Google Opens Its SynthID Detector to Everyone
Google has opened SynthID Detector to the public at synthid.com, where anyone can submit an image, video or audio clip to check for AI generation. TechCrunch The launch was reported on Oct. 7, 2026, and moves the tool out of limited testing.
Previously, access was restricted to select journalists, media professionals and researchers. General availability removes that gate and puts the same check in reach of any user with a browser.
The site accepts common production and distribution formats. For images: JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF and GIF. For video: MP4, MOV and WEBM. For audio: WAV, MP3, OGG, FLAC, AAC and M4A.
SynthID dates to 2023, when Google introduced it to identify whether media is AI-generated. It is a digital watermarking technology that embeds a hidden signal in AI-made content, imperceptible to people but readable by software, like an invisible barcode for files. Google SynthID Detector is the verification portal that checks files for that signal. Google
Coverage inside Google spans models and tools. The Nano Banana, Veo and Lyria models use SynthID to watermark media they generate, as do the Gemini, Flow, ProducerAI and Vids tools. Outside Google, OpenAI, NVIDIA and Kakao support SynthID, with Apple reported to be adding support soon. Verification is also built into the Gemini app and Google Chrome. Google said people currently make 1 million requests per day to verify content with SynthID.
Google first announced the Detector portal at I/O. Google At that time, it was also described as able to check a snippet of text to determine whether it was created using Google's AI tools. Yahoo Finance The public site now described by Google is positioned around image, video and audio upload and checking.
The broader context here is scope. A watermark detector answers a narrow question well: did this file carry a SynthID mark. It does not answer whether AI was involved in a file that never received one, or in one where the mark was removed upstream. A positive result is actionable. A negative result is not conclusive. It means no watermark was found, not that no synthesis occurred. Triage workflows, moderation queues and editorial checks will need to treat those two outcomes differently.
In my view, the list of supporters matters more than the portal itself. Watermarking only becomes useful infrastructure if competing generators mark content in ways that common detectors can read, and if verification lives where content is viewed and shared. Support from OpenAI, NVIDIA and Kakao, reported work toward Apple support, and native verification in Chrome and the Gemini app point in that direction. The practical test will be routine handling. Professionals know files get re-encoded, trimmed, captioned and re-uploaded across services. At public launch, a single URL and a stated volume of a million verifications per day give the ecosystem something concrete to build around. If that shared marking and checking layer holds, provenance work shifts from specialist forensics to everyday tooling.
