Background. Generative AI is reshaping popular-music creation and pedagogy, raising the dual challenge of developing production fluency while engaging responsibly with cultural heritage. This paper reports on a 2024 undergraduate composition course in China that embedded AI as a collaborative partner across the end-to-end workflow while engaging Qinghai’s intangible vocal traditions.Aim. The study examines: (1) how generative AI can be integrated to scaffold student creativity and production literacy; and (2) how such integration influences intercultural awareness and the stylistic fidelity of works drawing on Qinghai heritage.Method. A design-based approach combined project-based learning, supervised creative practice, and iterative reflection. Students employed AI tools for ideation, lyric drafting, composition, arrangement, mixing, and mastering, with staged human-in-the-loop checkpoints and mandatory disclosure of AI assistance. To ground creative decisions in living tradition, a field trip to Xining (Qinghai) documented vocal and song materials with recognized bearers of local intangible cultural heritage under informed consent and institutional guidelines; recordings served as culturally anchored references rather than training data. Evidence comprised student artifacts (tracks), reflective journals, peer/instructor rubrics, and external juried outcomes, analyzed through qualitative thematic coding and descriptive aggregation.Results. The course culminated in an original album themed on Qinghai heritage, co-produced with a provincial culture and tourism organization; selected tracks received awards at a juried international electronic-music composition competition in China (2025), adjudicated by an international jury. Students demonstrated expanded ideational fluency, more diverse harmonic/arrangement strategies, and improved production literacy across mixing and mastering. Reflections revealed sharper criteria for aligning generative outputs with heritage aesthetics, as well as heightened ethical sensitivity. Instructor analyses indicated faster iteration and broader option exploration via AI, while field recordings anchored students’ work against style drift.Conclusions and implications. When treated as a pedagogical partner, generative AI functioned both as a creative scaffold and as a careful cultural translator, enabling students to engage with and reinterpret heritage within contemporary popular-music frameworks without erasing stylistic integrity. The study proposes a replicable curricular model for higher education: (a) scaffold AI integration across stages with human verification; (b) pair classroom creation with consent-based field documentation; (c) triangulate assessment through artifacts, reflections, and external juries; and (d) embed ethics, attribution, and IP literacy. This dual focus on pedagogy and cultural responsibility can cultivate musicians who are technologically adept, culturally sensitive, and creatively innovative.