Odoriko: A Shape-Aware Multimodal Diffusion Framework for Human Motion

ECCV 2026
1Sony Group Corporation, 2Sony AI
Teaser

Abstract

Human motion generation has been widely studied across diverse input modalities — text, music, and video — and recent efforts have unified these into single multimodal frameworks. However, while morphological factors such as gender and body shape are known to produce distinct kinematic signatures, no existing unified framework incorporates this into generation, treating all subjects as morphologically equivalent.

We present Odoriko, the first unified multimodal motion generation framework that reflects subject bio-morphological information directly in synthesized motion output. Rather than averaging over subject variation, Odoriko generates motion that is consistent with who is moving, not just what they are asked to do — across text, music, and video conditions within a single model. When explicit morphological information is unavailable, Odoriko additionally recovers subject morphology alongside motion, unifying estimation and generation in one framework.

Extensive experiments across text-to-motion, music-to-dance, and video-to-motion benchmarks demonstrate that Odoriko matches or exceeds prior specialized models on standard metrics, while enabling morphology-consistent generation that no existing unified framework supports.

Human Motion Synthesis

Odoriko enables human motion generation and estimation from diverse input modalities—including text, music, and video—within a single unified framework.

Text-to-Motion

We compare Odoriko with existing multimodal motion generation models, MotionCraft and GENMO.

Music-to-Dance

Video-to-Motion

As our model focuses on local pose estimation, global motion is reconstructed using ground-truth translation. The subjects' shape and gender are directly predicted by Odoriko.

Shape-aware Motion Generation

Odoriko generates motions that faithfully follow the input shape conditioning. Even with identical text or music inputs, the resulting motions vary according to the subjects' shape, producing biomorphologically plausible and consistent movements. During inference, we fix the random seed so that differences arise only from the shape conditioning.

Text-to-Motion

Music-to-Dance

BibTeX

@inproceedings{shim2026odoriko,
  title     = {Odoriko: A Shape-Aware Multimodal Diffusion Framework for Human Motion},
  author    = {Shim, Dongseok and Tanke, Julian and Uchida, Kengo and Simon, Christian
               and Saito, Koichi and Shibuya, Takashi and Takahashi, Shusuke and Mitsufuji, Yuki},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}