Article —
Google Genie 4 generates interactive worlds in real time
Genie 4 is a Google DeepMind world model that builds explorable simulations you can change with text commands. It could shorten the path from 3D design to finished visuals.
Genie 4 is a “world model” from Google DeepMind. It generates interactive environments in real time, and you can move through them and change them while they run.
What it is
Genie 4 is made by Google DeepMind and builds on the Gemini model family. According to the report on Mobiili.fi, Gemini Omni grounds the simulations in how the real world looks and behaves. Other Gemini models handle reasoning and keep the scene coherent. A Gemini 3.8 Flash control agent tracks goals and progress in complex events.
You navigate with conventional controls and use text or voice commands to change the weather, move the camera, or add objects and people. The model simulates physics such as collisions and aerodynamics, and it generates human figures that move naturally. It is described as DeepMind’s first real-time world model with native audio: sounds, music and dialogue.
Access starts as a research preview for selected testers. Google AI Ultra subscribers are to get it through Project Genie, and developers through the Gemini API and Google AI Studio. All output carries SynthID watermarks.
What it is not
- It is not a finished product. Google calls it an early-stage research model, and says long-term consistency and visual detail still need work.
- It is not a 3D modelling tool. The report does not describe any link to CAD or design software.
- It is not widely available yet. Resolution, simulation length and pricing beyond the Ultra tier are not stated.
What it is good for
DeepMind names robot and agent training, film previsualization, games and industrial simulation. For anyone who works in 3D design, the interesting part is rendering. Today you draw the structures, then move images into a separate rendering program. If world models reach the quality described, high-quality images and video could come straight from the design tool.
That is a big “if”, given the stated limits. The pace of development is still hard to ignore. Treat this as a direction to watch, and test it once API access opens.
Contains AI-generated content