Tencent Hunyuan Unveils WorldClaw: Text-to-Editable 3D Open Worlds
Tencent Hunyuan has released WorldClaw, a new 3D generation project with a demo website. The system uses an agent-based workflow to generate large-scale, explorable 3D open worlds directly from text prompts, showcasing impressive 3D asset generation.
Confirmed
- Core mechanism: WorldClaw is an agent-based workflow that decomposes prompts into regions, terrain, and assets, then automatically generates and places textured mesh objects in Blender, with agents also fixing objects.
- Fine-grained editability: Unlike video or Gaussian Splatting, scenes consist entirely of editable independent 3D objects, allowing individual movement, deletion, or replacement of elements like trees, rocks, and houses.
- Use cases: Suitable for urban planning visualization, educational exhibits, and tourism concept drafts.
Unconfirmed
- Only a demo site is live; author @UncleMarty hopes for open-source weights for local testing, but official plans are unconfirmed.
Why it matters
WorldClaw bridges text and entire editable 3D worlds, with its agent architecture and fine-grained control overcoming previous limitations, offering high efficiency and freedom for concept design and content creation.
2026-08-09 ~ 2026-08-11 · 5 related posts
Primary sources
- Tencent Hunyuan Announces WorldClaw 3D Generation Project — Uncle___Marty ·
- Tencent Debuts WorldClaw: Generating Editable 3D Worlds from a Single Prompt — Scobleizer · 2026-08-09
- [source] Tencent Hunyuan Announces WorldClaw 3D Generation Project — Uncle___Marty · 2026-08-09
- Generate Editable 3D Worlds from Text: Tencent's WorldClaw — aigclink · 2026-08-11