DeepMind Unveils Genie 4 With Sharper Physics and More Lifelike People

Google DeepMind has announced Genie 4, the latest version of its world model, a system that generates interactive environments that users can move through. The company says this release targets two areas where earlier versions fell short: how objects behave under physical forces, and how people look and move inside a generated scene.

Realism is the main limit on what world models can be used for. Simulated spaces that bend the rules of motion, or that render people awkwardly, are hard to rely on for training robots or for testing game and product ideas. More believable physics and human figures would make the output more useful for those purposes, though the initial details do not say how those gains were measured or how large they are.

Access will be staged. DeepMind says Genie 4 will roll out first to developers and to subscribers of Google AI Ultra, the company's top consumer AI plan. The announcement gave no firm date for that rollout, so wider availability will depend on how the first phase goes.

Until outside testers have tried the model, the improvements should be read as the company's own claims. Coverage will be worth revisiting once the rollout begins and examples or independent comparisons are published.