Game2World (G2W) Engine is a full-stack framework designed to extract clean, world-model-ready training data from in-the-wild gameplay videos. It introduces a GameUI-Taxonomy to systematically categorize and ground game interface elements, enabling their removal from raw footage that otherwise entangles screen-space UI with underlying game dynamics. By stripping these game-specific biases and irrelevant visual elements, the framework aims to produce more generalizable training data for video world models.