AesMood: An Intelligent System for Generating Mood Boards with Aesthetic Computing

Rong, Rong and Liu, Shuyi and Zhang, Hanling (2025) AesMood: An Intelligent System for Generating Mood Boards with Aesthetic Computing. In: P/References of Design. Cumulus conference proceedings series, 1 . Cumulus Association, Moholy-Nagy University of Art and Design Budapest, Budapest, pp. 715-734. ISBN 978-952-7549-02-5

Text 388ee797-7020-4afc-abb7-a4f1e4acc1f1 - Published Version Download (5MB)

Abstract

In the design process, manually constructing mood boards is a difficult, multi-step task requiring significant time and effort to gather and select images. To improve the efficiency of mood board creation, we introduce AesMood, an intelligent mood board generation system based on image generation and aesthetic computing. AesMood utilises the Stable Diffusion 2 image generation algorithm, which enhances the diffusion model by introducing a cross-attention layer into the model architecture and substantially improves visual fidelity. To mitigate the issue of inconsistent aesthetic quality among generated images, AesMood integrates the Style-specific Art Assessment Network (SAAN). SAAN efficiently extracts both style-specific and generic aesthetic features to evaluate and ensure the high aesthetic quality of the images for mood boards. In the preliminary phase of designing the system, interviews were conducted, which yielded three design requirements: meet user needs, editability, and simplicity of use. Three important functionalities are encoded in the AesMood system, including intelligent mood board generation, aesthetic scoring, and mood board editing. We invited 20 designers to employ our system, and its performance was assessed across four dimensions - usefulness, ease of use, learnability, and satisfaction - using the Likert scale in conjunction with semi-structured interviews. The results demonstrate high satisfaction levels, with participants noting the system's ability to broaden inspiration, stimulate creativity and imagination among designers, and enhance the efficiency of the ideation capture phase.

Actions (login required)

Edit Item