The Landscape Architect as Developer and Designer: Optimisation and Control of Diffusion Models for Landscape Architecture

The integration of Generative Artificial Intelligence (GenAI), particularly image-generating AI, into landscape architecture marks a decisive step in the digital development of this discipline. This

work examines how diffusion models can be demonstrably optimised and controlled to support creative and site-specific design processes within landscape architectural workflows. Focusing on the training

and fine-tuning of AI models, the study investigates how customised data and proprietary parameters influence the generation of spatially specific images and design results. A conceptual and technical

framework developed in this work, adapted to the open-source generative AI model Stable Diffusion, provides the basis for the practical implementation of training using Low Rank Adaptation (LoRA) and

DreamBooth as fine-tuning training methods. This study adopts the classical design process as a holistic framework: Three specific learning tasks – location, concept control and look – define the experimental

setup and function as identifiable integration points within the design workflow. The AI models are evaluated both qualitatively and quantitatively through these learning tasks. This paper contributes a

structured framework for controlling diffusion models in landscape architectural design. It further demonstrates how these control dimensions can be operationalised within practical design tasks. By

examining the interaction between human intent and algorithmic interpretation, this work reveals new possibilities for creative control and optimisation within AI-driven design workflows and positions

landscape architects as both designers and active developers of their tools.

Mehr lesen