Google Plugs Street View Into Project Genie - Generate Worlds From Any Real US Location

Google Plugs Street View Into Project Genie - Generate Worlds From Any Real US Location

0:00 / 0:58
News

Google Plugs Street View Into Project Genie - Generate Worlds From Any Real US Location

calendar_today Date:
schedule Duration: 0:58
visibility Views: 42
database
Summary Report

Google Labs has plugged Street View into Project Genie, so users can generate playable worlds anchored to a real US location via a Maps pin. The update also adds a Library and external sharing.

  • 01. Project Genie can now generate playable worlds tied to real US Street View locations via a Google Maps pin.
  • 02. Users pick a starting place, choose a style like Desert Sands or Ocean World, and describe their character.
  • 03. A new Library stores and organises generated worlds, with remix support for iterating on past creations.
  • 04. External sharing lets anyone watch a video preview, while Ultra subscribers can explore or remix the world.
  • 05. Rolling out globally to Google AI Ultra subscribers (18+) over the coming weeks.
Google Labs has integrated Street View data into Project Genie, Google DeepMind's interactive world model used by Waymo for training self-driving cars on edge cases. Users can now drop a Maps pin anywhere in the United States, describe a character, and Genie 3 will generate a playable world anchored to those real-world coordinates. The update introduces three key features: Street View grounding that ties generated content to actual locations, a Library system for saving and remixing creations, and external sharing capabilities that allow others to preview generated worlds. This marks a significant step beyond previous world generation systems that operated in isolation. The Street View integration represents the most notable advancement, as it bridges the gap between synthetic training environments and real-world navigation challenges. By grounding generated worlds in actual coordinates, the system becomes more valuable for training agents and robots that must operate in physical spaces. The feature is rolling out to Google AI Ultra subscribers globally over the coming weeks. Current limitations include video game-level visual quality rather than photorealistic rendering, and imperfect physics simulation—demonstrated by a character running through a cactus in the demo. Despite these constraints, the real-world grounding capability suggests potential applications beyond gaming into robotics and autonomous systems training.