Saved in:
Bibliographic Details
Main Authors: Li, Xuan, Ma, Qianli, Lin, Tsung-Yi, Chen, Yongxin, Jiang, Chenfanfu, Liu, Ming-Yu, Xiang, Donglai
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.01204
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917973838528512
author Li, Xuan
Ma, Qianli
Lin, Tsung-Yi
Chen, Yongxin
Jiang, Chenfanfu
Liu, Ming-Yu
Xiang, Donglai
author_facet Li, Xuan
Ma, Qianli
Lin, Tsung-Yi
Chen, Yongxin
Jiang, Chenfanfu
Liu, Ming-Yu
Xiang, Donglai
contents We present Articulated Kinematics Distillation (AKD), a framework for generating high-fidelity character animations by merging the strengths of skeleton-based animation and modern generative models. AKD uses a skeleton-based representation for rigged 3D assets, drastically reducing the Degrees of Freedom (DoFs) by focusing on joint-level control, which allows for efficient, consistent motion synthesis. Through Score Distillation Sampling (SDS) with pre-trained video diffusion models, AKD distills complex, articulated motions while maintaining structural integrity, overcoming challenges faced by 4D neural deformation fields in preserving shape consistency. This approach is naturally compatible with physics-based simulation, ensuring physically plausible interactions. Experiments show that AKD achieves superior 3D consistency and motion quality compared with existing works on text-to-4D generation. Project page: https://research.nvidia.com/labs/dir/akd/
format Preprint
id arxiv_https___arxiv_org_abs_2504_01204
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Articulated Kinematics Distillation from Video Diffusion Models
Li, Xuan
Ma, Qianli
Lin, Tsung-Yi
Chen, Yongxin
Jiang, Chenfanfu
Liu, Ming-Yu
Xiang, Donglai
Graphics
Computer Vision and Pattern Recognition
We present Articulated Kinematics Distillation (AKD), a framework for generating high-fidelity character animations by merging the strengths of skeleton-based animation and modern generative models. AKD uses a skeleton-based representation for rigged 3D assets, drastically reducing the Degrees of Freedom (DoFs) by focusing on joint-level control, which allows for efficient, consistent motion synthesis. Through Score Distillation Sampling (SDS) with pre-trained video diffusion models, AKD distills complex, articulated motions while maintaining structural integrity, overcoming challenges faced by 4D neural deformation fields in preserving shape consistency. This approach is naturally compatible with physics-based simulation, ensuring physically plausible interactions. Experiments show that AKD achieves superior 3D consistency and motion quality compared with existing works on text-to-4D generation. Project page: https://research.nvidia.com/labs/dir/akd/
title Articulated Kinematics Distillation from Video Diffusion Models
topic Graphics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.01204