Genie 3
DeepMind's world model that generates explorable 3D worlds from a text prompt in real time.
About
Genie 3 is Google DeepMind's frontier world model: type a short prompt and it generates an interactive 3D environment you can walk around in, rendered at 720p and 24 frames per second. Unlike video generators, the world responds to your actions and holds together for several minutes, with about a minute of visual memory, so objects stay where you left them. You can also change the world mid-session with promptable world events: alter the weather, add characters, or reshape the landscape by typing. A 2026 upgrade grounded generation in Street View imagery, letting it build navigable simulations of real places, and Waymo uses it to simulate rare driving scenarios for robotaxi training.
It is a research model, not a game engine. Worlds last minutes, not hours, so there are no quests or persistent progression. Access has been limited, with newer features rolling out first to select Google AI subscribers, and independent benchmarks of its consistency claims are thin.
What it can do
- Real-time interactive 3D world generation
- Promptable world events during runtime
- ~1 minute visual memory for consistency
- Street View-grounded real-location simulation
- 720p at 24 fps generation
Setup
Access has been through Google's research previews and subscriber rollouts; check DeepMind's announcements for current availability.
Prerequisites
A Google account; newer features may require a Google AI subscription.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 52; Trending score 82/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.