Hugging Face researchers have unveiled WorldClaw, a novel agentic framework designed to generate large-scale, freely explorable 3D worlds from open-ended text prompts. The system tackles the long-standing challenge of maintaining global spatial coherence while delivering rich local content and explicit assets suitable for downstream editing and reuse.
WorldClaw operates on a coarse-to-fine principle. Planning agents first translate a text prompt into a structured specification covering regions, terrain, assets, materials, and spatial relations. The framework then constructs a globally coherent terrain foundation using semantic layouts, reusable assets, generative or procedural materials, and a region-aware height field.
For areas requiring high detail, WorldClaw generates terrain-conditioned compositions, reconstructs editable textured meshes, and recovers their placement on the terrain. Render-based agents further refine terrain, objects, appearance, and contacts to ensure visual fidelity and physical plausibility.
According to the paper, WorldClaw produces large-scale scenes with coherent spatial organization, visually compelling local content, and editable instance-level assets, all while preserving a consistent global terrain structure across diverse open-world prompts.