in with a Google, Apple or ChatGPT account, and you get about 10 checks a day.
  • Google deletes the upload after the check but keeps a fingerprint of it for 24 hours.

Use it as one clue alongside other checks, and avoid uploading private photos to any online detector.

Anyone can now upload a suspicious photo, video or voice clip to Google and get a straight yes or no on whether one of the big AI tools made it. In early October Google opened its SynthID AI detector to everyone in the world in English, turning a tool that used to be reserved for journalists and researchers into a free website for the public.

That matters because fakes are getting harder to spot by eye. A viral photo of a politician, a voice note that sounds like your boss, a too-perfect product shot on a shopping site: the detector gives you one more piece of evidence before you share, believe or pay. But it has a catch that the cheerful launch pitch skips over, and it is worth understanding before you trust a result.

01

How the SynthID AI detector actually works

The SynthID detector does not guess whether something looks fake. It searches a file for an invisible watermark that supported AI tools hide inside the content when they create it.

Google built SynthID in 2023. The mark is woven into the pixels of an image or video, or into the sound of an audio track, rather than stored as a label in the file’s metadata (the hidden details like date and camera that often get stripped when you share or screenshot something). Google says the watermark survives common changes such as cropping, filters and compression, and that audio marks hold up through added noise, MP3 compression and speed changes.

The scale is large. Google says it has stamped more than 180 billion images and videos with SynthID, plus what it describes as 240,000 years of audio. The same check already runs inside Gemini, Search and Chrome, which together handle over 1 million verification requests a day, so the new website mainly puts an existing system in one easy place.

Overhead angle of hand placing printed photo next to audio recorder
02

Which AI tools it can and cannot spot

The detector recognizes watermarks from Google’s own AI and from its partners OpenAI, Nvidia and Kakao. Apple joined too, and that part has already moved on since launch.

When the public version opened, Google said detection for Apple-made media was coming soon. Apple then shipped iOS 27.2 in late October, and its release notes list SynthID support in both Photos and Image Playground. Every picture made in Image Playground now carries the mark. In Photos it depends on the tool: early testing found that Extend and Reframe add the watermark, while Clean Up only adds it for larger edits, so a small touch-up can slip through unmarked.

What the detector cannot see is just as important. It knows nothing about AI tools that use a different watermark or none at all. Meta, for example, uses its own mark called Content Seal and runs a separate checker. Open-weight models that people run on their own computers can also produce images with no SynthID mark, and so can free tools like ComfyUI.

Play video: Samsung Galaxy A18 review
03

The catch: a clean result is not proof something is real

This is the part to remember. If the SynthID detector finds a watermark, you can be fairly confident a supported AI tool was involved. If it finds nothing, you have learned very little.

Because the tool only reads a deliberate mark, false alarms should be rare. The flip side is that a fake made with a non-partner tool will come back clean, and Google admits watermarks can be removed in some cases. The detector also can’t tell you how much of a file was AI. A photo with one small AI edit and a picture generated from scratch can both simply come back flagged.

That has already caused confusion. iPhone owners have reported photos edited with Apple’s AI tools showing up as AI-edited on Instagram, though it isn’t clear how Instagram spotted them, since Meta’s documentation does not mention SynthID and AI labels can also come from metadata standards. Treat any single result as a clue, not a verdict.

Back view of person at kitchen table with printed photo and smartphone face-down
04

What this means for you before you upload anything

For most people the SynthID detector is a useful extra step when something online feels too good, too shocking or too convenient to be true. It is free, it takes a few seconds, and you don’t need to ask a chatbot for its opinion.

There are a few practical limits:

  • You must sign in. The site needs a Google, Apple or ChatGPT account, so you can’t check files anonymously. If you use Apple or another outside sign-in, Google keeps basic account details such as your email address and user ID, with no stated time limit.
  • Checks are rationed. Google caps use at about 10 checks per person per day. It says this makes it harder for people to keep testing files while trying to strip the watermark out.
  • Your file is not kept, but a fingerprint is. Google says uploads are processed in real time and deleted once you see the result, but it holds a digital signature of each file for 24 hours to enforce the daily limit and block automated abuse.

The privacy point matters most for personal or sensitive pictures. It is wiser not to upload private family photos or intimate images to any online checker. Checking inside the Gemini app is a different deal again, because there your file is treated like any other chat upload and may be kept and reviewed depending on your settings.

05

How to check a photo, video or audio clip

Using the SynthID detector takes under a minute. Open Google’s SynthID Detector website in any browser, sign in with one of the supported accounts, upload the file and wait a few seconds for the answer.

Most everyday files work: the site takes 11 picture formats such as PNG, JPG and GIF, three video formats including MOV and MP4, and six kinds of audio, among them FLAC, WAV and MP3. If you’d rather not leave what you are doing, Google’s Chrome browser can also read SynthID marks, and so can Search through Circle to Search, Lens and AI Mode. Gemini can as well, though Google says Gemini currently only recognizes content from Google’s own AI.

On an iPhone there is no built-in checker in Photos, so the website is the way to test a picture made or edited with Apple’s tools. For anything important, such as a news photo you plan to share or a clip used to ask you for money, pair the result with other checks: look for the original poster, search for the same image elsewhere, and see whether trusted outlets have reported the event.

06

Quick answers

Is the SynthID detector free?

Yes. It costs nothing, but you need to sign in with a Google, Apple or ChatGPT account and you get roughly 10 checks a day.

Can it detect images made on an iPhone?

Yes, for images made or edited with Apple’s AI tools since iOS 27.2. Everything from Image Playground carries the mark, while some small Clean Up edits in Photos do not.

If SynthID finds no watermark, is the picture real?

No. It only spots marks from Google and its partners, so content from other AI tools, such as Meta’s or open models run at home, will come back clean.

You're reading the quick version.