Gemini can now tell you if an image was made by Google AI

How many times have you stopped mid-scroll lately, squinting at a bizarre photo and wondering if it was cooked up by an AI?

As image generators get eerily realistic, counting fingers and hunting for blurred background text just doesn't cut it anymore. Google wants to help you spot the fakes—starting with its own.

Google just brought AI image verification straight into the Gemini app. You can now upload any picture and simply ask whether it was created or edited with Google AI.

Just ask Gemini to check the watermark

The feature is powered by SynthID, Google's proprietary watermarking tech that embeds imperceptible signals directly into AI-generated pixels. Google first introduced the tech in 2023, and over 20 billion pieces of AI content carry it today.

Up until now, verifying those marks was mostly restricted to a dedicated SynthID Detector portal tested with journalists and media pros. Putting it directly inside Gemini brings that capability to anyone with the app.

All you have to do is drop an image into the chat and ask a natural question like "Was this created with Google AI?" Gemini scans for the hidden watermark and uses its reasoning to explain the context.

Google isn't stopping at its own garden wall either. Starting this week, images created by Nano Banana Pro (Gemini 3 Pro Image) across Gemini, Vertex AI, and Google Ads will carry C2PA metadata. In the coming months, Google plans to expand verification to video and audio, bring it to Search, and support C2PA credentials from third-party tools outside Google's ecosystem.

Why it matters

Google is doing something clever here: taking digital provenance out of dusty developer portals and dropping it right into a casual chat interface.

Most everyday users will never visit a specialized forensics website just to fact-check an odd photo they saw on social media. Making the check as simple as asking a chatbot lowers the barrier to zero. If you frequently find yourself wondering if a trending visual is real, it's worth testing Gemini on it right now.

That said, there is a giant elephant in the room. Right now, Gemini can only reliably spot images generated within Google's own walls using SynthID. When it comes to Midjourney, Flux, or un-watermarked open-source models, the app still can't provide absolute proof.

Until universal standards like C2PA become mandatory across the entire industry, watermarking only solves part of the problem. But turning verification into an everyday consumer feature is a solid first step toward keeping online media honest.