What problem does it solve? Creating coherent, free-viewpoint 3D environments from text briefs is difficult: ad-hoc scene assembly produces disconnected shots, untextured primitives, and worlds that cannot be edited after rendering. This Skill builds a persistent, deterministic Three.js scene graph with semantic regions, procedural terrain, explicit landmarks, and camera fly-throughs that remain editable after delivery. ## Core Features & Use Cases - Structured world specification: Author a single world_spec JSON defining seed, terrain, semantic regions, atmosphere, landmarks, and camera paths, keeping explicit user constraints separate from inferred details. - Two fidelity tiers: Use blockout for fast layout validation or production for licensed GLTF/GLB asset catalogs, PBR terrain layers, and textured hero frames. - Render-guided refinement: Inspect global, regional, and walk-level snapshots in cinematic, semantic, or wireframe modes, then fix only the affected region while preserving seeds and stable IDs. - Use Case: Given a brief like "a volcanic rift dividing a living valley with a dawn fly-through", produce an editable HyperFrames workspace with terrain, scattered assets, landmarks, and a 60-second camera path, then iterate up to three bounded refinement rounds before delivery. ## Quick Start Ask the agent to use threejs-world-generation to turn your world brief into a deterministic, editable Three.js world with a camera fly-through.
