Google SynthID Detector Review (2026): The Free AI-Content Checker — and Its Honest Limits
An honest note up front: we haven't run hands-on tests on the public portal yet — this analysis is built from Google's October 7, 2026 announcement and independent reporting from MacRumors, Unite.AI, and others. We'll update once we've uploaded files ourselves. No affiliate links in this article.
On October 7, 2026, Google did something quietly significant: it opened its SynthID Detector to everyone, worldwide. The portal had spent the previous year as a restricted tool for journalists, media professionals, and researchers. Now anyone can upload an image, video, or audio file and check whether it carries an invisible SynthID watermark identifying it as AI-generated.
The timing matters. AI-generated images, voiceovers, and video clips are flooding every platform creators work on — and telling real from synthetic by eye is increasingly impossible. A free, credible detection tool from Google is worth understanding. But it's also worth understanding what this tool doesn't do, because the name invites over-trusting it.
What is SynthID Detector?
SynthID is Google's invisible-watermarking technology, launched in 2023. When AI-generated media is created, a statistical signal is embedded into it — into the pixel values of images and video, into the waveform of audio — at the moment of generation. It's imperceptible to human eyes and ears, and Google says it survives common edits like compression and cropping.
SynthID Detector is the verification portal that reads those watermarks. You upload a file; it tells you whether the file carries a SynthID mark from a participating generator. Per Google's own FAQ, quoted by Unite.AI: "This is not a general AI detector." It can identify only media from companies that have adopted SynthID.
Before October 7, 2026, the portal was available only to a small group of journalists, media professionals, and researchers who got access at Google I/O in 2025, per the announcement. The October expansion makes it available globally in English — still English-only for now, and sign-in is required, per MacRumors.
How watermark detection works
The key difference from the "AI detectors" you've probably tried: SynthID doesn't analyze a file's style and guess whether it looks AI-made. It's checking for a cryptographic-style signature that was deliberately embedded at creation time. That's why it's far more reliable than guesswork tools — but only within its coverage area.
According to Google's announcement (via Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind):
- Images and video: the watermark signal is encoded directly into pixel values. Uploads supported include JPG, PNG, HEIC, and GIF images, and MP4, MOV, and WEBM video.
- Audio: the signal is embedded in the waveform (used in tools like Lyria and NotebookLM). Supported uploads include MP3, WAV, and FLAC.
- Text: large language models embed a signal through token-probability scoring, subtly influencing generation without changing readability — though the public portal focuses on media uploads.
The scale Google claims is enormous: since 2023, SynthID has watermarked over 180 billion images and videos and 240,000 years of audio. Google also says verification built into Search, the Gemini app, and Chrome now handles over 1 million verification requests daily. (These are Google's own figures, not independently audited.)
What it detects — and who's covered
The October 7 expansion is a big deal partly because of who else is now covered. Per Google's announcement, the Detector now reads SynthID watermarks from:
| Provider | What's covered | Status |
|---|---|---|
| Gemini, Veo, Lyria, Nano Banana, Imagen | Covered ★ | |
| OpenAI | ChatGPT images, DALL·E, OpenAI API output | Covered (new) ★ |
| NVIDIA | Watermarked partner output | Covered (new) ★ |
| Kakao | Watermarked partner output | Covered (new) ★ |
| Apple | Planned support | Coming soon |
| Meta, Microsoft | Run their own watermark systems | Not covered |
| Hailuo, Runway, Kling, Pika | Popular AI video generators | Not covered |
OpenAI's adoption is the headline addition: images generated through ChatGPT and DALL·E now embed SynthID watermarks, making this the first watermarking standard shared across multiple major AI platforms — per Google's announcement and independent coverage.
The honest limits: what it can't do
This section matters more than the feature list, because the tool's name invites a dangerous assumption: that a clean scan means "human-made." It doesn't.
- It's not a general AI detector. Google says so explicitly in the portal's FAQ. It only recognizes SynthID watermarks. If a clip was generated by a tool that doesn't watermark with SynthID, the portal shrugs — and plenty of the most viral AI video generators don't.
- A negative result proves nothing. No watermark detected does not mean no AI was involved. The watermark could be absent because the generator never embedded one.
- It can't tell AI-created from AI-edited. Per MacRumors, the tool cannot distinguish between fully AI-generated media and media that was merely edited or touched up with AI.
- Coverage gaps are exactly where creators feel pain. Independent analysis notes that viral clips from popular AI video tools (Hailuo, Runway, Kling, Pika) carry no SynthID — and Meta and Microsoft run their own separate systems. Those are some of the most common sources of misleading viral media.
- English only, sign-in required. The portal is global but English-only for now, and you must sign in to use it.
The bottom line: treat this as a positive-only tool. A detected watermark is strong evidence a file came from a participating AI generator. The absence of a watermark tells you almost nothing. Pair it with source-checking, not instead of it.
