Google Photos now powers Gemini’s personalized image creation
▼ Summary
– Google’s Personal Intelligence feature now uses personal data from connected apps and its Nano Banana 2 model to generate context-aware images for users.
– The feature creates images, like a dream house or desert island essentials, that reflect a user’s tastes by analyzing data from their Google Photos labels.
– Google states it will not directly train its AI models on users’ private Google Photos libraries if they opt into this feature.
– The company does, however, train models on limited information from user interactions, such as specific prompts and the AI’s responses.
– The feature is rolling out first to eligible US subscribers of AI Plus, Pro, and Ultra plans, with expansion to desktop and more users planned.
A new integration is bringing a highly personalized dimension to AI-generated imagery. Google has announced that its Personal Intelligence feature within Gemini can now create custom images by drawing on a user’s private data from connected apps like Google Photos. This capability is powered by the company’s Nano Banana 2 image model, which uses information from a user’s digital life to tailor visual outputs to their specific context.
According to Google, subscribers can now input prompts such as “Design my dream house” or “Create a picture of my desert island essentials.” The resulting images will automatically incorporate elements reflecting an individual’s tastes and lifestyle, inferred from their connected Google services. A company spokesperson explained that the system works by utilizing labels within a user’s Google Photos library to identify key people, including the user themselves, their friends, and family members. The Nano Banana 2 model then generates the image based on this contextual data.
Google has addressed potential privacy concerns, stating that for users who opt into the Personal Intelligence feature, the company will not directly train its broader AI models on the contents of their private Google Photos libraries. The training process instead involves what Google describes as “limited info,” such as the specific prompts entered into Gemini and the corresponding model responses.
This new functionality is beginning a phased rollout. Over the next several days, it will become available to eligible AI Plus, Pro, and Ultra subscribers in the United States. Google plans to expand access to the feature on Gemini for Chrome desktops and to more users in the near future.
(Source: The Verge)
