EmoCtrl: Generating Images with Controlled Content and Emotion

This paper addresses a significant gap in text-to-image generation by focusing on both content fidelity and emotional expression. Existing models often struggle to balance these two aspects. EmoCtrl's approach of using a dataset annotated with content, emotion, and affective prompts, along with textual and visual emotion enhancement modules, is a promising solution. The paper's claims of outperforming existing methods and aligning well with human preference, supported by quantitative and qualitative experiments and user studies, suggest a valuable contribution to the field.