Google DeepMind has released a research preview of Genie 4, its latest world model.
World models are a core building block of physical AI. They are designed to help AI systems understand the real world, predict the outcomes of actions and make decisions on their own.
Genie 4 generates realistic, interactive virtual worlds, and users can enter prompts during a simulation to change the scene. The model also produces real-time audio matched to the simulated world, including music, ambient sound and dialogue characters can interact with.
The release comes just months after AI video startup Runway launched GWM Worlds 2, its second world model, which can generate dynamic responses and audio. It signals Google DeepMind's intent to lead the world model market as the sector reshapes. World Labs, founded by Fei-Fei Li, has been acquired by AMD, and Amazon is reportedly in talks to buy Decart. Rivian has acquired Third Dimension to gain autonomous-driving test simulation capabilities.
Compared with its predecessor, Genie 4 offers more realistic graphics and audio. It also supports flexible editing, such as changing weather or camera angles in real time and adding new people or objects. It can maintain a consistent scene for up to two minutes, double the previous limit, and features an improved physics engine and more lifelike characters.
Genie 4 can also be tightly integrated with Google's Gemini AI models. In this setup, Gemini acts as a director, drawing up step-by-step plans and reviewing frames while Genie generates the visuals. In a Gemini Robotics research experiment, the Gemini 4 Argon model broke tasks into steps while Genie 4 simulated them in real time.
Potential commercial uses for Genie 4 include gaming, interactive learning and robotics. Some observers also expect it could be used to prototype early-stage film productions. However, the model is still under development and has limitations, so its performance is expected to keep improving.
“The real-time nature of the model is most valuable when it's important to stay in a 'flow state' during the creative process,” said Diego Rivas, an engineer at Google DeepMind. “It can be useful when you need to generate ideas quickly, such as when brainstorming or storyboarding.”
The research preview is initially available to “trusted testers.” Within weeks, it will roll out to Google AI Ultra subscribers through Project Genie, Google's AI tool that generates interactive worlds from text or image prompts. Access will then expand to early-access partners for the Gemini API and Google AI Studio.
