Breaking the Group Size Barrier: Parameter-Efficient Group Dance Generation with Chain-of-Dancers

2026-10-08

AI Summary

This study addresses the limitations of existing group dance generation methods, which struggle to scale to variable group sizes and frequently confuse dancer identities. To overcome these challenges, this work proposes ChainDance, a framework that reformulates group dance generation as a chained decomposition of individual dancers. Built upon a frozen single-dancer diffusion backbone, the method introduces role-aware text and group-aware motion encoders to preserve individual characteristics. Furthermore, it ensures spatial coordination by integrating graph convolutional networks, conditional distribution sequence modeling, and training-free noise optimization, thereby enabling cross-scale generalization within a single model. Experimental results demonstrate that ChainDance achieves state-of-the-art performance on the AIOZ-GDance dataset while reducing parameter count by 3–4× and training time by 3–6×.

Abstract

Group dance generation aims to synthesize coordinated multi-dancer choreography from music, with broad applications in animation and interactive content creation. This task requires modeling dense inter-person dependencies to ensure spatial coordination, while naturally preserving individual dancer identities. Existing approaches model all dancers jointly with end-to-end transformers, which tie the architecture to a fixed group size and entangle per-dancer identities across frames. We propose ChainDance, a scalable framework that reformulates group dance generation as a Chain-of-Dancers: a sequential decomposition over per-dancer conditional distributions, allowing a single model to scale across variable group sizes without retraining and naturally preserving per-dancer identity. Built on a frozen single-dancer diffusion backbone, ChainDance introduces two lightweight modules: a Role-Aware Text Encoder (RATE) for per-dancer semantic conditioning, and a Group-Aware Motion Encoder (GAME) that aggregates previously generated dancers via a distance-weighted graph convolutional network, and incorporates a training-free noise optimization procedure at inference time to enforce global spatial coherence. Experiments on AIOZ-GDance demonstrate that ChainDance achieves state-of-the-art motion quality and group coordination while structurally preserving per-dancer identity, with $3$-$4\times$ fewer parameters and requiring $3$-$6\times$ less training time compared to prior approaches.

Problem

• Group dance generation
• Variable group size
• Dancer identity preservation
• Spatial coordination
• Parameter efficiency

Innovation

• Group Dance Generation
• Chain-of-Dancers
• Parameter-Efficient
• Diffusion Model
• Graph Convolutional Network