Background: Due to the rapid development of artificial intelligence (AI) and the widespread adoption of online learning post-COVID-19, the metaverse has become an important strategy for innovative teaching. Objectives: This study aimed to investigate the impact of the metaverse on learning engagement, learning emotions, and creative performance in AI applications by analysing academic emotions through facial expressions. Methods: The study was conducted as a non-equivalent group pretest--posttest quasi-experimental construct involving 97 students from a public high school. The experimental teaching theme focused on AI image recognition and speech recognition, with the experimental group receiving online metaverse instruction and the control group receiving instruction through Google Meet. Results and Conclusion: Key findings include the following: the metaverse significantly and positively influenced cognitive, emotional, and social dimensions of learning engagement, but no impact was observed on the behavioural dimension; the metaverse had a substantial impact on the creative performance of AI application design, particularly in the empathy, definition, ideation, and testing stages; and the metaverse significantly affected academic emotions, with increased expressions of anger and sadness observed, particularly during the empathy, definition, and ideation stages of design thinking. Thus, the results of this study provide a pedagogical foundation for the adoption of the metaverse and auxiliary tools in teaching.