MusiChat Enables Conversational, Iterative AI Music Creation

Recent advancements in AI music generation have empowered users to create full musical pieces from natural language prompts. However, most existing systems operate on a "prompt-and-regenerate" model, which makes iterative refinement cumbersome. Users are often forced to recreate compositions from scratch rather than evolving existing musical ideas. To address this, researchers have developed MusiChat. MusiChat is a novel conversational "vibe composing" system designed for collaborative human-AI music creation through natural language interaction and iterative refinement. At its core is a hierarchical controllable music generation framework that distinctly separates the generation of lyric-aligned musical structure from the expressive surface realization. This separation allows for flexible stylistic transformations and edits that preserve the underlying musical structure. The system integrates a large language model with a hybrid symbolic music engine, utilizing a memory-augmented architecture to maintain the active composition state and user history across interactions. An intelligent intent-routing mechanism further enables efficient interpretation of both precise musical edits and open-ended creative requests. Instead of regenerating entire compositions, MusiChat incrementally transforms an evolving musical artifact while retaining relevant structure and user intent. Objective analysis and human studies confirm MusiChat's effectiveness, showing high accuracy in multi-turn interactions and favorable user ratings for melody naturalness and musical quality, demonstrating its capability for coherent, interactive co-creation.