From Pattern to Prompt: Datasets and Workflows

From Pattern to Prompt: Datasets and Workflows is the supplementary repository for the article “From Pattern to Prompt: Translating Alexander into Latent Space,” published in the proceedings of SIGraDi 2025. The project explores how Latent Diffusion Models—specifically Stable Diffusion 1.5, SDXL, and SDXL Lightning—interpret architectural patterns from Christopher Alexander’s A Pattern Language. Across three experimental stages, we measured model performance in terms of generation time, variability, and accuracy, comparing different prompt strategies and their impact on output diversity. This repository provides: ComfyUI workflows (.json) used to generate all image sets. Image folders organized by stage and model. Prompt collections corresponding to each image and stage. All materials are openly shared to ensure transparency, reproducibility, and reusability in computational design and architectural research.