Human-AI Co-Creation for Interior Design: Integrating Scene Graphs, Diffusion Models, and Lighting Transfer

Abstract

The interior design process is often complex and time-consuming, posing significant challenges for designers. To address this issue, this study proposes an AI- assisted design platform that integrates Large Language Models (LLMs), ControlNet, and Diffusion Models, while incorporating the LumiNet model to enhance the realism and consistency of lighting in generated interior design images. The platform architecture is grounded in the Double Diamond design thinking framework, following four key stages: problem discovery, requirement definition, idea development, and prototype testing. This approach aims to significantly improve both the efficiency and creativity of the interior design process. An interactive prototype platform was developed, enabling designers to generate design drafts and photorealistic images with realistic lighting by simply inputting spatial layout images and style preferences. By utilizing prompts, designers can quickly adjust object relationships and obtain outputs that align with their expectations. The usability and user experience of the platform were evaluated through practical testing conducted by professional interior designers. The evaluation employed the System Usability Scale (SUS) and the User Experience Questionnaire (UEQ) to assess the platform’s practical effectiveness and application potential.