Group dance generation method scales to variable group sizes efficiently
Breaking the Group Size Barrier: Parameter-Efficient Group Dance Generation with Chain-of-Dancers
Computer Vision and Pattern RecognitionSound
Summary
Generating dances for multiple people that stay coordinated and keep each dancer’s unique style is hard. The paper’s authors propose a new way to create group dances one dancer at a time, which allows their model to work for any number of dancers without retraining. They improve coordination while making the system smaller and faster to train than previous methods. Their method uses special encoders to understand each dancer’s role and how dancers relate to each other in space.
What this means in practice
- •For animation studios: Create synchronized group dance animations of varying sizes efficiently without needing to retrain for each group size.$Commercial implications: Enables production of customizable multi-dancer dance animations for entertainment and media applications with reduced computational resources.
- •For virtual reality developers: Generate coordinated group dance motions in interactive environments while preserving distinct character identities dynamically.
Authors
Jing Xu, Cunjian Chen, Qiuhong Ke
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.