Google DeepMind Unveils Genie 2, AI Model for Generating Playable 3D Worlds

The new AI tool can create interactive virtual environments from simple prompts, advancing possibilities in game design and AI training.

Genie 2 is a large-scale AI model capable of generating playable 3D environments from a single image or text prompt.
The model supports interactive elements like object manipulation, physics effects, and multiple perspectives, including first-person and isometric views.
Generated worlds can be explored by humans or AI agents for up to 60 seconds, though longer durations introduce visual inconsistencies.
DeepMind envisions Genie 2 as a tool for prototyping game environments and training AI agents in diverse, controllable scenarios.
The model builds on advancements in diffusion and transformer architectures, using large-scale video datasets for training.