Skip to content

MeshLog 2.0

Contributors

  • Zixi Zhang (M.S. DT '27)
  • Ching-Wen Chen (M.S. DT '27)

Class

DESIGN 6298 Coding for Design II

Instructors

This project develops MeshLog 2.0, an LLM-driven computational design framework for editable mesh grammar modeling. It addresses the gap between manual procedural modeling, which requires low-level operational knowledge, and text-to-3D generation methods, where step-by-step construction logic is often not exposed. The system incorporates base geometry selection, supports both global mesh-level generation and local face-level editing, builds a ground-truth mesh-JSON dataset for few-shot prompt optimization, and uses VLM inference to connect visual geometric descriptors with procedural transformation parameters. Together, these components enable natural-language-driven mesh generation with procedural editability, real-time iteration, and sustained designer control.