Until now, people outside Google’s trusted testing program had limited ways to verify whether media contained SynthID. Users could ask Gemini to inspect content, but that process could produce more explanation than a simple verification result required.
The new SynthID.com site provides a more direct workflow. Users upload a file and receive a message indicating whether SynthID was detected as part of the media’s generation or editing process.
The broader partner support also addresses a limitation of Google’s earlier detector. Previously, Google’s tool primarily recognized its own implementation of SynthID, even though other companies were also using the technology. The updated version is designed to recognize the distinct watermarks used by participating partners.
Google says OpenAI, Nvidia and Kakao are among the companies supported, while Apple is expected to be added later. That means media created with participating AI systems can be checked from one location rather than through separate company-specific tools.
The public website offers less detail than Google’s internal detection system. Internal tools can identify areas of an image where the watermark signal appears, while the consumer version provides a simpler result stating whether SynthID was involved.
Users must sign in before using the service. Google supports logins through Google, OpenAI and Apple accounts.
The company is also limiting how frequently the detector can be used. Google says users receive a daily allowance of about 10 checks across images, video and audio.
According to Google engineers, those limits are intended to make it harder for people to repeatedly test modified files while developing techniques to remove or defeat the watermark. The number of available checks may also vary if a user submits multiple files that are very similar.
The new detector does not identify every form of AI-generated media. Some companies use different watermarking systems, while others do not embed watermarks at all. Open-weight models running on local hardware can also create synthetic media without SynthID or other built-in labeling.
Google is making that limitation clear on the site. A file that does not contain SynthID cannot automatically be treated as authentic, because it may still have been generated by an AI system that does not use the technology.
The watermark itself also depends on remaining intact. If techniques are developed that can reliably strip or damage SynthID, some AI-generated media could potentially pass through the detector without being flagged.
Google has separately supported approaches that authenticate original media rather than only marking synthetic content. Its Pixel phones are among the devices using C2PA-based credentials to attach provenance information when media is created.
For now, the new SynthID website gives users a single public tool for checking supported AI-generated images, video and audio. Its usefulness will depend on how widely SynthID is adopted, since the detector can only recognize content carrying one of the compatible watermarks.
This analysis is based on reporting from ars TECHNICA.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
