Research on Enhancing the Generalization Ability of Generative AI Based on K-means Clustering Algorithm: Text-to-Image Task for Wind-Responsive Passive Building Facade
Rongrong Liu, Adzrool Idzwan Ismail, Ye Lu · 2025
In recent years, as AI has grown fast, generative AI has been used a lot in building design.But when generative AI tries to create Wind-Responsive Passive Building Facades, it doesn't understand the moving look of the facade very well. This makes it bad at telling the important features and not good at working with different design cases.This study tries to make generative AI better at designing Wind-Responsive Passive Building Facades by using the K-means clustering algorithm.First, this study built a dataset with facade images, weather conditions, material types, and size info. Then, using the K-means clustering algorithm, we picked out key features and grouped the buildings by type. The goal is to give generative AI a better kind of visual data to learn from.Based on that, we made a three-level prompt structure to help generative AI work better with different design tasks.This study not only builds a clear method between finding building features and making prompts, but also gives ideas and steps for using generative AI more in sustainable building design.