Google’s Dreambeans turns account data into daily illustrated stories

Google is promoting Dreambeans, an experimental Google Labs app that analyzes selected Google services overnight and returns a personalized set of illustrated story cards the next morning. Each batch stops at roughly a dozen cards, giving the experience a defined endpoint.

Dreambeans is available on Android and iOS in the United States for users 18 and older. The app tests how one system can combine email, appointments, photos, viewing habits, searches, and Gemini activity to produce suggestions before a user enters a prompt. Google currently presents it as a consumer Labs experiment and has announced no API or SDK.

The pipeline runs overnight

With permission, Dreambeans uses the same Personal Intelligence technology that supports personalization in the Gemini app and AI Mode in Google Search. Personal Intelligence is Google’s system for finding relevant connections across a user’s authorized account data.

1. Source selection: Users can connect any combination of Gmail, Calendar, Google Photos, YouTube, Search history, and Gemini.
2. Nightly processing: Dreambeans reviews the selected information and curates a fresh set of cards overnight.
3. Image generation: Google’s Nano Banana 2 image model creates an illustration for each card.
4. Delivery: The app sends one notification when the next batch is ready.

Google’s product lead described a card generated from two separate signals: a Gmail confirmation that puppy treats had arrived and a Calendar reminder that a friend was visiting. Dreambeans used them to suggest puppy-training tips and dog-friendly restaurants.

Cards open into a fullscreen, stories-style view with one-tap actions for watching a trailer, viewing ticket details, or opening a suggested purchase. Suggestion chips, a tuning chat behind the sparkle icon, and feedback controls let users confirm a relevant card, correct a detail, or mark a topic as “not for me.” Corrections take effect during the next overnight run and do not alter the current batch.

Broad access, separate controls

Dreambeans can combine information from email, calendars, photos, searches, and media histories, allowing it to infer details that no single service contains. Connections are optional and managed source by source.

• Service access: Individual sources can be connected or disconnected from the profile menu.
• Recognizable faces: When Google Photos face grouping is enabled, generated artwork can depict recognizable versions of the user and people in their library.
• Feedback records: Users can review their previous Dreambeans feedback from the profile menu.
• Data deletion: The app includes a control for deleting all Dreambeans data.

Dreambeans settings remain independent from equivalent controls in Gemini and AI Mode, so changing a connection in the app does not alter personalization elsewhere. Google also says data supplied to Dreambeans is not used to train its models.

A daily interface that initiates

Google Labs is also testing CC, an agent that uses Gmail, Calendar, and Drive to compile a daily “Your Day Ahead” email. CC delivers a text digest through the inbox, while Dreambeans packages cross-service recommendations as illustrated, tappable cards.

A finite daily batch lets Google test whether proactive AI can remain relevant, correctable, and manageable over time. For developers and researchers, the notable product choices are bounded output, source-level permissions, delayed feedback processing, and generated imagery tied to personal context.

Try it with a narrow data scope

A controlled trial can begin with one or two services, followed by additional connections only when the cards need more context.

1. Connect the minimum set of sources needed for the experiment.
2. Review which details Dreambeans combines and whether its inferences are accurate.
3. Correct one card, then inspect the following day’s batch for changes.
4. Test the disconnect, feedback-history, and deletion controls before expanding access.

Because feedback is processed overnight, evaluating Dreambeans requires several daily batches. The clearest measures are factual accuracy, sustained relevance, the sensitivity of inferred details, and whether the account controls behave as described.