Researchers have introduced EmoScene, a large-scale dataset designed to improve controllable affective image generation. The dataset contains 1.2 million images across over 300 scene categories, featuring a dual-space representation that includes discrete emotions and continuous VAD (valence-arousal-dominance) dimensions, alongside measurable appearance attributes. To demonstrate its utility, the team developed AffectCtrl, a system that enhances frozen diffusion models for categorical emotion generation and continuous control over VAD, brightness, and saturation, achieving high accuracy and strong correlations across control axes. AI
IMPACT Enables more nuanced control over emotional expression in AI-generated images, potentially impacting creative tools and media.
RANK_REASON The cluster describes a new dataset and associated model for image generation research, published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
