Websites streamed live directly from the model

Flipbook is an experimental service built around a visual browser generated on-demand in real time.

• Navigation begins when a user types into the search bar or uploads an image.
• Each "page" is generated as an image, and clicking on an object within the image regenerates a deeper exploration screen.
• Text on the screen is also rendered not as a separate overlay but as pixels drawn by the image model.
• Information is generated by combining agentic web search with the model's world knowledge, and the service notes that some inaccuracies may occur.

The live video stream feature is an experimental feature that turns static images into something closer to continuous video. It smoothly transitions between and animates the images explored, but it is currently offered only as a toggle due to low predictability and high resource consumption.

The current architecture combines a custom, highly efficient video generation model with an image generation system. Going forward, there are plans to unify the two systems into one.

The team views Flipbook as still experimental, but expects that as the accuracy and performance of image and video models improve, it will become possible to incorporate more real-world data and interactivity, execution capabilities, and even data storage.