Google has expanded the functionality of the Gemini app by introducing an option to verify videos generated or edited with its own artificial intelligence models. Users can upload a video and ask the app directly if it was made using Google AI, according to TheVerge.
Watch the latest TechRider.ro video productions
Table of Contents
– article continues below –
To make this check, Gemini analyzes both the visual and audio components of the file. The application searches for a proprietary watermark called SynthIDdeveloped by Google specifically for identifying AI-generated content.
How SynthID detection works
According to the company, the answer provided is not limited to a simple “yes” or “no”. Gemini also indicates the exact moments in the video or audio track where the watermark is detected. Google claims that SynthID is an “imperceptible” watermark, designed so that it does not affect the viewing experience.
Limitations of watermarking technology
However, it remains unclear how easily it can be removed, or whether other platforms will be able to identify and use this information to clearly flag AI-generated content. In the past, other markup solutions were removed relatively easily, and the lack of a common standard across platforms continues to be a problem.
Extension from images to videos
The company launched similar functionality for images in November, but even then the check was limited to content created or modified with Google AI. The same limitation applies to videos: Gemini cannot confirm whether a material was generated by other AI models outside the Google ecosystem.
C2PA metadata and labeling challenges
In parallel, Google also uses metadata (information about who and how the content was created) of type C2PA (an international initiative for authenticating digital content, which uses metadata to show who created or modified a material and with what tools) in certain image generation models within Gemini. However, the lack of coordinated labeling at the level of social networks means that fake material or deepfakes can circulate without being properly flagged.
Technical specifications and availability
The new verification feature supports videos up to 100 MB and a maximum length of 90 seconds. This is available in all languages and regions where the Gemini app is already accessible.