With the rapid development of artificial intelligence technology, large models have become the key force to promote multimodal learning. However, the performance of these models is largely limited by the choice and design of prompt words. As the bridge between the model and the data, the optimization of prompt words is very important to improve the performance of the model in image description, video content understanding, question-answering system and other tasks. In this paper, we propose a series of innovative cue-word optimization strategies, including automatic cueing, active cueing, directed stimulus cueing, multimodal thought chain cueing, and graph cueing, and demonstrate their effectiveness in boosting model performance through extensive experiments on publicly available multimodal datasets.