Show HN: Agents infer Lego mechanisms from sources, then animate them

ldraw-mecha is a toolchain for turning existing LDraw LEGO models into Blender animations with AI agents. The project’s central claim is that an agent can infer how a model works by examining its annotated .MPD source: it identifies submodels and mechanisms, uses part descriptions to reason about their behavior, and plans how their components interact. To connect that reasoning to the 3D scene, the project preserves matching annotations on nodes in an exported GLB file. The agent can then use Python tools in Blender to build control rigs and animate the corresponding parts. The developer reports that several agents produced fairly consistent mechanism inferences, and that the end-to-end process has worked, though not perfectly.

The repository provides the tools, guides, sample assets, and setup scripts needed to run this workflow locally, including Blender, LeoCAD, an LDraw parts library, and utilities for inspecting and annotating models. Users can select pre-annotated models and follow a step-by-step process guide. Demonstrations include a Mercedes-Benz Arocs crane, a Fabuland plane, and a minifigure movement catalog; the first example was simplified to animate only the crane because the full truck was too complex. The animations use original LEGO designs rather than AI-generated models. Linux and Windows/WSL support are described as untested or in progress, and setup requires Python 3.12 or later. The project is aimed at experimenting with agent-assisted mechanical interpretation and animation, not promising reliable automation for every model.