Google Gemini Is Now Rummaging Through Your Photos to Make AI Images

Google started pushing “personal intelligence” in Gemini earlier this year, giving subscribers a more tailored chatbot experience. Today, they’re taking it a step further: tying the image generation model directly to Google Photos.

If you opt in, Gemini can dig through your photo library and use the labels and context it finds there to improve generated images. The idea is you don’t have to describe every detail in your prompt. Just say “my family at the beach” or “my dog wearing a hat,” and the model does the heavy lifting by matching your photos to the request.

Honestly, this streamlines a workflow that already existed. Nano Banana 2 (yes, that’s the real name) is one of the better image generators out there, and you could already feed it reference images of yourself or others. But that required manual effort—uploading, selecting, describing. Now the bot just roams free in your library.

Google’s examples make it sound harmless: “You won’t have to pack as much context into your prompts.” True. If you want a picture of your family in a fantasy setting, the AI can pull visual cues from actual photos of them. That’s genuinely useful.

But let’s be real—this is also handing the keys to your personal visual history to a model that’s already training on everything you say. The privacy implications aren’t trivial. Google says you have to opt in, but we all know how those menus work. Most people will click “yes” without reading the fine print.

Still, for the use case, it delivers. The outputs I’ve seen are noticeably more accurate when the model has real reference material. It’s less generic AI slop and more something that actually resembles the people and places you care about.

Is it worth the trade-off? That depends on how much you trust Google with your dog photos. I’m not here to scare you off—I’ve used similar features myself—but I’d recommend at least reviewing what labels Gemini is pulling from your library before you let it run wild.

Either way, this is the direction we’re heading: AI that doesn’t just generate from scratch, but personalizes itself to your data. Whether that’s helpful or invasive is mostly a matter of where you draw the line.