Google opened a public SynthID verification website on 6 October, allowing anyone to check an image, video or audio clip for AI-generated content, TechCrunch reported on 7 October. The company had offered the tool for testing to selected journalists, media professionals and researchers at Google I/O in 2025. The website now makes that check available beyond those testers and Google’s own apps.
Key points
- The public website accepts image, video and audio files.
- Google uses SynthID watermarks on media generated by its named models and tools.
- OpenAI, Nvidia and Kakao also support SynthID; Google says people make 1 million verification requests a day.
SynthID moves beyond Google I/O testers
Google introduced SynthID in 2023 to watermark AI-generated media. A watermark is an identifying mark placed in the material when it is made, rather than a label someone attaches when passing the file along. Verification looks for that mark later, much as a mark on a manufactured object can identify where it came from. Google’s Nano Banana, Veo and Lyria models use SynthID on the media they generate, as do its generation tools Gemini, Flow, ProducerAI and Vids.
Checking a picture before passing it along could therefore involve submitting the file for a watermark check, rather than deciding from its appearance alone whether it was made with AI. The same reasoning would apply to a clip or recording, provided the media carried a watermark that the check could identify. That condition matters: the reported use of SynthID by particular models and tools is narrower than a claim about every AI-generated file someone might encounter.
The move from selected testers to a public website changes who can make the request, without changing when a watermark was put in the media. For a company building a generation tool, adding SynthID would concern the output as it is created; for someone checking that output, the new site provides a separate place to examine it. TechCrunch reported that OpenAI, Nvidia and Kakao also support SynthID, while OpenAI runs its own content-checking site. Support by those companies does not, on its own, describe which of their outputs carry the mark.
Image, video and audio files reach the site
The website accepts JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF and GIF images. Its video formats are MP4, MOV and WEBM, and its audio formats are WAV, MP3, OGG, FLAC, AAC and M4A. Those lists describe the kinds of files that can be submitted. The watermark check concerns the content within an accepted file: a familiar file extension is not itself evidence that SynthID was added when the media was generated.
Google has also put SynthID verification in the Gemini app and Chrome. The website gives that check a destination of its own, alongside those existing routes. Google says people currently make 1 million requests to verify content every day. That is the company’s figure for verification requests, rather than a count of visits to the new website.
The range of accepted files matters differently from the range of participating generators. An audio recording in a supported format can be submitted to the website, while Google’s account of watermarking names Lyria among the models that add SynthID to generated media. In the same way, accepting a video file permits a check of that file; the watermarking described for Veo concerns media the model generates. Submission and identification are separate parts of the process.
OpenAI supports SynthID alongside its own checker
OpenAI’s support for SynthID sits alongside its separate checking site. It is another example of verification depending on both a way to mark generated material and somewhere to check it afterwards. AI Affairs has also reported on OpenAI’s plans for EU text watermarks for ChatGPT and Codex, with API use on an opt-in basis. Those plans concern text, while Google’s new website handles image, video and audio files.
Google’s watermarking work extends beyond those three media formats. AI Affairs previously covered Google DeepMind’s introduction of SynthID Bio for AI-designed proteins. The public verification website, by contrast, is built around files people can submit as images, videos or recordings.
Other watermarking systems face a detection problem even with content from the organisations that made them. Microsoft and Meta have their own standards for marking and checking media, but their tools often fail to identify material generated by their makers’ models, TechCrunch reported. A watermarking standard and a verification tool can therefore exist together without every check identifying the content they were built to recognise.
