New ways to create personalized images in the Gemini app
Key Points
- Personalization via Google Photos and Personal Intelligence
- Powered by Nano Banana 2; minimal prompting required
- Opt-in Google Photos use; not used to train models
Summary
Gemini now uses Personal Intelligence plus Nano Banana 2 and an optional Google Photos connection to generate personalized images without long prompts or manual uploads. The app can automatically select reference photos and apply your saved preferences and labeled people/pets to produce tailored outputs. Controls let you swap reference images, inspect sources, and provide corrections. This feature is rolling out in the U.S. over the next few days to Google AI Plus, Pro, and Ultra subscribers and remains opt-in and privacy-preserving.
Key Points
- Integration: Personal Intelligence + Nano Banana 2 uses context from your connected Google apps (including Google Photos) to ground image generations.
- Hands-off personalization: Simple prompts (e.g., “create a claymation image of me and my family”) can automatically include people/pets and style choices based on your library and preferences.
- User controls: You can refine results by telling Gemini what’s wrong, selecting a different reference photo via the ‘+’ icon, and viewing the auto-selected source via the Sources button.
- Privacy: Gemini does not train on your private Google Photos library; connecting Google Photos is opt-in and attribution metadata is available per image.
- Availability: U.S. rollout to Google AI Plus, Pro, and Ultra subscribers in the Gemini app first; broader desktop and user rollout planned.
Practical notes for engineers
- Model: image generation is powered by Nano Banana 2 integrated with the Personal Intelligence layer.
- Data flow: Gemini reads labeled/person context from a linked Google Photos library to select reference images but does not use private photos for model training; limited prompt/response telemetry may be used for functionality improvements.
- UX hooks: expect UI affordances for selecting reference images, a Sources view for attribution, and inline correction feedback APIs/controls in the app.
Bottom line
This update reduces prompt engineering overhead by surfacing personal context automatically while preserving opt-in privacy controls and offering interactive refinement mechanisms.