Google has begun rolling out a significant update to Gemini, integrating the assistant with Google Photos. This new feature is a core component of the “Personal Intelligence” concept, allowing the AI to generate custom images based on a user’s personal photo library.

Streamlined Creative Workflow

Currently, the update is being released primarily for subscribers of Google’s paid AI plans (AI Plus, Pro, and Ultra). The integration aims to make image generation more intuitive by reducing the need for exhaustive text descriptions. By accessing the Google Photos library—including existing labels for people and pets—Gemini can identify relevant subjects automatically.

For example, instead of describing a specific person or dog in detail, users can use simple prompts like “my family” or “my dog.” The system then uses those real-world references to ground the generated art. This process leverages the Nano Banana 2 model (Gemini 3.1 Flash Image), which is designed for high-speed, conversational image creation and editing. While users could previously upload reference images manually, this direct connection eliminates several steps, making the creative process more fluid.

Contextual Coherence and Precision

The primary benefit of this update is the reduction of “friction.” By interpreting personal context directly from a user’s gallery, Gemini can produce more coherent scenes—such as recreating frequent activities or simulating specific visual styles based on past photos.

However, the feature is still evolving. Google acknowledges that the system may occasionally select the wrong reference image. To address this, the interface includes tools that allow users to:

• Review the sources Gemini selected to guide the creation.

• Manually adjust the results for better accuracy.

Privacy and Control

Privacy remains a central focus of the rollout. Google emphasizes that personal photos are used only as context for specific generation requests and are not used to train the underlying AI models. Furthermore, the Google Photos integration is disabled by default; users must explicitly grant permission within the “Personal Intelligence” settings to enable the feature.