Gemini’s Nano Banana 2 Can Now Make Images From Your Actual Life

Google quietly pushed an update to the Gemini app today that makes its image generation feature significantly less useless for everyday people. Nano Banana 2—yes, that’s the actual name—now pulls from your personal context and Google Photos to create images that reflect your actual life.

This is one of those features that sounds obvious in hindsight but took way too long to arrive. We’ve had AI image generators for years now, but they’ve always been weirdly disconnected from the person using them. You’d type “a cat sitting on my sofa” and get some generic feline on a generic couch that looked nothing like your actual cat or your actual sofa. Nano Banana 2 aims to fix that.

How it works is straightforward: the model has access to your Google Photos library and whatever personal context Google has on you (which is a lot, let’s be honest). When you ask it to generate an image, it can reference your actual pets, your actual furniture, your actual travel photos, and even your actual face. The result is supposed to feel like a photo you could have taken, not a stock image from the uncanny valley.

I’ve been testing this for a few days and it’s genuinely impressive in some areas. I asked it to generate an image of my dog wearing a birthday hat in my living room, and it got both the dog’s coloring and the specific layout of my apartment right. The lighting was off, and the hat looked a bit like a CGI accident, but the recognition of my actual space was better than I expected.

Privacy-wise, you should probably think about this one before diving in. Google says the image generation happens on-device as much as possible, and your photos aren’t used to train the model further. But you’re still handing over access to your personal photo library to an AI system. If that makes you uncomfortable, you can opt out entirely in the settings.

The feature is rolling out now to Gemini Advanced subscribers in English-speaking countries. Free tier users get the old generic version. Typical Google move—dangle the actually useful feature behind a paywall.

What I’d really like to see next is this same personalization applied to other modalities. Imagine an AI assistant that can generate outfit suggestions from your actual wardrobe, or meal ideas from the actual contents of your fridge. Nano Banana 2 feels like a proof of concept for that future.

Is it perfect? No. Faces still come out weird sometimes, and the model struggles with complex scenes involving multiple people. But it’s a solid step toward AI that actually understands your context instead of just generating random pixels that sort of match your prompt.

Go try it if you have Gemini Advanced. Just maybe don’t ask it to generate anything you wouldn’t show your mom.