Google DeepMind Researchers Explain How World Models Create Navigable Environments

Google DeepMind

Feb 25, 2026 · Updated Apr 25, 2026

Google DeepMind explains world models through Project Genie: they simulate environments moment-by-moment as an agent acts, not just predicting text. A single image generates a navigable world — objects respond, rooms are walkable — without any game engine.

Project Genie, Google DeepMind's experimental world model, turns image and text prompts into interactive, navigable environments. Co-leads Shlomi Fruchter and Jack Parker-Holder explain the key distinction: language models predict the next word; world models predict the next visual state based on what an agent does. Push a ball and it rolls. Walk into a room and lighting adjusts. No game engine — the model learns environment dynamics from data alone.

The researchers see three use cases: safe AI agent training (simulate before real-world deployment), interactive education (walk through ancient Rome in class), and game and film prototyping. Project Genie is available to Google AI Ultra subscribers in the US.

For developers, the agent training application is the key signal — world models are sandboxed environments where AI agents safely learn physical tasks before deployment.

View the full update on blog.google

Google DeepMind

@GoogleDeepMindFeb 25

How does a single prompt become a navigable environment? 🌐 We asked the researchers behind Project Genie to explain the mechanics of world models and their potential for training future AI agents. 🧵

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from Google →

Keep reading

Google Integrates Street View Into Project Genie for Real World Simulations

Google DeepMind's Project Genie now allows users to transform real-world U.S. locations from Google Maps Street View into interactive, navigable 3D environments. By anchoring generative world models in real-world imagery, the update shifts AI simulation from purely imaginative landscapes to playable versions of actual places.

NVIDIAMay 9

NVIDIA Outlines Technical Roadmap for Scaling Robot Dexterity and Physical AGI

NVIDIA's Jim Fan presented Robotics: Endgame at Sequoia AI Ascent, a 20-minute technical roadmap for solving Physical AGI as a parallel to the LLM success story. He walks through why current VLAs fall short, the case for video world models as a second pretraining paradigm, World Action Models, EgoScale, a Dexterity Scaling Law, and DreamDojo, an end-to-end neural physics engine for scaling reinforcement learning in silico.

Google Research Benchmarks Gemini's 3D Object Generation Through Code

GoogleJun 7

Google Research Benchmarks Gemini's 3D Object Generation Through Code

Google Research introduced 3DCodeBench, a new benchmark evaluating AI models' ability to generate 3D objects using code. This benchmark, presented at CVPR2026, demonstrates how agentic AI can autonomously create complex 3D assets, highlighting the role of iterative refinement in improving model performance.

Runway Joins NVIDIA to Open Source World Models for Physical AI

RunwayJun 1

Runway Joins NVIDIA to Open Source World Models for Physical AI

Runway has become a founding member of the Cosmos Coalition, a global initiative with NVIDIA to develop open-source world models. The partnership aims to accelerate physical AI research by providing a shared ecosystem for models that can reason about and simulate the real world.