Google's Genie 4 lets you build worlds that sound and move like real life

It’s easy to think generative AI is all about video and image generation tools like Gemini Omni or Nano Banana. However, world models are where a lot of AI labs are focusing their research. World models build an internal simulation of the environment and use physics, spatial relationships, and time progression to create consistent output. While large language models (LLMs) use token data to predict the next word, world models predict how space, time, gravity, and other physical forces will evolve in an environment.

Google has announced Genie 4, its latest world model, which improves on last year’s Genie 3 in several key areas. It builds on Gemini Omni, which Google says anchors Genie 4’s simulations in real-world appearance and behavior. Genie 4 can also keep a scene consistent for up to two minutes, double Genie 3’s limit. With Genie 4, you can shape the world in real time with natural language prompts. Google says you can change the weather, move the camera, and even introduce new people or objects into the scene, while Genie 4 keeps everything consistent.

It’s also Google’s first real-time world model to support native audio generation. The company says Genie 4 can automatically generate ambient sounds for the world it’s simulating, so you don’t need to add any additional detail about background noise. Google also says that the model can generate human dialogue that adapts as the scene changes. You can make characters talk to each other or react to things happening around them in real time.

Genie 4 also works in tandem with Gemini 3.8 Flash to execute prompts with complex events. Gemini acts as a guiding agent, breaking down the prompt’s primary goal into smaller steps and continuously observing the scene as Genie 4 generates it to keep those steps on track. Gemini also adjusts those steps in real time if the user intervenes and asks for an unplanned change to the scene.

The model also supports more realistic human movement and world physics. Google says Genie 4 can successfully simulate scenarios such as aerodynamics, collisions, and more. It can also generate people who move naturally and respond organically to their environment. That said, Google calls Genie 4 an early, experimental model, and longer sessions can still produce visual inconsistencies. Everything it generates is also marked with SynthID, Google’s invisible watermark for AI-generated content.

Before you head over to Project Genie in the hope of creating your own Genie 4-powered world, it’s worth noting that access is currently limited. Genie 4 is currently available as a research preview for Google’s trusted testers. In the coming weeks, it will roll out to users with a Google AI Ultra subscription in Project Genie, as well as early access partners of the Gemini API and Google AI Studio. You can sign up for the Genie 4 waitlist in Project Genie here.