Anyone with a Google, OpenAI or Apple account can now drag a photo, a video clip or an audio file onto a web page and learn whether it carries SynthID, the invisible watermark Google builds into media made by its AI tools. The checker lives at synthid.com, costs nothing, and opened to the general public on Oct. 7, 2026 after more than a year as a waitlist tool for reporters and researchers.
The page does one narrow job, and Google’s own copy says so in plain terms: “This is not a general AI detector,” and it can detect only media from the listed companies that have adopted SynthID technology.
SynthID checks for four companies’ watermarks
SynthID began as Google’s marker for its own generators. Google DeepMind’s announcement now describes a detector that also reads the watermarks of partners, and Engadget’s report lists the companies as Google, OpenAI, NVIDIA and Kakao. Apple has not yet joined: Google says support for content from Apple Intelligence is coming “soon,” with no date given, and ElevenLabs has committed to adding SynthID to its audio.
That list defines the tool’s reach. A picture from a generator outside those four companies, or one made with a model that never embedded SynthID, returns nothing, and a blank result means only that this particular signal was absent. Open-weight models can generate media with no watermark at all, which puts a whole category of synthetic content outside what any SynthID check can see, however many partners sign on.
Account sign-in, file types and daily scans
The site accepts images, video and audio. Search Engine Journal reports that it opens to everyone worldwide, in English only, from the day of the announcement. Users sign in with a Google, OpenAI or Apple account, which Android Headlines reported is meant to keep the detector from being used to train watermark-removal tools, and the same account has a cap of about ten scans a day.
Google had separately built SynthID checks into other products. Circle to Search and the Gemini app already look for SynthID watermarks and C2PA content credentials, and The Next Web notes that Google counts more than one million verification requests a day across Search, Gemini and Chrome. The dedicated website matters for people who never open a Google AI product but still want to test a viral clip.
Result wording and the limits of one watermark
Results are phrased as “made or edited with Google AI,” and the tool does not separate fully generated content from a real photo that was touched up with an AI feature. Its sample result adds that the media may have been edited further since the watermark was applied. Google’s own FAQ states that “no signal is completely robust” and recommends “collecting multiple points of data before making a decision.”
Accuracy also depends on what is uploaded. Android Headlines reports the site takes standard JPG and HEIC images, MP4 video and uncompressed WAV audio, and the site’s guidance is to submit the best-quality copy available, since heavy compression or re-saving can degrade a signal that was embedded at creation. A screenshot of an AI image is a weaker test than the original file.
Watermarks can also be stripped in some cases, and Meta’s competing checker covers only images and video made with its latest Muse Image models. The effect is a patchwork: each company’s marker answers only for that company’s output, and the new site stitches four of them together behind one upload box.
SynthID Detector leaves the journalist waitlist behind
The earlier version was announced on May 20, 2025 by Pushmeet Kohli, vice president of science and strategic initiatives at Google DeepMind, in a post introducing the portal. At that stage it was open to testers, with journalists, media professionals and researchers invited to join a waitlist, and Google said more than 10 billion pieces of content had been watermarked with SynthID. Video and text detection were still described as coming in the following weeks.
The 2025 portal also promised more granular output than the public site gives. For images it was designed to highlight the areas where a watermark was most likely, and for audio to pinpoint the segments where one was detected. The public page returns a verdict on the file as a whole, and Android Headlines notes it will not break down which parts of an image were edited, a trade that favors a fast answer for casual users over the forensic detail a newsroom might want from the same watermark.
The 2026 launch removes the waitlist and puts SynthID Detector at synthid.com, where the only gate is the sign-in. Google says it is continuously working to extend its partnerships so the detector can recognize more generated media, which means the four-company list is the one to check against any file before reading its result.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- Hybrids have 15% fewer problems than gas cars, while EVs and plug-in hybrids have about 80% more, Consumer Reports finds
- Regulators cleared the first U.S. small modular reactor, 4 months early
- Lake Powell sank to a record 3,517 feet, nearing the level that stops Glen Canyon Dam’s turbines
- Seven of Earth’s nine planetary boundaries are breached, and all seven are worsening
