Recently, text-to-image generation tasks have achieved remarkable progress, enabling the production of appropriate images from natural language descriptions. Although existing models can generate images that align with textual prompts, they still face significant limitations when dealing with abstract emotions. Thus, the EmoGen method introduces the task of Emotional Image Content Generation (EICG) for the first time. In this paper, we propose EmoPrompt \(+\) , a novel approach designed for the task of EICG. Specifically, we employ prefix language modeling to train an emotion-only text decoder, and we use dedicated emotion residual blocks to enhance the CLIP text encoder, improving its sensitivity to emotional content. This design endows abstract emotional concepts with richer semantics and enhances their intrinsic emotional representations. Experimental results demonstrate that our approach can effectively generate emotional image content and substantially outperforms existing state-of-the-art methods.