SynthID opens globally, but it verifies a mark rather than authenticity

Google has opened SynthID Detector to anyone on a dedicated website, globally and for now in English. The tool checks images, video and audio for imperceptible watermarks from Google and participating partners.

One website now reads watermarks from several vendors

SynthID launched in 2023, and Google says it has since watermarked more than 180 billion images and videos plus 240,000 years of audio. The new website can look for marks associated with Google as well as partners OpenAI, NVIDIA and Kakao. Apple support is due later.

The earlier detector targeted media professionals, while checking partner marks remained fragmented. The public site consolidates that step. Ars Technica reports that it requires a login and gives each user an approximate daily quota of 10 image, video or audio checks.

A shared detector turns watermarking into usable infrastructure

A watermark has limited value when only its maker can read it inside its own product. Support from several companies gives newsrooms, moderators and investigators one place to check supported media without switching among separate detectors. Google also says verification features in Search, Gemini and Chrome now handle more than 1 million requests a day.

The second layer is standard setting. Google is promoting more than its own generator. It is operating a reader for a partner ecosystem. Whoever controls that common verification point gains influence over which watermarks become practically visible to platforms and users.

A negative result is not a certificate of human origin

The detector reports whether supported SynthID technology was used to create or edit media. According to Ars Technica, the public version does not provide the detailed map of image regions shown by an internal tool. More importantly, it searches for a particular mark rather than general evidence of any AI system.

A file from an unsupported generator, an older model without watermarking or content altered later can therefore produce no match. That result means only that the detector did not find a supported SynthID mark. It does not identify the creator or prove that the depicted event occurred.

Robust marks and transparent results will determine trust

The useful test is how detection survives compression, cropping, transcoding, editing and other ordinary transformations. False results, a transparent list of supported systems and a way to check more than 10 files a day will matter as well.

Apple's promised support and the arrival of more vendors are worth watching. If the partner list expands and results can safely enter newsroom or platform workflows, SynthID can become a provenance layer. A green message on a website alone cannot become a universal truth detector.

Lilith's verdict

SynthID can find the signature of a known machine on a file. When the signature is absent, the detector has not found a human, only someone without a readable name badge.