October 12, 2026, 1:34 pm | Read time: 2 minutes
Images, videos, and voices today are often hard to distinguish from real recordings. A new tool from Google aims to help find clues to artificially generated content.
With increasingly powerful AI systems, it is becoming more difficult to recognize manipulated or entirely artificially generated content. Whether it’s a photo, video, or voice recording, the results often appear so realistic that distinguishing them with the naked eye is hardly possible. This is where Google’s “SynthID Detector” comes in.
The company introduced the tool back in 2025. At that time, only selected journalists and researchers could access it through a waiting list. Now, the service is publicly available, offering users another way to more easily identify AI-generated content.
Invisible Watermarks as Digital Markers
The foundation of the service is digital watermarks. These are invisible markers embedded directly into files. According to Google, this technology has been in use since 2023. The company reports more than 180 billion images and videos have already been marked accordingly. Additionally, there are audio recordings with a total duration of around 240,000 hours.
The “SynthID Detector” specifically searches uploaded files for such traces. If a matching watermark is found, it indicates creation or editing with artificial intelligence. Besides Google’s own markers, the system also recognizes watermarks from OpenAI, Nvidia, and Kakao, according to the company. Apple is expected to be supported in the future.
Also of interest: Google cuts large parts of the free version of Gemini
The Tool Also Has Clear Limits
Usage is relatively simple. After signing up on the SynthID platform with a Google, Apple, or ChatGPT account, images, videos, or audio files can be uploaded. The service then checks if a known AI watermark is present.
However, the result is not a reliable proof of the authenticity of content. If the appropriate watermark is missing, it does not automatically mean that no AI was used. Conversely, a found indication is not proof of misinformation. AI-generated graphics can indeed show correct content. Therefore, it is advisable to also look at other sources. For images, a reverse search can help trace the origin and previous publications of the recording.