From Abstract Noise to Architectural Form: Designing Diffusion Models for Efficient Floor Plan Generation
Keywords: Architectural Design Automation, Generative Models, Diffusion Models
TL;DR: This paper demonstrates how tailored diffusion models can efficiently generate detailed, professional-grade architectural floor plans.
Abstract: In contemporary architectural design, the generation of innovative and efficient floor plans remains a critical challenge. This research introduces a novel application of diffusion models, specifically adapted for the generation of architectural floor plans. Unlike traditional generative models that broadly target image generation, our approach harnesses the state-of-the-art in diffusion technology to produce detailed, functional, and visually appealing architectural designs. We demonstrate that diffusion models, when finely tuned and conditioned, not only embrace 'implicit, human-learned' architectural semantics but also enhance design efficiency and creativity. The paper details our methodology from adapting the U-Net architecture within diffusion frameworks to incorporating advanced upscaling techniques, significantly reducing computational overhead while maintaining high-resolution outputs. Our results show a promising direction for integrating AI in architectural design, opening new avenues for automated, creative design processes that could revolutionize the industry.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 14136
Loading