Nano Banana now makes images from your Google Photos

Google shipped a quietly significant update to Gemini this week: personalized image generation. If you opt in, Gemini’s Nano Banana image model can now reach into your Google Photos library and generate images that actually look like you, your family, or things you care about — without you uploading reference images into the chat.

The canonical demo prompt from Google’s April 16 announcement:

“Create a claymation image of me and my family enjoying our favorite activity.”

Gemini pulls context from your photo library — who your family is, what your favorite activity tends to look like, what room you usually do it in — and generates a plausible claymation version. No uploads. No “here are three photos of my son for reference.” It just knows.

How the rollout is happening

• April 16: Announced via Google’s Personal Intelligence feature.
• April 18-20: Rolled out to paid US subscribers (Google AI Pro $19.99/mo and above).
• Opt-in required: You have to explicitly connect your Google Photos library under Personal Intelligence settings. Default is off.
• Free tier: Not included, at least not yet.

The opt-in gate is the most important design decision here. Google has been consistent since Gemini’s rebrand about not hoovering photo data without consent; that position is what makes this feature legally viable in the EU and has kept Apple out of a comparable feature for 18 months.

What this actually changes for you

For casual creators: It’s genuinely useful. The old workflow for “make an image of my kid as a pirate” was: open Midjourney, upload 3-5 reference photos with --cref, adjust prompts, wait for queue. Now it’s: ask Gemini in plain English. The reference images are already there; you don’t think about them.

For photographers and designers: Less relevant. You probably don’t want Nano Banana riffing on your work portfolio. You want precise control, which specialist tools (Midjourney, FLUX, Stable Diffusion + ControlNet) still do better.

For parents and families: This is the feature that’ll drive consumer AI image generation past Midjourney for casual use. The magic isn’t the model — it’s that you don’t have to do the reference work. Grandma can ask for a cartoon of her grandkids without learning what a --cref parameter is.

The privacy footnote that actually matters

Google says the Gemini app does not directly train its models on users’ private Google Photos library, but does use “limited info, like specific prompts in Gemini and the model’s responses.”

Translate: your photos aren’t training data. Your prompts and Gemini’s responses — which can contain descriptions of your photos — are. If you don’t want Google storing the text description “a claymation of my 4-year-old son playing with a yellow truck in our garage,” don’t use the feature.

I wouldn’t overstate the risk. It’s a small marginal increment on the data Google already has from Gmail, Calendar, and general Gemini use. But the feature is clearly useful enough that a lot of people will opt in without thinking through what goes into the prompt-response log.

Where this fits in the image generation landscape

Nano Banana was already the speed leader in mainstream AI image generation — 1-3 seconds per image, versus 10-30 for DALL-E / GPT Image and 30-60 for Midjourney. Personalized generation makes it the ease-of-use leader too.

What Nano Banana still doesn’t have: artistic quality. If you want an image that looks like a professional illustrator made it — book covers, concept art, editorial illustration — Midjourney is still ahead. And for absolute benchmark quality on complex multi-element prompts, GPT Image 1.5 still holds the #1 Arena ELO.

My current stack for image work in April 2026:

• Nano Banana (free via Gemini) for quick casual images, now especially with personalization
• Midjourney ($30/mo Standard) for anything that needs to look designed
• ChatGPT Plus ($20/mo) for DALL-E/GPT Image when you’re already drafting text there

Personalized Nano Banana tips the “which free AI image generator should a casual user pick” answer decisively toward Google. Six months ago that answer was ChatGPT Free (because of DALL-E integration). This week it’s Gemini free tier — though personalization itself is paid-only.

See the full best AI image generation tools page for the ranked comparison. Full Nano Banana Pro review has my long-form take.

Questions or corrections? Email Pick Right. Want the full list? See all news.