Text-to-image models reveal specific color-emotion associations

Area of Science:

• Computational psychology
• Affective science
• Computer vision

Background:

• Text-to-image models excel at learning image-tag associations from vast datasets.
• These models offer a novel approach to investigating the psychological links between color and emotion.

Purpose of the Study:

• To explore the psychological relationships between colors and emotions using text-to-image generation.
• To identify specific color features and their impact on perceived emotions.

Main Methods:

• Generated images based on emotion descriptions varying in valence, arousal, and dominance.
• Extracted color features, including chroma and L*a*b* values, from generated images.
• Analyzed color-emotion associations using computational methods.

Main Results:

• A joint effect of red and chroma was found to evoke joy, rage, and negative powerlessness.
• Lightness emerged as crucial for conveying serenity, threat, and stress/relief.
• Dominance significantly influenced the understanding of nuanced color-emotion interactions.

Conclusions:

• Color-emotion associations are complex, involving interactions between multiple color attributes.
• Specific combinations of color elements, not just isolated colors, effectively convey emotions.
• Dominance is a key dimension for understanding these nuanced relationships.