MIRA: A Musical Intent Refinement Agent for Aligning Text-to-Music Generation with User Intent

2026-10-07

AI Summary

This study addresses the challenge that existing text-to-music generation systems struggle to precisely align with user intent, while single scalar scores fail to diagnose specific deficiencies. We propose MIRA, a test-time agent framework that decomposes user requests into verifiable, fine-grained intent criteria, guiding prompt optimization through iterative generation, verification, and feedback-driven tree search. As the first test-time agent framework based on fine-grained intent criteria, MIRA establishes closed-loop control transitioning from black-box generation to intent alignment. Furthermore, it introduces trajectory-aware tree search, automatic prompt revision, and MuRA-Bench, a benchmark constructed by multiple domain experts. Experimental results demonstrate that our approach significantly enhances the intent alignment capabilities of open-source models, achieving performance comparable to commercial systems such as Suno.

Abstract

Text-to-music systems produce increasingly convincing audio, yet evaluation reveals little about whether the result matches user intent. A global text-audio relevance score can overlook the implicit intent in underspecified prompts and mask failures in specific requirements, such as instrumentation, structure, rhythm, or mood progression. To bridge this gap, we formulate text-to-music intent alignment as satisfying a per-request rubric of independently verifiable items covering both a request's explicit requirements and its implied musical intent. Scoring items individually makes evaluation diagnostic by intent source and musical dimension, rather than a single opaque score. We instantiate this as MuRA-Bench, a benchmark of real-world platform requests curated by music experts. We further propose MIRA (Musical Intent Refinement Agent), a test-time agent that first grounds a request's intent into rubrics, then searches over prompt revisions for a black-box generator under a bounded budget, iteratively generating music, verifying it against the rubrics, and using this feedback to guide a trajectory-aware tree search. Experiments across open-source and commercial backends show that MIRA improves intent alignment, enabling an open-source generator to achieve performance comparable to representative commercial systems (e.g. Suno and Mureka). Project page: https://mirareview.github.io/.

Problem

• Text-to-Music Generation
• Intent Alignment
• Music Evaluation
• User Intent
• Prompt Underspecification

Innovation

• Text-to-Music Generation
• Intent Alignment
• Rubric-based Evaluation
• Test-time Agent
• Tree Search