EmoGen: Emotional Image Content Generation with Text-to-Image Diffusion Models

Recent years have witnessed remarkable progress in image generation task,where users can create visually astonishing images with high-quality. However,existing text-to-image diffusion models are proficient in generating concreteconcepts (dogs) but encounter challenges with more abstract ones (emotions).Several efforts have been made to modify image emotions with color and styleadjustments, facing limitations in effectively conveying emotions with fixedimage contents. In this work, we introduce Emotional Image Content Generation(EICG), a new task to generate semantic-clear and emotion-faithful images givenemotion categories. Specifically, we propose an emotion space and construct amapping network to align it with the powerful Contrastive Language-ImagePre-training (CLIP) space, providing a concrete interpretation of abstractemotions. Attribute loss and emotion confidence are further proposed to ensurethe semantic diversity and emotion fidelity of the generated images. Our methodoutperforms the state-of-the-art text-to-image approaches both quantitativelyand qualitatively, where we derive three custom metrics, i.e., emotionaccuracy, semantic clarity and semantic diversity. In addition to generation,our method can help emotion understanding and inspire emotional art design.