Google has extended its Personal Intelligence system for Gemini, enabling the AI to now generate images informed by a user's Google Photos library. This move connects the new Nano Banana 2 image model to a pipeline of personal context data. The technical architecture allows a user to prompt for concepts like "my dream house," and the system will attempt to reflect individual aesthetic preferences identified from connected app data.

From an engineering perspective, the system uses existing labels within Google Photos—such as identified people—as contextual signals for the generative model. A Google spokesperson confirmed this data informs the image creation process. A notable design decision is the stated data handling policy: while the system uses your photos for real-time inference, Google asserts it will not directly train its core AI models on the private photo libraries themselves. However, the company does acknowledge training on associated interaction data, including specific prompts and the model's outputs.

The feature represents a practical trade-off between personalized output and user privacy, managed through explicit opt-in. It is currently rolling out to paid Gemini subscribers in the United States, with a planned expansion to desktop Chrome and broader user bases. The implementation highlights a trend towards tightly coupling generative models with existing, structured personal data ecosystems.

Source: The Verge