Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FlowerDance: MeanFlow for Efficient and Refined 3D Dance Generation

Kaixing Yang, Xulong Tang, Ziqiao Peng, Xiangyue Zhang, Puwei Wang, Jun He, Hongyan Liu

TL;DR

FlowerDance addresses the twin challenges of quality and efficiency in 3D music-to-dance generation by integrating MeanFlow with a Physical Consistency Constraint, enabling high-fidelity motions with only a few sampling steps. The architecture uses a BiMamba backbone and Channel-Level Cross-Modal Fusion to support non-autoregressive, end-to-end generation and bidirectional temporal modeling. Extensive experiments on FineDance and AIST++ demonstrate state-of-the-art performance in both motion quality and inference efficiency, complemented by interactive motion editing capabilities. The work offers practical impact for real-time, controllable 3D character animation in virtual reality and entertainment applications.

Abstract

Music-to-dance generation aims to translate auditory signals into expressive human motion, with broad applications in virtual reality, choreography, and digital entertainment. Despite promising progress, the limited generation efficiency of existing methods leaves insufficient computational headroom for high-fidelity 3D rendering, thereby constraining the expressiveness of 3D characters during real-world applications. Thus, we propose FlowerDance, which not only generates refined motion with physical plausibility and artistic expressiveness, but also achieves significant generation efficiency on inference speed and memory utilization. Specifically, FlowerDance combines MeanFlow with Physical Consistency Constraints, which enables high-quality motion generation with only a few sampling steps. Moreover, FlowerDance leverages a simple but efficient model architecture with BiMamba-based backbone and Channel-Level Cross-Modal Fusion, which generates dance with efficient non-autoregressive manner. Meanwhile, FlowerDance supports motion editing, enabling users to interactively refine dance sequences. Extensive experiments on AIST++ and FineDance show that FlowerDance achieves state-of-the-art results in both motion quality and generation efficiency. Code will be released upon acceptance. Project page: https://flowerdance25.github.io/ .

FlowerDance: MeanFlow for Efficient and Refined 3D Dance Generation

TL;DR

FlowerDance addresses the twin challenges of quality and efficiency in 3D music-to-dance generation by integrating MeanFlow with a Physical Consistency Constraint, enabling high-fidelity motions with only a few sampling steps. The architecture uses a BiMamba backbone and Channel-Level Cross-Modal Fusion to support non-autoregressive, end-to-end generation and bidirectional temporal modeling. Extensive experiments on FineDance and AIST++ demonstrate state-of-the-art performance in both motion quality and inference efficiency, complemented by interactive motion editing capabilities. The work offers practical impact for real-time, controllable 3D character animation in virtual reality and entertainment applications.

Abstract

Music-to-dance generation aims to translate auditory signals into expressive human motion, with broad applications in virtual reality, choreography, and digital entertainment. Despite promising progress, the limited generation efficiency of existing methods leaves insufficient computational headroom for high-fidelity 3D rendering, thereby constraining the expressiveness of 3D characters during real-world applications. Thus, we propose FlowerDance, which not only generates refined motion with physical plausibility and artistic expressiveness, but also achieves significant generation efficiency on inference speed and memory utilization. Specifically, FlowerDance combines MeanFlow with Physical Consistency Constraints, which enables high-quality motion generation with only a few sampling steps. Moreover, FlowerDance leverages a simple but efficient model architecture with BiMamba-based backbone and Channel-Level Cross-Modal Fusion, which generates dance with efficient non-autoregressive manner. Meanwhile, FlowerDance supports motion editing, enabling users to interactively refine dance sequences. Extensive experiments on AIST++ and FineDance show that FlowerDance achieves state-of-the-art results in both motion quality and generation efficiency. Code will be released upon acceptance. Project page: https://flowerdance25.github.io/ .

Paper Structure

This paper contains 15 sections, 12 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Flow Matching Generative Models
  4. Music-to-Dance Generation
  5. Methodology
  6. Problem Definition
  7. Generative Strategy
  8. Model Architecture
  9. Experiment
  10. Dataset
  11. Quantitative Evaluation
  12. Qualitative Evaluation
  13. Ablation Study
  14. Qualitative Analysis on Motion Editing
  15. Conclusion

Figures (4)

  • Figure 1: FlowerDance not only generates refined motion with physical plausibility and artistic expressiveness, but also achieves generation efficiency that surpasses that of state-of-the-art (SOTA) methods in inference speed and memory utilization.
  • Figure 2: Illustration of model architecture of FlowerDance at different levels.
  • Figure 3: Qualitative Analysis on a typical Eastern Folk music clip.
  • Figure 4: Given start (yellow) and end (red) motions, FlowerDance generates smooth in-between motions with a soft mask (green), whereas a hard mask (blue) causes teleportation artifacts.