AlayaWorld: Long-Horizon and Playable Video World Generation

AlayaWorld presents an open-source, modular framework for building interactive generative worlds with autoregressive video world models. The system synthesizes future observations from the current world state and user actions, supporting open-ended navigation and actions such as combat, spell casting, and monster summoning. It covers the full development stack, including data preparation, model architecture, training, inference acceleration, and deployment. The authors also describe reproducible pipelines, reference implementations, evaluation tools, and documentation. The framework is trained on gameplay recordings and real-world videos, targeting both playable applications and future embodied-intelligence research.