Procedura Writes 3D Objects as LLM-Generated Assembly Code

The claim is unusual: an LLM writing code beats specialized 3D generators at making 3D things.

A paper on arXiv from Youtian Lin and colleagues introduces Procedura, which they describe as "a novel 3D modeling agent framework that writes an object as a procedural assembly, a parametric program whose named parts are joined by typed, machine-checkable mates." From a text prompt, the agent plans the object as an assembly graph and writes the program part by part, "solving each placement from the mated frames rather than guessing it, and admitting a part only once compile, mate, and connectivity checks pass."

A vision critic then refines the assembly "one diagnosed fix at a time." The same graph carries per-part materials and a simulator-validated articulation.

The headline result: evaluated on P3D-Bench "under its assembly judge," and on MechBench-36 (which the authors introduce as "our hard-surface benchmark"), "Procedura outperforms state-of-the-art native 3D generators and every prior 3D-code agent on judged quality, produces the sharpest edges of any method we evaluate, and is the only one whose output is an editable, part-structured program."

The abstract publishes no per-benchmark numbers, no ablations, and does not name the underlying LLM. The comparisons rest on a single assembly judge whose criteria are not detailed in the excerpt, and MechBench-36 is the authors' own construction.