<p>Artificial intelligence-generated virtual lecturers are increasingly adopted in educational contexts as a cost-effective and scalable alternative to human instructors in lecture video production. These virtual lecturers can simulate human-like facial expressions, gestures, and voices, offering new possibilities for emotionally expressive teaching. However, limited research has examined how the emotional expressions of virtual lecturers and the frame of instructional scripts jointly influence students’ learning outcomes. This study employed a 2 × 2 between-subjects experimental design to investigate the effects of instructor emotion (positive vs. neutral) and script frame (gain-frame vs. loss-frame) on students’ learning performance, intrinsic motivation, emotion, and cognitive load. A total of 114 middle school students from mainland China were randomly assigned to one of four experimental conditions. The emotional expression of virtual instructors is controlled through facial expressions and intonation, while the script is managed by employing either a gain-frame or a loss-frame. The results showed that virtual lecturers displaying positive emotional expressions significantly enhanced students’ learning performance and intrinsic motivation compared with neutral ones. Furthermore, optimal learning outcomes and lower cognitive load occurred when the lecturer’s emotional tone aligned with the script frame (e.g., positive emotion with gain-frame). The study provides empirical evidence and practical implications for designing emotionally coherent and pedagogically effective virtual instructors in multimedia learning environments.</p>