As existing LLMs expand in their real-world applications, the quality of the text generated has become increasingly off-target, outputting text that is coherent in terms of grammar and linguistics, but is unfactual and possibly entirely fabricated, a phenomenon known as hallucination. Though these LLMs have achieved tremendous success in their applications, this hallucination problem acts as a persistent bottleneck in their development. To make matters more dire, traditional strategies are often accompanied by a host of limitations in terms of parameter and data size, not to mention the high monetary and human workforce costs. This paper presents the possibility of an emotion-augmented interface based on the theory of the Wheel of Emotions to solve the issue of hallucinations in standing AI LLMs. The Wheel of Emotions theory presents emotions as a hierarchical and symmetric system. It delineates composite emotions as being a combination of basic emotions (like joy, sadness, anger, and fear) in varying intensities. Within the EAI foundation, this theory serves as a reference for emotional distribution and effectively aids in achieving the core objectives of emotion visualization, hallucination suppression, and emotional regulation. In this way, the EAI utilizes visual contrastive decoding + affective textual symbolization to produce enhanced veracity and emotional consistency in its outputs. This system recognizes emotional signals within the system and regulates them by converting emotional text into pictorial symbols to allow for emotional modulation in real time. The functionality of this system was verified through the core tasks of visual question answering, image captioning, and emotional vocabulary grounding. To test this, comparative experiments were conducted on the multimodal macro models CLIP, DALL-E, and ViLBERT, comparing the regular decoding strategy with the novel method proposed by the EAI on the MSCOCO and GQA datasets. It was found that the EAI decoding strategy improved the accuracy and F1 scores of the models in the MSCOCO dataset by 2% to 12%, with a similar range of improvement in the GQA dataset, hence proving that the EAI method both reduced hallucinations and improved accuracy, recall, and precision. Conclusively, EAI outperformed existing mainstream multimodal models regarding hallucination suppression, emotional coherence, and emotional control, solidifying its adaptability and functionality in generation tasks that are emotion-driven. This framework effectively paves the way for future innovations in LLMs in healthcare by way of trustworthy, empathetic AI systems.