AnyAct: Towards Human Reenactment of Character Motion From Video

Liuhan Chen*1, Lei Zhong*2, Jiawei Wang3, Qing Shuai4, Li Yuan1, Leidong Fan5, Qing Li†5, Kanglin Liu5

1Peking University 2Nankai University 3The University of Hong Kong 4Zhejiang University 5Pengcheng Laboratory

arXiv 2026

What is AnyAct?

AnyAct reenacts human motion from arbitrary character videos. It converts sparse 2D motion cues from animals, cartoons, objects, and stylized characters into plausible 3D human motions that preserve the source action semantics. Given monocular videos of non-human characters with diverse topologies, AnyAct reinterprets their characteristic motion patterns as plausible human performances rather than reproducing their source structures literally. Shown here are reenactments of (a) kangaroo-like jumping, (b) butterfly-like wing flapping, and (c) beckoning-cat-like waving.

Kangaroo-like jumping

Kangaroo-like jumping

Butterfly-like wing flapping

Butterfly-like wing flapping

Beckoning-cat-like waving

Beckoning-cat-like waving

Method Overview

AnyAct method overview

Given reference videos of characters, AnyAct first extracts local sparse 2D joint trajectories as transferable motion cues from the input video using our model-ensemble-based Versatile Feature Extractor (VFE). These cues are then injected into a human motion generator (MoMask++) through the ControlNet-like 2D Local Adapter (2D-LA) to produce the initial human reenactments that follow the observed character dynamics.

More Results

Dancing with side-to-side swaying

Ghost dance reference video and generated human motion

Deer-like walking

Deer walk reference video and generated human motion

Penguin-like walking

Penguin walk reference video and generated human motion

Monkey-like walking

Monkey walk reference video and generated human motion

Dinosaur-like walking

Dinosaur walk reference video and generated human motion

Seal-like walking with side-to-side swaying

Seal walk reference video and generated human motion

Mechanical-spider-like in-place jumping

Monster jump reference video and generated human motion

Toy-robot-like walking with side-to-side swaying

Robot walk reference video and generated human motion

Trajectory Control and Intuitive Editing

Cat-like walking with forward and half-circle trajectory control

Cat-like walking with forward and half-circle trajectory control

Editing the height of human when perform kangaroo-like jumping

Editing the height of human when perform kangaroo-like jumping

Editing the arm spread of human when perform penguin-like walking

Editing the arm spread of human when perform penguin-like walking

Generalization to the reenactment of human characters.

Human character reenactment example
Human character reenactment example

Physical-Proxies-Based Reenactment (Comparsed to DancingBox)

Physical-proxies-based reenactment comparison example
Physical-proxies-based reenactment comparison example

Citation

@article{chen2026anyact,
  title={AnyAct: Towards Human Reenactment of Character Motion From Video},
  author={Chen, Liuhan and Zhong, Lei and Wang, Jiewei and Shuai, Qin and Yuan, Li and Fan, Leidong and Li, Qing and Liu, Kanglin},
  journal={arXiv preprint arXiv:2605.15497},
  year={2026}
}