Google DeepMind Launches Project Genie Prototype To Create Model Worlds
Google DeepMind Launches Project Genie Prototype To Create Model Worlds
https://quantumzeitgeist.com/google-deepmind-project-genie-generative-ai/
Publish Date: 2026-01-30 10:28:00
Source Domain: quantumzeitgeist.com
Google DeepMind is today launching Project Genie, an experimental research prototype granting Google AI Ultra subscribers in the U.S. the power to build and explore interactive worlds. Leveraging the capabilities of Genie 3, this innovative system moves beyond static 3D environments by generating real-time paths and dynamic interactions as users navigate and remix creations built from text and image prompts. “Unlike explorable experiences in static 3D snapshots, Genie 3 generates the path ahead in real time as you move and interact with the world,” explains the company. This marks a significant step towards Google DeepMind’s broader ambition of building Artificial General Intelligence (AGI) by simulating the complexities of real-world environments and opening new avenues for applications ranging from robotics to storytelling.
Genie 3 World Model Enables Real-Time Interactive Environments
(18+). The core of Project Genie revolves around three capabilities: world sketching, exploration, and remixing. Users initiate creation by prompting with text and images, building “a living, expanding environment” and defining their mode of exploration – from walking to flying. A feature called “World Sketching” integrates with Nano Banana Pro, allowing for image previews and fine-tuning before entering the generated world. Beyond creation, the system facilitates real-time path generation as users navigate, adjusting the camera based on their actions.
Existing worlds can also be remixed, building upon previous prompts or drawing inspiration from a curated gallery. “Building on our model research with trusted testers,” Google DeepMind is now expanding access to gather user insights. While acknowledging current limitations – including occasional inconsistencies in realism and character control, and generation limits of 60 seconds – the team is actively working on improvements, with a future goal of broader accessibility.