From the source
Google previewed Genie 4, a world model that generates realistic three-dimensional environments from natural language descriptions, as the latest in a series of world models the company launched in 2024.
Google DeepMind staffers say the most dramatic leap is how Genie 4 simulates people, creating characters that move naturally, respond organically to their environments, and can be interactively steered in real-time.
- Compared with its predecessor, the model is better at rendering fluid movements and other complex details, and it can generate accompanying audio such as the sound of crashing waves.
- Users describe what they want to render, then navigate the simulation and make edits through follow-up prompts, for example changing the weather or adding objects, without losing photorealism or consistency; keyboard controls adjust the camera's position and angle.
- Genie 4 relies on other Google models under the hood: Gemini Omni, a video generator introduced in May, keeps generated visual assets realistic, while Gemini 3.8 Flash, a cost-efficient reasoning model, makes individual simulation frames consistent with one another.
- DeepMind is studying robotics uses, pairing Genie 4 with Gemini 4 Argon, Google's flagship reasoning-optimized large language model, so engineers can simulate robots from high-level instructions; Argon turns a brief request into dozens of detailed Genie 4 prompts describing each step of a workflow.
- Google is testing Genie 4 with a limited number of early adopters and plans to make it available through its Google AI Ultra subscription for consumers, the Gemini API and Google AI Studio.