Google unveils state-of-the-art Genie 4 world model

Google LLC today previewed Genie 4, an artificial intelligence model that can create realistic three-dimensional environments.

The algorithm is the latest addition to a series of world models that the company launched in 2024. Compared to its predecessor, Genie 4 is better able to display fluid movements and other complex details. In addition, it can produce accompanying audio signals such as the sound of crashing waves.

“The most dramatic leap is how Genie 4 simulates people and creates characters that move naturally, react organically to their surroundings, and can be controlled interactively in real time,” wrote Google DeepMind employees Raia Hadsell, Diego Rivas, and Hang Qi in one Blog post.

Users can create a virtual world by providing Genie 4 with a natural language description of what they want to render. They can then navigate through the simulation and make changes by entering follow-up prompts. For example, a user could ask Genie 4 to change the weather or add more objects. The model can make such changes in a way that does not affect the photorealism or consistency of the virtual world.

Genie 4’s natural language interface is available alongside keyboard controls. They can be used to change the position and angle of the camera.

Under the hood, Genie 4 leverages several other Google models to improve output quality. It’s based on Gemini Omni, a video generator the search giant launched in May to ensure the visual assets generated are realistic. Gemini 3.8 Flash, an argumentation model optimized for cost efficiency, ensures that the individual frames of a simulation are consistent with one another.

“Gemini continuously observes the scene, breaking targets into a sequence of steps and inspecting incoming frames to track progress and verify that each step was successful,” Hadsell, Rivas and Qi wrote. “As the world changes or the user intervenes, Gemini adapts its plan at that moment.”

DeepMind is currently exploring how Genie 4 can be applied to use cases in robotics. The project combines the algorithm with Gemini 4 Argon, Google’s flagship model for argument-optimized large languages.

The goal of the project is to enable engineers to simulate robots using high-level instructions. For example, a user might ask Argon to create synthetic footage of a robotic arm soldering two car parts together. Argon can convert the short query into dozens of highly detailed, multi-sentence Genie 4 prompts, each describing a different step of the soldering workflow.

Increasing the number and precision of prompts provided for Genie 4 improves rendering quality. This increase in quality, in turn, makes the model’s robot simulations more useful for AI training purposes. Developers train AI models to control robots by providing them with visual examples of how to complete the tasks expected of them.

Google is currently testing Genie 4 with a limited number of early adopters. It plans to make the model available through its Google AI Ultra consumer subscription, the Gemini API, and Google AI Studio.

Image: Google