Hybrid Spaces

C&C • Creativity and Cognition (2026)

MoodPrism

Surfacing the Subjective Mood of Visual Content with MLLM-Generated Mood Profiles

Description

Mood is a central factor in visual design, shaping both the creative choices of creators and the ways viewers interpret content. Current design tools typically reduce mood to coarse tags supplied by creators or generated by systems. However, mood is inherently subjective, and coarse tags often fail to align with how designers, viewers, and systems experience and communicate mood. We introduce mood profiles, a granular representation of mood as multiple weighted tags grounded in explanations that tie back to visual attributes. We present the MoodPrism pipeline which leverages Multimodal Large Language Models (MLLMs) to generate mood profiles from graphic designs. We evaluate our approach through a crowd-based study to assess how mood profiles align with human perception of mood, and how they differ from coarse tags specified by creators. Finally, we present interactions that demonstrate how nuanced mood profiles can support mood-driven design and search.