WorldClaw: AI Generates 3D Worlds from Text Descriptions

Tencent Hunyuan released WorldClaw — a system where an agent plans and assembles a 3D world on its own.

The user simply describes the desired scene. Then WorldClaw plans the landscape, generates buildings, plants, characters, and other objects, and then assembles everything in Blender.

Moreover, the system also checks the result. If objects are floating in the air, intersecting with each other, or look disproportionate, WorldClaw continues to make edits and re-renders the scene for verification.

After generation, everything remains editable: the landscape, buildings, and plants exist as separate 3D objects that can be moved, deleted, and modified, and then used in games or animation.

Currently, Claude Opus 4.8 handles planning, coordination, and verification, while specific tasks are performed by GPT-Image-2, SAM3, SAM3D, Hunyuan3D, and Blender.