End-to-end logic chain of the proposed image-to-parametric modeling framework
Natural language descriptions and architectural images are translated by an LLM into executable C# code, enabling the generation of adjustable parametric 3D models within a computational design environment.

A set of architectural and geometric references used to evaluate model generation from visually regular forms
The examples include buildings and parametric models that exhibit clear repetition, periodicity, or rule-based geometric patterns.

An iterative generation process starting from simple and highly regular visual inputs
Different image preprocessing strategies and prompt refinements are applied to improve the quality and controllability of the generated parametric models.