Google Gemini Gets New Image Verification Tool to Detect AI-Generated Content
Google has expanded the capabilities of its Gemini AI service by adding an image-verification tool to both the mobile app and the web version, designed to determine whether a picture was generated automatically. The introduction of this feature is a logical step: visual content is increasingly produced with AI models, and the demand for distinguishing real photographs from synthetic ones continues to grow.
The new detector is based on the SynthID system — invisible digital watermarks introduced in 2023. These markers are embedded into images created by Google’s generators and persist even after resizing or partial editing. For this reason, the verification tool works only with content produced specifically by Google’s models.
If an image does not contain an embedded watermark, the tool cannot confidently determine whether it was AI-generated. Testing content created by other models highlights this limitation: Gemini sometimes “guesses” based on secondary visual cues, but such results cannot be considered reliable.
SynthID has been released as open source, and Google has even secured partners such as Hugging Face and Nvidia. However, most generators rely on different methods. For example, ChatGPT uses the C2PA metadata framework — supported by Microsoft, Adobe, Meta, and others. Google has stated that it plans to add C2PA compatibility to expand detection beyond its own ecosystem.
Yet even this would not guarantee full protection. Last summer, researchers at the University of Waterloo developed a technique called UnMarker, capable of stripping AI watermarks — including SynthID — in just minutes on an Nvidia A100 GPU. Google DeepMind reached similar conclusions, noting that C2PA metadata can be even less resilient in certain scenarios.
In parallel, the company introduced an updated version of its image-generation system, Nano Banana Pro. Built on Gemini 3 Pro, the model focuses on more accurate text rendering in images — historically a weak point for visual AI.
The algorithm can now produce infographics and other materials where legible labels are essential. Content generation speed has also noticeably improved. At the same time, images still contain the visible Gemini badge and the invisible SynthID watermark.
In one test, Nano Banana Pro generated an illustration specifically for demonstration, after which researchers attempted to remove the SynthID markers. Even after the watermark was stripped away, the system still identified the picture as AI-generated.
Thus, the new Gemini feature helps detect some images created with Google’s tools, but it is not universal. Removing or altering embedded markers remains possible, meaning that provenance-verification tools are only one of several methods needed to navigate an increasingly synthetic visual landscape.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
