Google’s Gemini Now Rummages Through Your Photos to Make Custom Images

Google’s been pushing this “personal intelligence” thing in Gemini since early this year, letting subscribers get a more customized chatbot experience. Today, they’re taking it a step further by hooking it up to Google Photos.

If you opt in, Gemini’s image generator—Nano Banana 2, which I still think is a weird name—gets access to your photo library and the labels attached to those pictures. The idea is you can type something simple like “my dog playing fetch” and it’ll actually know what your dog looks like, instead of generating some generic golden retriever.

In practice, this streamlines something that was already possible. You could previously feed Gemini images of yourself or your family to use as reference for new generations. But that meant manually uploading or selecting photos each time. Now the model just digs through your library on its own.

Google’s examples make sense: instead of describing your family members’ appearances in excruciating detail, you can just say “my family at the beach” and let the model figure out who’s who. Same for pets, cars, or that weird lamp you bought at a garage sale.

Is this useful? For some people, absolutely. If you’re the type who generates AI images regularly—for social media, gifts, or just for fun—having the model already know your context cuts down prompt engineering significantly.

But let’s be real: this requires trusting Google with even more personal data. Your photos are already on their servers, sure, but now an AI model is actively crawling through them to learn what you look like, who your family members are, and what your dog’s name is. Google says this is opt-in, which is the bare minimum, but I’d be surprised if they don’t eventually push it harder.

The quality of the output depends heavily on how well your photos are labeled. If you’ve got a mess of untagged images from five different dogs, good luck getting the right one. Google Photos’ automatic tagging is decent, but it’s not perfect.

For now, this is rolling out to Gemini Advanced subscribers. If you’re already paying for the service and you’re comfortable with the privacy trade-off, it’s a nice convenience. If you’re not, well, the old manual method still works fine.