Google opens its SynthID detector to everyone, adds partner watermarks

Google has spent years stamping invisible watermarks on everything its Gemini models produce. Now it is letting the rest of us check for them.

The company announced that its SynthID detector is available to anyone through a new dedicated website. It now recognizes watermarks from every partner using the technology, not just Google’s own.

What changed with the SynthID detector

SynthID works by hiding a signal inside the content itself. For images and video, the marker lives in the pixels. For audio, it is woven into the waveform.

The scale is large. Google says Gemini alone has applied SynthID to over 100 billion images and videos, along with 60,000 years of audio.

Until this update, people outside Google faced a clunky process to confirm whether a file carried a SynthID tag. The new site removes that friction. You bring the file, the detector looks for the hidden signal, and the answer no longer depends on which partner’s model made the content.

That partner coverage matters. SynthID has picked up support from several major AI companies, including OpenAI, ElevenLabs, Kakao, and Nvidia.

How SynthID got here

Google DeepMind launched SynthID in 2023. It started as an image watermark and later stretched to cover text, video, and audio. The text version was open-sourced in 2024, which let outside developers use it in their own systems.

The detector has existed in some form since May 2025. Back then, Google opened a dedicated SynthID Detector portal, but access was limited to early testers and journalists.

Google widened the net at its I/O event in May 2026. The company folded SynthID checks into Google Search features such as Lens and AI Mode, and into Chrome. That let users ask directly whether a piece of content was AI-generated.

The verification feature inside the Gemini app had been used 50 million times before that May update.

Google has also brought C2PA Content Credentials into the picture. C2PA is an industry standard that attaches provenance data to files, describing where content came from and how it was edited. Pairing visible metadata with invisible watermarks gives verification two layers instead of one.

Apple has planned integration of SynthID into image tools in iOS 27. Google has also announced SynthID Bio, a version aimed at watermarking AI-designed protein sequences, slated for September 30, 2026.

What this means for AI and media

Opening the detector to everyone changes that dynamic. Journalists, fact-checkers, educators, and regular users can now test suspicious images or audio without asking permission.

Limits remain. A watermark detector can confirm that a tagged file came from a participating model. It cannot flag content from AI systems that never adopted SynthID in the first place. A clean result means no SynthID signal was found, not that the content is definitely human-made.

The next markers to watch are the planned iOS 27 integration and the September 30, 2026 launch of SynthID Bio. Both would carry the technology well beyond Google’s own ecosystem.