An "interactive world" with sound
Google on Oct. 27 (Oct. 28 Japan time) unveiled "Genie 4," an AI that instantly creates interactive 3D worlds. It creates worlds from text or images, and users can move around inside them with a keyboard. CEO Sundar Pichai introduced it on X.
Such AI is called a "world model," and it mimics how the real world behaves to predict what comes next. The previous generation, "Genie 3," released in August 2025, produced only visuals, but Genie 4 also generates ambient sound, music and dialogue at the same time. Typing text partway through can change the contents of the world.
Gemini plans the steps
Inside Genie 4, Google's lightweight model "Gemini 3.8 Flash" serves as a guide. It breaks goals into steps, checks whether each step succeeded and replans if the world drifts. This allows it to advance events made up of multiple stages in a logical sequence, according to Google.
Google cited film previsualization storyboards, game prototyping and training environments for robots as uses. It also showed an example in which the higher-end model "Gemini 4 Argon" plans the steps for a robot to play a board game, which are then tested in Genie 4.
Limited testers first
A research-stage version will go first to testers selected by Google. Early access through "Project Genie" in the paid monthly "Google AI Ultra" plan, and through the Gemini API and Google AI Studio for developers, is coming "soon," the company said. It has not disclosed pricing, image quality or performance figures.
Google also acknowledged weaknesses. When it runs for a long time or instructions are changed repeatedly, the details and positions of objects can drift. It cannot yet handle multiple people or agents acting in the same world or accurately reproduce real places. The visuals and sound it creates carry "SynthID," an invisible mark indicating they were made with AI.
