DeepMind's Project Genie Learns to Simulate Real Places from Google Street View

DeepMind has expanded access to Project Genie for all Google AI Ultra subscribers worldwide and introduced a new capability to integrate with Google Street View, enabling the simulation of interactive video of real geographic locations.

What is Project Genie

Project Genie is an advanced generative video model from Google DeepMind that creates interactive video scenes based on text descriptions or images. Unlike conventional video generators that simply reproduce pre-recorded content, Genie builds a dynamic virtual world that responds to agent actions in real time. The model can show a person (or robot) in a scene performing various actions: walking in different directions, manipulating objects, interacting with the environment. With each action, the video updates, reflecting physical laws and cause-and-effect relationships between events. This makes the experience similar to controlling a game character in a video game, but based on neural network predictions rather than pre-recorded material.

New Integration with Street View

The new feature combines Project Genie's capabilities with Google Street View — a massive archive of panoramic photos of millions of places on Earth. Now, instead of imaginary or synthetic scenes, the agent can interact with real locations: historic city centers, parks, public spaces, landmarks. This transforms Street View from a static photo gallery into an interactive virtual world. A user can not only view a panoramic photo of St. Peter's Square but literally 'walk' through it, exploring architectural details, peering into shop windows, interacting with objects, and watching the image change according to their actions.

Where This Can Be Useful

Interactive video simulation of real places opens numerous practical applications:

• Travel planning and tourism — tourists can virtually explore a landmark before visiting
• Rehabilitation and accessibility — people with mobility limitations can remotely explore public spaces
• Architecture and urban planning — designers can simulate how new buildings integrate into existing spaces
• Robotics — neural networks for autonomous robots can train on real urban scenarios
• Education and culture — virtual tours of historic sites become fully interactive

Each of these applications requires Genie to achieve a high level of realism in predicting physical processes and human behavior.

Technical Challenges

Generating real interactive video requires enormous computational resources. The model must not only predict subsequent video frames with high accuracy but also do so with minimal latency to make interactions feel smooth and realistic. Small errors in predicting physics or human movements can quickly accumulate, destroying the illusion of reality.

Expanded Access

Previously, Project Genie was available only to a limited number of users. Now Google is expanding access to all Google AI Ultra subscribers worldwide. This will allow more developers, researchers, and enthusiasts to experiment with interactive video generation of real places. Expanding access to such advanced technology signals that the model has reached a certain level of stability and readiness for use in real-world applications.

What This Means

The boundary between static information (photos, videos, maps) and interactive AI simulations is gradually blurring. Project Genie combined with Street View is a significant step toward creating an 'alternative interactive reality' based on real geographic data. In the future, people will not just view places but actively explore them, interact with objects and environments, and train AI systems based on simulations. This fundamentally changes how we consume information about the world and interact with geographic space.