AnyMachine guides AI agents to build and rig mechanical assets in Blender
By GameSlash
The experimental workflow turns reference images into articulated machines, including a 160-part walker built with Blender Python.

AnyMachine is an experimental Blender workflow that guides a coding agent through building, rigging and animating machines. Developer Apekshik published the project and examples on 18 November, including Stilt, a 30-metre walking machine built from 160 parts.
The creator says Claude Fable 5.2 in Claude Code followed the project's pipeline while a person chose the design and approved stages. Stilt has a 104-bone rig, working hydraulics and three animations. The Blender geometry, rig and actions are generated by a Python script; this is a documented example, not a one-click service.

The workflow begins with a concept image, then generates views and close-ups of the joints. The agent breaks the design into parts, scripts them in Blender, and compares renders with the references before refining the model. The repository includes the script and a GLB example, giving game creators a concrete starting point for articulated machines and props.
The project is experimental: its author notes that some model details are simpler than the reference and that the examples are not a hosted endpoint. The repository describes separate static-prop and generated-mesh workflows as well; results should be checked and adapted for the target game.