Google DeepMind has officially unveiled Project Genie, an ambitious experimental research prototype that transforms static prompts into infinite, interactive digital environments. Built upon the Genie 3 foundation model, the system moves beyond traditional video generation by creating "world models" that simulate the laws of physics and causal interactions in real-time as users navigate them. Unlike standard 3D snapshots, Genie 3 predicts the path ahead based on user input, effectively acting as a generative game engine that adapts to every movement and action.

The prototype, currently rolling out to Google AI Ultra subscribers in the U.S., integrates multiple models including Gemini and a new utility called Nano Banana Pro. This integration enables a feature called "World Sketching," where users can fine-tune their environments through generated or uploaded images before entering a first- or third-person perspective. The system emphasizes creative freedom, allowing users to define character movement—from walking and driving to flying—while simulating how those specific actions affect the surrounding digital ecosystem.

While the technology represents a significant leap toward Artificial General Intelligence (AGI) by teaching systems to understand real-world dynamics, DeepMind maintains a cautious rollout strategy. Current limitations include 60-second generation caps and occasional deviations from consistent real-world physics or prompt adherence. However, the ability to "remix" existing worlds and share explorations marks a turning point for generative media, shifting AI from a tool that creates static content to one that builds entire interactive experiences.

Google DeepMind has officially unveiled Project Genie, an ambitious experimental research prototype that transforms static prompts into infinite, interactive digital environments. Built upon the Genie 3 foundation model, the system moves beyond traditional video generation by creating "world models" that simulate the laws of physics and causal interactions in real-time as users navigate them. Unlike standard 3D snapshots, Genie 3 predicts the path ahead based on user input, effectively acting as a generative game engine that adapts to every movement and action.

The prototype, currently rolling out to Google AI Ultra subscribers in the U.S., integrates multiple models including Gemini and a new utility called Nano Banana Pro. This integration enables a feature called "World Sketching," where users can fine-tune their environments through generated or uploaded images before entering a first- or third-person perspective. The system emphasizes creative freedom, allowing users to define character movement—from walking and driving to flying—while simulating how those specific actions affect the surrounding digital ecosystem.

While the technology represents a significant leap toward Artificial General Intelligence (AGI) by teaching systems to understand real-world dynamics, DeepMind maintains a cautious rollout strategy. Current limitations include 60-second generation caps and occasional deviations from consistent real-world physics or prompt adherence. However, the ability to "remix" existing worlds and share explorations marks a turning point for generative media, shifting AI from a tool that creates static content to one that builds entire interactive experiences.