From a Pig to a Tentacled Mech: Teaching an Agent to Rig
Meshy introduces MeshyRig, an experimental project where an AI agent analyzes each 3D asset and decides how to build its skeleton instead of fitting a template. The team tried it on three generated assets, a pig, a tank and a two-tentacled mech, through skeleton, skinning and animation.
What the article says
Meshy, a 3D asset generation company, published a research post by Sky Shi on October 19, 2026: generation is measured in seconds, but rigging, meaning the skeleton, skin weights and controls that let a model animate, is still measured in days or even weeks. Automatic tools mostly handle humanoids and quadrupeds, there is no general skinning strategy, and modeling and rigging are coupled. Meshy presents MeshyRig, an experimental project in which an AI agent reads each asset and decides on its own where joints go, how weights are computed and how the animation is built.
The agent chooses among general templates, skeleton primitives and custom sandboxed code, then checks the result against metrics; export is blocked if skeleton validation fails. A human confirms the skeleton before weights are computed and polishes the final result. Three tests are described: a pig (34 joints, 12 animations), a tank (125 joints, 10 animations) and a tentacled mech (41 joints, 8 animations). This is research: the post does not say whether or when MeshyRig will be offered to users.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Meshy. The content and rights belong to the publisher: please read the original article.