On October 27, Google DeepMind unveiled its new world model Genie 4. The model generates interactive virtual environments that users can move around in, in real time, and does so with sound for the first time.
What happened?
According to DeepMind, Genie 4 is the company's first real-time world model that generates sound directly. Each world carries its own ambient sound: wind in the trees, waves hitting the shore or music. According to Ars Technica, the biggest novelty is human speech: dialogue that adapts as the simulation changes. People in the scene also move more naturally than in the previous version.
Users can steer a running world with plain text. It is possible to change the weather, move the camera or add new people and objects to the scene; Gemini models interpret the commands. DeepMind positions the model as a training environment for robots and AI agents, a prototyping tool in game design and a previsualization tool in film. Examples that turn a single archive photo into an explorable scene were also shared.
For now the model is open only to selected testers. Access will soon expand step by step to Google AI Ultra subscribers through Project Genie and to early-access partners of the Gemini API and AI Studio.
Why does it matter?
World models mean AI generating not just text or images but environments that run by their own rules. This could let robots be tried out in thousands of scenarios before they go into the real world, and could speed up game and film production. The limits are clear, though: according to Ars Technica, the worlds stay consistent for only a few minutes. DeepMind also acknowledges that objects can drift over time and that complex physical interactions and long-term consistency are still research topics.