This paper introduces PromptMorph, a design workflow that automates the generation of 3D parametric models using Large Language Models (LLMs) like ChatGPT, Python, and Grasshopper. It bridges natural language and image prompts with computational modeling, enabling architects to create editable parametric models without extensive coding. Central to the method is a novel image-based prompting technique tailored to design disciplines, translating conceptual and visual inputs into parametric geometries. As a proof of concept, PromptMorph demonstrates that LLMs can generate 3D models from prompts and refine them through iterative feedback. The study applies the workflow to eight geometry-based case studies, assessing performance through metrics such as accuracy, usability, iteration count, and error rates. Two additional freeform cases extend the method’s applicability. The framework highlights the potential of LLMs to streamline design workflows, reduce manual scripting, and accelerate exploration of alternatives. While limitations remain in AI reasoning and platform integration, the approach provides a foundation for further research in LLM-assisted computational design.