AI watermark checks are now open to everyone. Google opened its SynthID Detector worldwide in English on October 7, 2026, giving users a public way to inspect images, video, and audio for embedded AI watermarks.
The portal at synthid.com checks for watermarks created by Google and partners OpenAI, NVIDIA, and Kakao, with support for Apple’s AI-created and edited content coming soon. The service is free, but users must sign in with a Google, OpenAI, or Apple account—because apparently even suspicion now needs an account.
Google first introduced an early version of the Detector in 2025 for journalists and media professionals. The company has worked on the tool for more than a year, and the public release expands access beyond the professional groups that previously received it.
What SynthID can and cannot verify
SynthID watermarks embed a near-invisible marker into AI-generated media at the moment of creation. The marker sits inside the content rather than its metadata, and Google designed it to remain detectable after common edits such as cropping, filters, frame-rate changes, and compression.
For audio, the watermark is inaudible and designed to withstand noise, MP3 compression, and playback-speed changes. For text, SynthID adjusts the probability scores of words to create a watermark that users cannot notice and that does not affect output quality, although the public Detector checks images, video, and audio.
The Detector supports JPG, PNG, WEBP, HEIC, MP4, MOV, WEBM, WAV, MP3, OGG, FLAC, AAC, and M4A files. Its result is blunt: “watermark detected” or “watermark not detected.”
That second result is not a declaration of human authorship. Google warns that a negative result does not rule out AI generation because different companies use different watermark standards, and watermarks can be stripped. The company’s own FAQ puts the boundary plainly: “This is not a general AI detector.”
A large watermarking system meets a public test
Google launched SynthID in 2023 and says it has since watermarked more than 180 billion images and videos, along with 240,000 years of audio content. A May 19, 2026 announcement cited more than 100 billion watermarked images and videos and 60,000 years of audio, showing how the reported totals have grown over time.
The public Detector joins built-in verification tools in Search, the Gemini app, and Chrome. Those features handle over 1 million requests each day, while SynthID verification in the Gemini app had reached 50 million uses worldwide as of May 19, 2026.
Google expanded SynthID verification to Search on May 19, 2026, then brought it to Chrome over the following weeks. The global Detector now adds a separate place for users to upload media instead of relying only on verification features built into Google’s products.
That makes SynthID useful for checking content from participating systems, but not for settling every dispute about whether a file came from AI. It can identify a watermark when one survives and when the relevant system uses a supported standard. It cannot turn the entire internet into a laboratory for proving origin.
Support for Apple’s AI features is still coming, and the Detector already depends on watermarks from Google, OpenAI, NVIDIA, and Kakao. The partnership list matters because detection becomes more useful when several companies place compatible markers in their generated media rather than each building an isolated signal.
For now, SynthID offers a practical test with a narrow promise: it can look for known watermarks in supported images, video, and audio. That is less glamorous than universal AI detection, but it is also a claim the tool can actually make.
Based on
