Enhancing Local Ecological Adaptation in Multispecies Eco-Building Design Through Multimodal System

EasyChair Preprint 15736

Abstract

Multispecies ecological architectural design aims to create buildings that coexist harmoniously with the surrounding natural environment, playing an increasingly important role in enhancing the biodiversity and ecological resilience of urban environments. However, ecological architectural design requires expertise to tailor designs to local ecosystems, especially in terms of species selection and growth environment matching. This study introduces a multimodal system that enables non-professional designers to create ecologically adaptive, site-specific architectural designs. The system combines ChatGPT and diffusion models to analyze and synthesize visual and textual data, embedding local ecological characteristics into the design process. It supports the collection of local species data, which is then processed by GPT-4V (Vision) to generate detailed material descriptions, optimized through expert feedback. The minimally trained ChatGPT model, supported by the Segment Anything Model (SAM), predicts the ecological suitability of species integration across different areas, segmenting images to identify regions conducive to multispecies growth. Identified regions are further used to generate ecological designs via latent space diffusion, with a low-rank adaptation (LoRA) model, trained on local species data, enhancing the accuracy of ecological simulations. ControlNet and advanced prompt engineering are utilized to optimize the final design outcomes. This multimodal system integrates AI technologies such as transformer models and diffusion models, distinguishing itself from previous multimodal applications that mainly focused on style guidance and aesthetic generation. In contrast, this approach emphasizes improving building performance, offering a new method for incorporating ecological principles into architectural practice, and providing a practical tool for developing urban environments with biodiversity and resilience.

Keyphrases: Generative AI, Multimodal Large Language Model, Multimodal system, Multispecies Design, ecological architecture

@booklet{EasyChair:15736,
  author    = {Daxu Wei and Christiane Herr},
  title     = {Enhancing Local Ecological Adaptation in Multispecies Eco-Building Design Through Multimodal System},
  howpublished = {EasyChair Preprint 15736},
  year      = {EasyChair, 2025}}