If you’ve spent any time online this year, you’ve probably had that moment—staring at a photo of a politician in a questionable outfit or a celebrity doing something improbable, wondering, “Is this real?”

Google is finally giving us a tool to answer that question, or at least, it’s trying to. Starting this week, if you’re using the Gemini app, you can simply upload an image and ask, “Is this AI-generated?” It’s a straightforward feature that feels long overdue, but as with all things in the world of synthetic media, the reality is a bit more complicated than a simple “yes” or “no.”

To understand why this update matters, you have to look at what Google just released alongside it. We are officially entering the era of Nano Banana Pro.

Yes, you read that right.

In what might be the most “internet” branding accident of 2025, Google’s latest state-of-the-art image model, officially named Gemini 3 Pro Image, has been embraced by the community (and now Google itself) under its leaked internal codename: Nano Banana.

What started as a viral meme when the model was being anonymously tested has become the face of Google’s most powerful creative tool. But don’t let the silly name fool you. Nano Banana Pro is a beast. It offers 4K resolution, “studio-quality” lighting controls, and arguably the best text rendering we’ve seen yet (goodbye, gibberish signs in AI backgrounds).

But with great power comes great responsibility—or at least, a great need for guardrails. Because Nano Banana Pro makes it easier than ever to create hyper-realistic fake photos, Google is rolling it out with two layers of digital armor: SynthID and C2PA.

The invisible fingerprint: SynthID

The new “Is this AI?” feature in the Gemini app relies primarily on SynthID, Google’s proprietary watermarking technology.

Think of SynthID as an invisible digital fingerprint. When an image is generated by Google’s tools, SynthID embeds a signal directly into the pixels. You can’t see it, but Gemini can read it. It’s designed to be robust; even if you crop the photo, throw a filter on it, or compress it into a crusty JPEG, the signal usually survives.

According to Google, SynthID has already been applied to over 20 billion images. So, if you ask Gemini about a suspicious photo and it spots that watermark, it can give you a definitive “Yes, this was made with Google AI.”

However, there’s a catch: SynthID currently only works for content made by Google. If someone generates a deepfake using Midjourney, Flux, or OpenAI’s tools, Gemini’s SynthID scanner won’t catch it. That’s where the second piece of the puzzle comes in.

The industry standard: C2PA

While SynthID is Google’s home-cooked solution, C2PA (Coalition for Content Provenance and Authenticity) is the potluck dinner where everyone brings a dish.

C2PA is an open technical standard that essentially attaches a “nutrition label” to digital files. This metadata tracks where an image came from and every edit made to it along the way. If an image is generated by AI, the C2PA credentials will say so. If it’s a real photo taken by a camera, the credentials can verify that, too.

Google is now embedding C2PA metadata into all images generated by Nano Banana Pro. This is a big deal because it means these images can be verified not just by Gemini, but by any platform that supports the standard.

And the industry is finally rallying.

• TikTok recently announced it is rolling out “invisible watermarking” and will automatically label content that carries C2PA metadata.
• Samsung has made waves by integrating C2PA support natively into the Galaxy S25, meaning photos taken on the device have their authenticity baked in from the moment of capture.
• Meta (Facebook/Instagram) has also joined the C2PA steering committee, committing to labeling AI content across its massive social networks.

Is the problem of AI misinformation solved? Definitely not.

The biggest hurdle is that these tools are “opt-in” for the good actors. Responsible companies like Google, Adobe, and OpenAI are playing ball, but bad actors using open-source models on private servers aren’t going to watermark their deepfakes.

Furthermore, while SynthID is hard to break, C2PA metadata can be stripped. Taking a screenshot of a C2PA-protected image often wipes the “nutrition label” clean. It’s a classic cat-and-mouse game: as detection gets better, evasion gets smarter.

Google’s new verification tool in Gemini is a welcome step. It empowers regular users to do a quick “vibe check” on images that feel off. But we aren’t at the point where we can outsource our critical thinking to an app just yet.

For now, if you see a photo of a politician eating a banana that looks too perfect, maybe check with Gemini. Just don’t be surprised if the answer involves a “Nano Banana.”