CCMSAI 2026

Current discussions of artificial intelligence in music often focus on AI-generated artifacts and whether machines can compose autonomously. This paper shifts attention from finished outputs to the broader creative workflow, asking how generative AI can participate in audiovisual composition without directly generating the final score. Through two practice-based case studies, this paper examines AI across conceptual development, prompt mediation, material generation, research, evaluation, and reflection. In the first case, conversational AI supported concept and prompt development, while an image-to-video system produced visual material that required substantial human editing before independently composed music was added. The second case examines a Dunhuang-inspired dance film in which AI supported cultural research, theoretical inquiry, and audiovisual ideation, while verification, composition, orchestration, synchronization, and aesthetic judgment remained human-directed. Drawing on human–AI co-creativity research, Vilém Flusser’s theory of technical apparatuses and play, and Michel Chion’s and Nicholas Cook’s theories of audiovisual relations, the paper proposes a shifting-role co-creative workflow. It argues that AI can shape composition before a single note is written, while artistic judgment, cultural responsibility, and final creative accountability remain human.