Google's Project Genie Combines Street View Data to Simulate the Real World — A Key Breakthrough for AI World Models

Overview

Google recently announced the expansion of global access to its Project Genie, now available to Google AI Ultra subscribers, along with a new capability powered by Google Street View — simulating real-world locations. This move marks a deep integration of AI generative technology with geospatial data, opening up entirely new possibilities for virtual world construction.

What Is Project Genie?

From Game Generation to World Simulation

Project Genie is a Foundation World Model project developed by Google DeepMind. The concept of a "foundation world model" represents one of the cutting-edge directions in AI research — unlike large language models (LLMs) that focus on text understanding, world models aim to help AI build internal representations of the physical world, understanding how objects move, how space is organized, and how causal relationships work. This concept traces back to cognitive scientist Kenneth Craik's "mental model" theory proposed in 1943, and in the deep learning era, researchers like Yann LeCun have redefined it as a core pathway toward artificial general intelligence. Project Genie's "Foundation" positioning means it's analogous to pre-trained large models like the GPT series — learning universal world understanding capabilities through self-supervised learning on massive amounts of unlabeled visual data, then fine-tuning for specific tasks rather than training separately for each scenario.

Its core function is to generate interactive virtual environments from inputs such as images and text using AI. Unlike traditional game engines, Genie doesn't require developers to manually build scenes. Instead, it learns from vast amounts of visual data to automatically understand physical rules, spatial relationships, and environmental logic, thereby creating explorable 2D and even 3D worlds "from scratch."

Previously, Project Genie demonstrated the ability to generate playable game levels from a single image, attracting widespread industry attention. Now, its integration with Google Street View extends this capability from fictional scenes to the real world. Notably, Project Genie's technical evolution builds on recent advances in 3D scene generation research: Neural Radiance Fields (NeRF), proposed in 2020, first achieved the breakthrough of reconstructing high-quality 3D scenes from a small number of 2D images, followed by 3D Gaussian Splatting, which dramatically improved rendering speed. Project Genie's innovation goes a step further — it not only reconstructs static scenes but also endows them with "interactivity," enabling AI to understand which elements can be manipulated and how physical laws constrain motion. This represents a qualitative leap from "passive perception" to "active understanding."

A New Use for Google Street View Data

Since its launch in 2007, Google Street View has accumulated a massive database covering more than 220 countries and territories, with over 100 billion street-level images and a total distance exceeding 16 million kilometers. Its capture system combines multi-eye fisheye cameras, LiDAR, GPS, and IMU (Inertial Measurement Unit) sensors, simultaneously collecting 360-degree panoramic imagery and precise 3D point cloud data. This data is processed through Google's Structure from Motion (SfM) algorithms to reconstruct accurate three-dimensional spatial structures. It is precisely this multimodal data characteristic of "image + depth + pose" that makes it ideal training material for world models — AI can not only see how a scene looks but also understand its 3D geometric relationships. This forms the data foundation for Project Genie's generation of spatially coherent environments.

Previously, this street view data primarily served map navigation and virtual tours. Project Genie's new capability means that these street view images can be "understood" by AI models and transformed into interactive simulated environments. Users simply specify a real-world location, and the AI generates a virtual simulation of that location based on street view data, allowing users to explore and interact within it. This is not static panoramic browsing but rather a dynamic environment with physical logic and spatial coherence.

Global Availability for AI Ultra Subscribers

Subscription Strategy and Commercialization Path

Google AI Ultra is Google's flagship AI subscription service launched in 2025, priced at approximately $249.99 per month. It represents the highest tier in Google's subscription hierarchy, integrating the Gemini Ultra model, NotebookLM Plus, Google One cloud storage, and various cutting-edge experimental features. This pricing strategy directly competes with OpenAI's ChatGPT Pro (at $200/month) and Anthropic's Claude Max, collectively forming the premium battlefield of the current AI subscription market. Google's decision to bundle Project Genie within the Google AI Ultra subscription service — tying computationally intensive world model generation capabilities to a premium subscription tier — is partly a practical consideration for cost management and partly the familiar tech industry approach of "feature tiering."