Google’s SynthID Detector Checks Images, Video, and Audio for AI Watermarks

Google is expanding access to its SynthID detector, a tool that checks images, video, and audio for supported AI watermarks. The service is available to everyone, but users must sign in with a Google, OpenAI, or Apple account before they can use it.

Google says each user will receive a daily usage quota of approximately 10 image, video, and audio checks. That limit is similar to restrictions on the company’s earlier detection tools.

Google engineers told Ars that access must be restricted to prevent people from using the detector to develop tools for removing SynthID watermarks. The exact limit may vary, particularly when users submit multiple files that are very similar. That behavior could suggest an attempt to refine a watermark-bypass workflow.

SynthID cannot identify every AI-generated file

Making SynthID detection easier to access is useful. It gives people a convenient place to check media for watermarks associated with several popular AI tools. However, Google and its partners are not the only sources of AI-generated content.

Some model providers use different watermarking methods, including Meta, while others do not watermark their output at all. Open-weight models can also run on a user’s own hardware and generate increasingly realistic content without adding even basic metadata labels.

As a result, much AI-generated content will not be identified by SynthID. Google’s website includes a similar disclaimer. It is also possible that someone could eventually find a way to remove SynthID, making existing AI-generated media appear more legitimate.

SynthID is therefore not a perfect AI detector. Its effectiveness depends on AI companies proactively adding watermarks to content—a particularly challenging requirement as synthetic images, video, and audio continue to spread across the internet.

Why content credentials may be a better long-term solution

Another approach is to use encryption-based systems such as C2PA to label images, video, and audio at the moment they are created. Google’s Pixel smartphones are among the few devices that currently do this.

Until content credentials become more widely adopted, SynthID offers one of the more convenient ways to check media for supported AI watermarks. Just remember that a negative result does not prove that a file was created by a person.