Google’s Gemini Now Digs Through Your Photos to Make AI Images. I Have Thoughts.

Google started pushing its “personal intelligence” thing in Gemini earlier this year. The idea was simple: let the AI know more about you, and it’ll give you better answers. Today, they’re taking that concept and throwing Google Photos into the mix.

If you opt in, Gemini’s image generation model—Nano Banana 2, which is genuinely one of the better AI image generators out there—can now rummage through your photo library. It reads the labels and content of your photos to figure out what you mean when you say “my dog” or “my family.” The result is that you don’t have to write a novel of a prompt to get a recognizable picture of your actual golden retriever.

Honestly, this was already possible before, just clunkier. You could manually upload images of yourself or your pet and ask the AI to work with that context. This new integration just automates the boring part. It’s like the difference between handing someone a photo album and saying “this is my dog” versus letting them just go through your entire camera roll. Convenient, but also a little invasive if you think about it too hard.

Google’s examples make sense. Instead of typing “a golden retriever sitting on a beach at sunset wearing sunglasses, the dog has floppy ears and a red collar,” you can just say “my dog on a beach at sunset.” The AI pulls the visual details from your photos. It works because more personal data usually means better AI output. That’s a truism at this point.

But here’s the thing: this is Google we’re talking about. They already have access to your photos if you use Google Photos. They already train models on your data if you’ve agreed to it. This feature doesn’t change that. What it does change is how much of that data gets fed directly into a generative model that creates new images of you, your family, and your pets. That’s a different category of risk than just having Google scan your photos for search.

I’m not saying don’t use it. I’m saying think about it. If you’re the kind of person who already uploads face scans to AI tools, this is a quality-of-life improvement. If you’ve been keeping your photo library private, nothing changes—you have to opt in. But the rollout is seamless enough that a lot of people will just click “yes” without reading the fine print.

The bigger picture is that this is where all the major players are heading. Apple, Meta, Google—they all want their AI to know you personally. The trade-off is always the same: convenience for privacy. This feature is a good example of that trade-off in action. It’s useful, it works well, and it makes me uneasy in exactly the way most Google products do these days.