Hyper3D launches Agentic Mode, pushing 3D generation from Production-Ready to Agent-Ready

量子位 · wechat · 2026-09-22

After GPT-6 Astra made one-prompt 3D modeling viral, Hyper3D (Yingmo) argues general models handle task understanding while specialized 3D models still own high-quality assets with rich surfaces — and that the new user of 3D tools is the Agent itself.

Quantbitai tested the newly launched Agentic Mode: five messy product photos (mixed angles, scene shots, spec cards) went in raw; the system identified outlines, distinguished angles from lighting, filled in gaps across images, excluded desk clutter, and produced a complete retro desk fan. A full set of residential architectural drawings (floor plan, elevations, dimension notes) was also parsed into a 3D building without prompts.

Generated assets stay editable: natural-language parameter tweaks, localized material swaps, animation presets, and movable-part support for things like robotic hands. Combined with the earlier MCP interface, external agents can call Hyper3D Rodin directly. The competitive bar is shifting from Production-Ready to Agent-Ready.

Related event: Hyper3D Launches Agentic Mode: 3D Modeling Goes Agent-Ready(5 posts)→

Original post →

More from Multimodal

Multimodal channel →