Prompt-Optimized Music Generation: A User Feedback-Adaptive AI System

This paper presents a music generation system that adapts to individual user preferences by combining generative AI with interactive evolutionary computation (IEC). Current music generation systems struggle to produce content aligned with users’ personal tastes because users find it difficult to express their preferences in the specific prompts these systems require. Our approach uses IEC to automatically optimize prompts for a music generation AI based on user feedback. Users evaluate generated music pieces, and the system iteratively refines prompts to better match their preferences, eliminating the need for explicit prompt engineering or technical expertise. We evaluated the system through experiments with 15 participants who used the interface to generate personalized music over multiple generations. Results show that the system successfully adapts to user preferences, with approximately 93% of participants reporting that the final optimized music reflected their personal taste. This research contributes to the IEC field by (1) providing a computational framework for personalized content generation through prompt optimization of generative AI, (2) demonstrating the effectiveness of an efficient search method in word vector space based on user ratings, and (3) showing how users can explore their latent preferences without verbalizing them. These findings deepen our understanding of how to design generative systems that enable intuitive human-AI collaboration in creative applications.