Last updated: October 13, 2026.
The short version
Google’s SynthID Detector went free for everyone on October 7, 2026. Go to synthid.com, sign in with a Google, Apple, or OpenAI account, and upload a photo, video, or audio file. It tells you whether the file carries an invisible SynthID watermark, which means AI was involved in making it. That is genuinely useful. It is not a magic fake-detector, and the most important thing to understand is what a negative result means: nothing. More on that below.
What just changed
SynthID is the invisible watermark Google has been embedding in AI-generated content since 2023. Until this month, the detector that reads it was limited to journalists, media professionals, and researchers who got test access starting at Google I/O last year. On October 7, Google opened it to everyone, worldwide, in English, for free. The only price is a sign-in.
The timing is not subtle. AI-generated photos now flood every feed, and the old advice, “look for weird hands,” stopped working about two model generations ago. Google’s answer is infrastructure: watermark at creation, verify at consumption. The detector is the verification half.
What it actually detects (and what it doesn’t)
Here is the honest core of this article, because most coverage glosses over it. SynthID Detector does not detect “AI-ness.” It detects one specific watermark. If the file has it, AI from a participating system touched it. If it doesn’t, you have learned almost nothing.
What carries the watermark: content from Google’s models (Gemini, Imagen, Veo, Lyria, and the Nano Banana image model), Google’s tools (Gemini app, Flow, Vids), and partners OpenAI, NVIDIA, and Kakao. OpenAI’s participation covers images made through ChatGPT, Codex, or the OpenAI API. Apple support is expected soon, with no date given.
Read Also: ChatGPT Is Getting Visual Ads Inside Image Generation
What it will miss: anything from a generator that never watermarks (basic Stable Diffusion, most open models), Microsoft and Meta content (they run their own marking standards), and files where heavy editing or recompression degraded the signal. Researchers have even demonstrated an attack tool, UnMarker, that cut SynthID image detection from nearly 100% to 21%. Treat the detector as one clue, not a verdict.
How to use it
- Go to synthid.com in any browser.
- Sign in with your Google, Apple, or OpenAI account.
- Upload the file. Images: JPG, PNG, WEBP, AVIF, HEIC, HEIF, TIFF, GIF, BMP. Video: MP4, MOV, WEBM. Audio: WAV, MP3, OGG, FLAC, AAC, M4A.
- Read the result: “watermark detected” or “watermark not detected,” plus where in the file it was found.
Google applies usage quotas, so this is not a bulk-scanning tool. For quick checks inside Google’s own apps, you can also upload a file to the Gemini app and ask “Was this created by AI?” That route handles one file at a time, up to 100 MB, video under 90 seconds, audio under an hour. Google said the Gemini verification feature had already been used 50 million times as of May.
How to read the result
“Watermark detected” is the strong result. It means a participating AI system generated or edited the file, and the detector can usually say which one. For video, note that a clip can mix AI and real footage; a watermark does not prove every frame is synthetic. Same for audio: a watermarked segment does not mean the whole recording is fake.
“Watermark not detected” is the weak result, and this is where people get burned. It means exactly one thing: no SynthID watermark was found. The photo could still be AI-generated by any of the dozens of systems that don’t watermark. Never use a negative result to declare a photo real, especially if the stakes are high (news, evidence, money changing hands).
Other free ways to check
The standalone detector is the most direct tool, but Google has scattered verification across its products. Google Lens and Circle to Search can check images for SynthID and C2PA content credentials. Chrome has verification built in. And since 2026, YouTube automatically labels videos with significant photorealistic AI content instead of relying on creators to disclose it. If you live inside Google’s ecosystem, you may never need the standalone site.
Our take: use the detector as a first pass, not a last word. A positive result settles the question. A negative result settles nothing. For anything that matters, that asymmetry is the whole story, and any guide that buries it is selling you certainty that doesn’t exist.
Frequently asked questions
Is Google’s SynthID Detector really free?
Yes. Since October 7, 2026, it is free for everyone worldwide, in English. The only requirement is signing in with a Google, Apple, or OpenAI account. Google applies usage quotas, so you cannot bulk-scan hundreds of files, but casual checks cost nothing.
Can it detect images made with ChatGPT?
Yes, for supported outputs. OpenAI is a SynthID partner, and its participation covers images created through ChatGPT, Codex, or the OpenAI API. That does not mean every image ever touched by an OpenAI product carries the watermark, but current supported generations do. Images from other generators that don’t watermark, like basic Stable Diffusion, will come back as “not detected.”
Does “watermark not detected” mean the photo is real?
No, and this is the single most misunderstood thing about the tool. It means no SynthID watermark was found, nothing more. The image could be AI-generated by any system that doesn’t apply the watermark, or the watermark could have been weakened by editing and recompression. A positive result is meaningful. A negative result is not proof of anything.