How to use it
Based on Google's announcement and launch coverage, the flow is simple:
- Open the SynthID Detector portal (Google's verification site for SynthID watermarks) and sign in.
- Upload a file — an image (JPG, PNG, HEIC, GIF), video (MP4, MOV, WEBM), or audio (MP3, WAV, FLAC).
- Read the result. The portal checks for the invisible SynthID signal and reports whether a participating generator's watermark is present.
- Interpret correctly. Watermark found = strong evidence of AI generation by a covered provider. No watermark = inconclusive, not proof of authenticity.
Pricing note: the portal is free to use — no paid tier has been announced. Availability details come from Google's October 7, 2026 announcement; check the portal itself for the current supported formats and providers.
Use cases for creators
This isn't a niche research tool — it solves real, weekly problems for working creators:
- Vetting viral clips before you repost. That "unbelievable" video in your group chat? Run it through the Detector before it becomes your next post's centerpiece — or its embarrassment.
- Provenance for client work. If you deliver AI-assisted visuals or voiceovers to clients, a clean watermark check (plus your own disclosure) is documentation that your pipeline is traceable.
- Protecting your audience from deepfake scams. Scam ads increasingly use AI-generated celebrity faces and voices. Creators who warn their communities can now point to a free, first-party verification tool.
- Checking your own outputs. If you generate thumbnails, B-roll, or music with Gemini, Veo, or ChatGPT image tools, verifying the watermark survives your export settings confirms your content stays transparent downstream.
It pairs well with the verification features Google already built into Search, the Gemini app, and Chrome, which process over a million checks daily — for files you encounter in the wild rather than ones you upload deliberately. For more on AI workflows, see our best AI writing tools roundup.
Pros and cons
Pros
- Free, from a first-party source — no sketchy third-party "AI detector" with a paywall
- Checks actual embedded watermarks, not vibes — far more reliable than style-guessing detectors
- Covers OpenAI output too — the first cross-platform watermark standard
- Works on images, video, and audio in one portal
- Backed by enormous watermarked corpus (180B+ images/videos per Google)
- Complements built-in checks in Search, Gemini, and Chrome
Cons
- Not a general AI detector — misses every generator outside SynthID
- A clean scan proves nothing; false confidence is the real risk
- Can't distinguish AI-created from AI-edited media
- Sign-in required; English-only for now
- Apple support still "coming soon"
- We haven't hands-on tested the public portal yet
SynthID vs generic "AI detectors"
| SynthID Detector (Google) | Generic "AI detectors" | |
|---|---|---|
| Method | Reads embedded watermark ★ | Guesses from style/statistics |
| Reliability when positive | High — cryptographic-style signal ★ | Low — false positives are common |
| Coverage | SynthID adopters only | Attempts everything (unreliably) ★ |
| Meaning of negative result | Inconclusive — be careful | Claimed as "human" (often wrong) ★ |
| Price | Free ★ | Often freemium/paywalled |
| Media types | Image, video, audio ★ | Usually text only |
The honest read: for covered providers, SynthID is the most trustworthy check available to the public. For everything else, nothing has replaced old-fashioned source verification — who posted it, where it first appeared, and whether it passes basic scrutiny.
Our verdict
Early verdict: genuinely useful, dangerously easy to over-trust
SynthID Detector is the best free public AI-content check available right now — a first-party watermark reader covering Google and OpenAI output, with the scale to matter. Every creator who reposts viral media should bookmark it. But its name is a trap: it's not a general AI detector, a clean scan proves nothing, and some of the most common sources of viral AI clips aren't covered at all. Use it as a positive-evidence tool alongside source-checking, never as a "human-made" stamp. We'll update this review after hands-on testing.
FAQ
When did Google open SynthID Detector to the public?
October 7, 2026. It was previously limited to journalists, media professionals, and researchers who received access at Google I/O in 2025.
Is SynthID Detector free?
Yes — the public portal is free to use. You do need to sign in with a Google account.
What file types does it check?
Per launch coverage: JPG, PNG, HEIC, and GIF images; MP4, MOV, and WEBM video; MP3, WAV, and FLAC audio.
Can it detect AI images from ChatGPT?
Yes, if they carry the SynthID watermark. OpenAI has adopted SynthID, so images generated through ChatGPT, DALL·E, and the OpenAI API are covered.
If no watermark is found, was the file made by a human?
No. Google explicitly says this is not a general AI detector. A negative result only means no SynthID watermark was found — the file could still be AI-generated by a tool that doesn't use SynthID.
Can it tell whether a video is fully AI-generated or just AI-edited?
No. Per MacRumors' launch coverage, the tool cannot distinguish between AI-created and AI-edited media.
Does it detect video from tools like Runway or Kling?
No. Those popular AI video generators don't use SynthID, and Meta and Microsoft run their own separate watermark systems. Apple support is listed as coming soon.