"Prompt-based AI music generation platforms provide users a way to generate music audio recordings by giving textual prompts. Yet to be studied is how the generated music relates to the corresponding text prompts in terms of sentiment. This paper begins to examine the alignment between the valence (pleasantness) and arousal (intensity) of text prompts to that of the generated music. We fine-tune a masked language model to infer valence and arousal (V-A) of text prompts,and train a bimodal deep neural network to predict V-A for music generated from those prompts. We apply these models to a dataset of 6,086 text prompt-music pairs collected from two state-of-the-art AI music generation platforms (Suno and Udio). We examine specific pairs and the inferred V-A ratings for them, and explore how the two platforms respond to prompts involving adjectives specific to each quadrant of the V-A space. Our preliminary results indicate a significant but weak correlation between the inferred V-A ratings of text prompt and generated music, but formal human listening tests are necessary to validate this. We find some clear examples of text for which sentiment is not clear or relevant."