Scholarlink

This article explores the capabilities of large-scale diffusion models in generating architectural representations, focussing on insights derived from the Large City Architecture—Generated Cityscape dataset (LCA-GCS). This dataset, comprising 1,060,166 AI-generated images depicting architectural features across 5,856 global cities, provides a unique lens for examining how AI models interpret and represent architectural concepts across diverse urban contexts. I analysed patterns of architectural representation, regional variations, and inherent biases within these models, highlighting the potential of synthetic data to illuminate previously obscured relationships in architectural production and urban form. The findings demonstrate the capacity of Generative AI to map latent architectural worlds, revealing both the strengths and limitations of these models in understanding and generating the built environment.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Generating Latent Worlds: Mapping Architectural Types and Their Regional Differences Within Large-Scale Image Models

  • 章节

  • 发表日期：

    2026年

查看全文
  • Daniel Koehler

摘要

This article explores the capabilities of large-scale diffusion models in generating architectural representations, focussing on insights derived from the Large City Architecture—Generated Cityscape dataset (LCA-GCS). This dataset, comprising 1,060,166 AI-generated images depicting architectural features across 5,856 global cities, provides a unique lens for examining how AI models interpret and represent architectural concepts across diverse urban contexts. I analysed patterns of architectural representation, regional variations, and inherent biases within these models, highlighting the potential of synthetic data to illuminate previously obscured relationships in architectural production and urban form. The findings demonstrate the capacity of Generative AI to map latent architectural worlds, revealing both the strengths and limitations of these models in understanding and generating the built environment.