Overview
SynthID Detector is the verification portal Google opened to everyone on October 7, 2026 at synthid.com. Upload an image, a video or an audio file and it tells you whether it can find a SynthID watermark, the invisible mark Google and its partners embed at the moment content is generated.
The portal itself is not new. It first appeared in waitlist form at Google I/O in May 2025, limited to journalists, researchers and media professionals. What changed is that the door is now open: anyone can sign in from a browser and upload a file. Google puts the scale in context by noting that since SynthID launched in 2023 it has watermarked more than 180 billion images and videos along with the equivalent of 240,000 years of audio, while the verification features already built into Search, the Gemini app and Chrome handle over a million requests a day.
The boundary needs stating up front. Google writes directly on the portal that this is not a general purpose AI detector. It only recognizes content from companies that adopted SynthID, which today means Google, OpenAI, NVIDIA and Kakao with Apple listed as coming soon. Something made by any other model, even thoroughly synthetic, will not trigger an alert.
Key Features
- Watermarked at generation: SynthID has been in use since 2023, writing an imperceptible mark into the content itself the moment an AI creates it. Neither eye nor ear can detect it, but a detector can read it. Unlike metadata that gets stripped by accident or on purpose, it travels with the content and usually survives sharing and common compression
- Three modalities in one upload: The current version accepts images (JPG, PNG, HEIC, GIF), video (MP4, MOV, WEBM) and audio (MP3, WAV, FLAC). Text is not in the format list for this public release
- It marks which parts carry the watermark: The result is more than yes or no. When a watermark is found it indicates which portions of the file carry it, and for video it lays out the video track and audio track separately on a timeline. A partly synthetic file behaves very differently from a fully generated one, and that distinction is the useful part
- A login gate: You need a Google, Apple or OpenAI account to upload, with no plain email sign-up option. Commentators have pointed out that two of those three belong to companies whose output is among the things being checked
- A partner list, not the whole web: It recognizes content generated or processed by Google, OpenAI, NVIDIA and Kakao, with Apple listed as coming. Microsoft and Meta each run their own watermarking and verification standards and appear nowhere on the list, so content from their models reads as nothing
- Limits stated in public: The page says plainly that it is not a general purpose AI detector, and that a clean result only means no supported watermark was found, which is not evidence the content is human made. Google recommends uploading the highest quality file available and acknowledges the watermark can be circumvented
Use Cases
- Checking whether an image or clip of unknown origin carries a partner watermark before you share it
- Newsrooms and fact-checking desks using the portal as a fast first pass alongside source, date and context
- Teachers, recruiters and platform reviewers who want one more checkable signal on suspicious material
- Provenance researchers observing how much of a watermark survives cropping, compression and screen recording
Pros
- Free and global, usable straight from a browser with no waitlist to apply for
- Covers image, video and audio, with track-level detail on video
- Points to the specific parts carrying the watermark, separating full generation from partial synthesis
- The mark is written into the content at generation, so it survives transfer better than metadata
- Backed by 180 billion watermarked images and videos, so hits inside the partner list mean something
- Google states the limits itself, putting the not-a-general-detector line in a visible place
Pricing
Completely free, with no tiers and no quota. A Google, Apple or OpenAI account is required to sign in, and that is the only barrier. It is available worldwide with an English-only interface for now.
Summary
SynthID Detector works better as a specific key than as a lie detector. It opens one particular lock: the watermark Google and its partners write into a file at generation. When the key fits you learn something concrete, and when it does not you learn nothing at all.
The real change here is reach rather than capability. In the waitlist era the tool sat with journalists and researchers, and now anyone can check a file before passing it on, which raises the cost of passing off fabricated media. Combined with the checks already inside Search, Gemini and Chrome, Google has pushed verification out of a professional workflow and into an everyday action.
Three things are worth holding onto when you use it. A negative result reads as unknown, not as human made, which Google says on the page itself. Vendors outside the partner list are invisible to it, Microsoft and Meta run their own standards, and anything from a local open weights model has no mark to find. Apple is the next marker to watch: iOS 27 began rolling out on September 14, images from Image Playground already carry a SynthID watermark, and yet Apple is not counted as an official partner, so closing that gap would noticeably widen what a single upload can confirm.
Treat it as one check among several. Source, date and context still have to be examined by hand. SynthID just makes one of those steps verifiable.
Version History
- Detector portal opens to the public (2026-10-07): Google expanded SynthID Detector from a waitlist for journalists, researchers and media professionals into a verification portal open to everyone worldwide at synthid.com, English only for now. It supports images (JPG, PNG, HEIC, GIF), video (MP4, MOV, WEBM) and audio (MP3, WAV, FLAC) with text excluded from this release, highlights which parts of a file carry the watermark, and shows video and audio tracks separately on a timeline. Coverage covers content from Google, OpenAI, NVIDIA and Kakao with Apple listed as coming, while Microsoft and Meta remain outside the partner list. Google also reported more than 180 billion watermarked images and videos and the equivalent of 240,000 years of audio since 2023, over a million daily verification requests through its built-in features, and stated clearly that the portal is not a general purpose AI detector and that a clean result does not prove content is human made