Google Genie 2

Google Genie 2 is a large-scale foundation world model developed by Google DeepMind that generates interactive, playable 2D environments directly from images or text prompts. It learns a general-purpose representation of game-like worlds, enabling it to create coherent physics, character dynamics, and environment interactions without task-specific training. Users can upload a single image or provide a short description, and Genie 2 will synthesize a controllable scene where agents can move, jump, and interact with objects according to learned dynamics.

The model supports frame-by-frame prediction and action-conditioned generation, making it suitable for rapid prototyping of game mechanics, simulation environments, and interactive demos. Researchers can use Genie 2 to study world modeling, planning, and reinforcement learning in rich, visually diverse settings without manually designing levels. Game developers and designers can explore early gameplay concepts, level layouts, or visual styles by turning sketches or reference art into playable prototypes. Educators and hobbyists can create simple interactive worlds for experimentation, learning, or creative exploration. By abstracting away low-level physics and environment coding, Genie 2 enables faster iteration on ideas and provides a flexible testbed for AI agents and human–computer interaction research.