Carview!

Interactive Motion Editing

MotionReFit supports interactive motion modification by using previously edited motions as input for subsequent processes. Users can build upon earlier edits by feeding the modified motion back into the model with new prompts.

Results on Body Part Replacement

Results on Style Transfer

Results on Fine-grained Adjustment

Abstract

Text-guided motion editing enables high-level semantic control and iterative modifications beyond traditional keyframe animation. Existing methods rely on limited pre-collected training triplets (original motion, edited motion, and instruction), which severely hinders their versatility in diverse editing scenarios. We introduce MotionCutMix, an online data augmentation technique that dynamically generates training triplets by blending body part motions based on input text. While MotionCutMix effectively expands the training distribution, the compositional nature introduces increased randomness and potential body part incoordination. To model such a rich distribution, we present MotionReFit, an auto-regressive diffusion model with a motion coordinator. The auto-regressive architecture facilitates learning by decomposing long sequences, while the motion coordinator mitigates the artifacts of motion composition. Our method handles both spatial and temporal motion edits directly from high-level human instructions, without relying on additional specifications or Large Language Models (LLMs). Through extensive experiments, we show that MotionReFit achieves state-of-the-art performance in text-guided motion editing. Ablation studies further verify that MotionCutMix significantly improves the model's generalizability while maintaining training convergence.

Method Overview

MotionRefit

Research Poster — MotionRefit processes the original motion through sliding windows, where body keypoints are encoded for input to a transformer-based motion diffusion model. To ensure motion continuity, noise is applied starting from the third frame while preserving the first two frames. The model incorporates an additional token integrating the editing instruction, diffusion step, and progress indicator. The generated keypoints undergo SMPL-X optimization and merging to create the final edited motion. To enhance body part coordination, we employ a discriminator trained to identify motion segments composed of multiple source motions, which guides the denoising process through classifier guidance.

Spatial Motion Blending

The STANCE Dataset

BibTex

@article{jiang2025dynamic,
  title={Dynamic Motion Blending for Versatile Motion Editing},
  author={Jiang, Nan and Li, Hongjie and Yuan, Ziye and He, Zimo and Chen, Yixin and Liu, Tengyu and Zhu, Yixin and Huang, Siyuan},
  journal={arXiv preprint arXiv:2503.20724},
  year={2025}
}

Dynamic Motion Blending for Versatile Motion Editing

Overview Video