Skip to content
Open access

Dance arrangement and style transfer methods driven by diffusion models and spatiotemporal graph convolutional networks

Sep 2026 · Discover Artificial Intelligence · Vol 6 · 0 citations · 21 references
Computer Science

TL;DR

A multi-task learning framework based on diffusion models and spatio-temporal graph convolutional networks is proposed, which has the potential to approach the level of professional choreography in terms of artistic expressiveness and technical indicators.

Abstract

As the problems of traditional dance choreography methods, such as high dependence on manual experience and limited creation efficiency, have become increasingly prominent, this study is committed to exploring new methods driven by artificial intelligence (AI) technology to improve the degree of choreography automation and artistic diversity. This study proposes a multi-task learning framework based on diffusion models and spatio-temporal graph convolutional networks. Through the framework’s strong sequence modeling and style decoupling capabilities, two core tasks, dance motion generation and style transfer, are realized. The study adopts a Transformer-based music-motion cross-modal encoder to achieve semantic alignment of input music features. Then, the study uses a conditional denoising diffusion probabilistic model to generate basic dance motion sequences, and finally realizes accurate control and conversion of different dance styles through a style encoder and an adaptive instance normalization module. Experimental results on the AIST + + dataset show that: the proposed method achieves a Fréchet Initial Distance (FID) value of 5.23 in terms of dance motion generation quality, the Dynamic Time Warping (DTW) index for music synchronization is reduced to 5.11, and the subjective evaluation score reaches 4.33. All indicators are significantly better than those of existing baseline models. Ablation experiments confirm the core role of the adaptive instance normalization module in style control, and the cross-style FID value in the style transfer task is stably around 6.0. This study provides an effective technical path for solving the problems of automation and stylization in dance choreography. Its generated results show the potential to approach the level of professional choreography in terms of artistic expressiveness and technical indicators, which is of great value for promoting the application of AI in the field of creative content generation.

Read PDF

Similar papers

Oct 2026

Dance Movement Style Transfer Based on 3D Pose Estimation and Spatiotemporal Feature Fusion

Experimental evaluations show that the proposed method can generate coherent target-style dance movements while maintaining the choreographic structure and motion characteristics of the original sequence and offers potential applications in digital choreography, motion synthesis, and creative performance analysis.

Xi-Wei-Li-Jia-Xing-Wang-Duo-Wei-Ya-Nan Wang · 0 citations
Book Open access Nov 2026

MATHDance: Mamba-Transformer Architecture with Uniform Tokenization for High-Quality 3D Dance Generation

A retrieval-based evaluation protocol, which leverages the intrinsic connection between retrieval and generation tasks and utilizes contrastive learning to encode cross-modal music-dance semantics, is introduced.

Kai-Xing Yang, Xu-Long Tang, Zi-Qiao Peng et al. · 0 citations
Open access Sep 2026

Replication as Pedagogy: The Social and Cultural Dimensions of Transmitting Dance Knowledge

Dance is an important intangible cultural heritage, the means by which cultural identity, artistic traditions, and embodied knowledge are transmitted from one generation to another. Movement replication has traditionally been a key technique for teaching dance in terms of which movement skills can be acquired through o...

Deng-Yu Zeng · 0 citations
#artificial intelligence Preprint Oct 2026

Self-Representation Alignment in Diffusion Models for Video Background Music Generation

Diffusion models have been widely used for various generative tasks. However, recent studies have demonstrated that the internal representations learned in the diffusion models'denoising process are insufficiently meaningful, which can hinder the generation performance. To address this, we propose a novel diffusion U-N...

Jin-Cheng Zhang, György Fazekas, C. Saitis · 0 citations
Oct 2026

Enhancing AI-Driven Sports Dance Therapy via Hybrid Dilated Convolution: A Motion Recognition and Evaluation Optimization Framework

This study addressed the need for motion process monitoring and real-time feedback in AI-assisted sports dance rehabilitation training. It proposed a multi-scale temporal evaluation framework incorporating Hybrid Dilated Convolution (HDC). Unlike traditional single-scale temporal models, this study addressed the diffic...

Rui-Ling Liu · 0 citations
Open access Sep 2026

Animation Storyboard Dynamic Layout and Rhythm Control Strategy Based on Optimized Reinforcement Learning

This research presents a Dynamic Shuffled Frog Leaping Fine-tuned Adaptive Soft Actor-Critic (DSFL-ASAC) model capable of dynamically adjusting frame composition (layout) and controlling temporal transitions (rhythm) between AI-assisted storyboard panels. A dataset comprising animation scripts, existing storyboards, an...

Meng-Jun Xie · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.