---
title: "Music-to-Dance Generation via Atomic Movements"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.13978v1"
permalink: "https://feed.astralor.com/items/itm_9dab8691cf13d1673ac165c5"
published: "2026-07-15T16:03:55.000Z"
score: 79
channel: "ai"
tags: ["arXiv","Two-stage Generation","Atomic Movements","Large Language Models","Dance Generation","Music-driven"]
---

# Music-to-Dance Generation via Atomic Movements

## AI commentary

> This paper deserves attention as it offers a novel structural perspective for music-driven dance generation—atomic movements as semantic units, significantly enhancing controllability and interpretability. It is especially important for researchers in AI-generated content, virtual human animation, and interactive choreography.

## At a glance

This work introduces a structure-aware framework for music-driven dance generation, modeling choreography as a sequence of semantically interpretable atomic movements. By segmenting large-scale dance data and clustering into atomic movement groups, and using a large language model for semantic relabeling, a vocabulary of atomic movements is constructed. A two-stage generation pipeline first plans atomic movement types, durations, and timings conditioned on music, then synthesizes smooth and coherent motion. Experiments show significant improvements in structural coherence, rhythmic alignment, and perceptual naturalness over baselines, with enhanced interpretability and controllable editing.

## Key points

- Proposes atomic movements as semantically interpretable building blocks of dance
- Uses a large language model to semantically relabel and refine clustered movement groups
- Two-stage generation: atomic movement planning and transition-aware completion
- Outperforms baselines in structural coherence, rhythmic alignment, and perceptual naturalness

## Full text

Music-driven dance generation aims to produce human motion that is both rhythmically synchronized and semantically consistent with music. While recent neural approaches have achieved impressive visual realism, they typically model motion as a continuous signal and neglect its compositional nature, making generated dances structurally incoherent and difficult to control. In this work, we introduce a structure-aware framework that models choreography as a sequence of atomic movements-semantically interpretable motion events that serve as the building blocks of dance. To construct this atomic movement vocabulary, we first segment large-scale dance data and cluster them into atomic movement groups. We then employ a large language model to semantically relabel and refine the clusters, yielding a set of interpretable and reusable atomic movements. Based on these atomic movement annotations, we design a two-stage generation framework that mirrors the human choreography process. In the atomic movement planning stage, the model predicts the type, duration, and timing of atomic movements conditioned on the input music, forming a symbolic dance allocation. In the completion stage, a transition-aware generator synthesizes smooth and stylistically coherent motion conditioned on the planned structure. Extensive experiments demonstrate that our method produces dances with significantly improved structural coherence, rhythmic alignment, and perceptual naturalness compared to existing baselines, while providing enhanced interpretability and controllable editing through explicit structural representation.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_9dab8691cf13d1673ac165c5