Google opened its SynthID Detector to the public on October 7, giving any user the ability to check image, video, and audio files for the company’s invisible AI watermark. Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind, announced the global expansion in a company blog post, aiming to give internet users clear context about the digital media they encounter online.
Global Public Launch at SynthID Dot Com
The free verification tool now operates at synthid.com alongside built-in checkers already embedded in Google Search, Chrome, and the Gemini app. Prior to this public rollout, access was limited to an early version distributed to journalists, media professionals, and researchers who joined a waitlist. Users can upload various file formats directly to the site, including JPG, PNG, WEBP, GIF, and HEIC for images, MP4, MOV, and WEBM for videos, and MP3, WAV, and FLAC for audio files.
Detector Partnerships Across Generative AI Platforms
The system reads imperceptible watermarks embedded directly into pixels or audio waveforms rather than interacting with file metadata. According to Google’s announcement, the detector identifies media generated by Google alongside participating partners such as OpenAI, NVIDIA, and Kakao. ElevenLabs also supports SynthID. Apple is slated to join the partner network soon, though no official release date has been set. Meanwhile, Apple is also testing its own reference image method in the iOS 27 beta to connect pictures directly to originating camera sensors.
SynthID detects AI edits but cannot distinguish synthetic images
Google emphasizes that synthid.com functions specifically as a targeted verification checker rather than a general-purpose AI detector. The interface returns results stating either that a SynthID was detected or that no signal was found, accompanied by a label indicating whether the media was made or edited with participating AI systems. Because the detector flags media that was either created from scratch or merely touched up using supported tools, a touched-up photo and an entirely synthetic image can yield the same result. A negative result only indicates the absence of a SynthID signal and does not definitively prove a human created the file.
Google provides details on SynthID usage and accuracy
What file quality gives the best detection accuracy?
Google’s site advises users to upload the highest-quality copy of a file available to ensure the most accurate watermark detection result.
How many requests do Google’s existing AI verification features handle?
Built-in verification features across Google Search, Chrome, and the Gemini app regularly handle over 1 million requests daily, with the Gemini app check having been used 50 million times before May.
How much content has Google watermarked with SynthID so far?
Google has applied its invisible watermark to over 180 billion images and videos, alongside 240,000 years of audio content.
Worth a look