Exploring Multistyle LoRA Models for Architectural Image Generation
Faculty Mentor
Travis Masingale
Presentation Type
Poster
Start Date
May 2025
End Date
May 2025
Location
PUB NCR
Primary Discipline of Presentation
Design
Abstract
Exploring Multistyle LoRA Models for Architectural Image Generation
This creative research project investigates innovative approaches to architectural visualization and rendering through use of artificial intelligence image generation, focusing on the development and integration of multiple Low-Rank Adaptation (LoRA) models with ControlNet refinement techniques. A primary objective of this exploration was to dramatically reduce the time required for high-quality architectural visualization—transforming a process that traditionally demands hours of rendering time into one that generates compelling results in mere seconds. Rather than pursuing a single multistyle model, the research evolved to concentrate on developing distinct, style-specific LoRAs that were then strategically combined within a unified workflow. Through systematic experimentation over ten weeks, the study addressed fundamental challenges in maintaining stylistic coherence while achieving precise compositional control in AI-generated architectural imagery. Key innovations included techniques for model integration that preserved the distinctive characteristics of architectural styles while allowing for creative hybridization. The resulting portfolio demonstrates how these approaches can generate architectural visualizations with greater stylistic diversity and compositional refinement than conventional methods. This work contributes valuable insights into workflow optimization for architectural visualization and explores the creative possibilities that emerge at the intersection of artificial intelligence and architectural representation, with potential applications in design exploration, conceptual development, and visual communication.
Recommended Citation
Carlson, Antoni WJ, "Exploring Multistyle LoRA Models for Architectural Image Generation" (2025). 2025 Symposium. 2.
https://dc.ewu.edu/srcw_2025/ps_2025/p1_2025/2
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Exploring Multistyle LoRA Models for Architectural Image Generation
PUB NCR
Exploring Multistyle LoRA Models for Architectural Image Generation
This creative research project investigates innovative approaches to architectural visualization and rendering through use of artificial intelligence image generation, focusing on the development and integration of multiple Low-Rank Adaptation (LoRA) models with ControlNet refinement techniques. A primary objective of this exploration was to dramatically reduce the time required for high-quality architectural visualization—transforming a process that traditionally demands hours of rendering time into one that generates compelling results in mere seconds. Rather than pursuing a single multistyle model, the research evolved to concentrate on developing distinct, style-specific LoRAs that were then strategically combined within a unified workflow. Through systematic experimentation over ten weeks, the study addressed fundamental challenges in maintaining stylistic coherence while achieving precise compositional control in AI-generated architectural imagery. Key innovations included techniques for model integration that preserved the distinctive characteristics of architectural styles while allowing for creative hybridization. The resulting portfolio demonstrates how these approaches can generate architectural visualizations with greater stylistic diversity and compositional refinement than conventional methods. This work contributes valuable insights into workflow optimization for architectural visualization and explores the creative possibilities that emerge at the intersection of artificial intelligence and architectural representation, with potential applications in design exploration, conceptual development, and visual communication.
