Google introduces cutting-edge Genie 4 world model

Google has unveiled Genie 4, an advanced artificial intelligence model capable of producing lifelike three-dimensional spaces. This is part of a series of world models the company has introduced since 2024. Genie 4 significantly improves upon its predecessor, excelling at rendering smooth movements and intricate details. Notably, it can also produce accompanying sound effects, such as the roar of crashing waves.

According to DeepMind researchers Raia Hadsell, Diego Rivas, and Hang Qi, the most significant advancement is Genie 4's ability to simulate human-like characters. These characters move naturally, interact organically with their surroundings, and can be controlled in real-time. Users input a natural language description of the desired virtual world, after which they can navigate the simulation and make changes using subsequent prompts.

For instance, a user could instruct Genie 4 to alter the weather or introduce additional elements. The model accommodates these modifications without compromising the virtual world's realism or coherence. Access to Genie 4's natural language interface is available alongside keyboard controls, enabling users to adjust the camera's position and angle.

Under the hood, Genie 4 leverages several other Google models to enhance its output quality. Gemini Omni, a video generator introduced in May, ensures the generated visuals are realistic. Gemini 3.8 Flash, a cost-efficient reasoning model, maintains the consistency of individual frames within a simulation. Gemini continuously monitors the scene, decomposes goals into sequential steps, and reviews incoming frames to verify progress and ensure each step is successful.

DeepMind is currently exploring potential applications of Genie 4 in robotics. By combining the algorithm with Gemini 4 Argon, Google's top-tier reasoning-optimized large language model, the objective is to enable engineers to simulate robots using high-level instructions. For example, a user could instruct Argon to create synthetic footage of a robotic arm soldering two car parts together.

The more and precise prompts provided to Genie 4, the better the resulting rendering quality. This increased quality, in turn, makes the model's robot simulations more valuable for AI training purposes. Currently, Google is testing Genie 4 with a select group of early adopters and plans to make it available through its Google AI Ultra subscription, the Gemini API, and Google AI Studio.