3D-PreMise: Can Large Language Models Generate 3D Shapes with Sharp Features and Parametric Control?
Zeqing Yuan, Haoxuan Lan, Qiang Zou, Junbo Zhao
TL;DR
The paper tackles generating explicit 3D shapes with sharp engineering features under strict parametric control by leveraging LLMs to produce Blender bpy code that drives 3D modeling. It introduces the 3D-PreMise dataset of 57 prompt–code pairs and a specialized test program to evaluate parametric fidelity, enabling systematic study of generation strategies. The work finds that in-context learning and chain-of-thought prompting improve code generation quality, while a multimodal visual feedback interface can aid self-correction but has limitations in commonsense and complex geometry. The results highlight both the promise of LLM-guided parametric 3D modeling for industrial design and the need for dataset scale-up and targeted fine-tuning to enhance spatial reasoning and geometric accuracy.
Abstract
Recent advancements in implicit 3D representations and generative models have markedly propelled the field of 3D object generation forward. However, it remains a significant challenge to accurately model geometries with defined sharp features under parametric controls, which is crucial in fields like industrial design and manufacturing. To bridge this gap, we introduce a framework that employs Large Language Models (LLMs) to generate text-driven 3D shapes, manipulating 3D software via program synthesis. We present 3D-PreMise, a dataset specifically tailored for 3D parametric modeling of industrial shapes, designed to explore state-of-the-art LLMs within our proposed pipeline. Our work reveals effective generation strategies and delves into the self-correction capabilities of LLMs using a visual interface. Our work highlights both the potential and limitations of LLMs in 3D parametric modeling for industrial applications.