HyperSketch: Controllable Video Sketching in a Style Hyperspace

Xinding Zhu1 Xinye Yang1 Yingping Yang1 Mengjian Li2 Fei Gao1 Jiazhou Chen 1
1 Zhejiang University of Technology 2 Zhejiang Lab
Paper Code
teaser
Our method models the sketch as a point in a 4D style hyperspace, enabling continuous control of styles with temporal coherence.
The intermediate column shows input frames (top) and output sketches (middle), each of which corresponds to a point along a trajectory in this 4D style hyperspace (bottom).
Users can adjust temporal trajectories for fidelity, simplicity, and text guidance via editable curves (left).
The right panel presents additional discrete sketches with 4 other inputs, demonstrating the validity of HyperSketch across style combinations.

Overview

Abstract

Vector sketch animation offers tremendous advantages for multimedia and creative design through concise line expressions and flexible editing. Learning-based generation methods of sketch animation have made significant progress in the last decade, but still suffer from limited style diversity and controllability. This paper presents a controllable video sketching method that automatically converts videos into multi-style vector sketch animations. A continuous style hyperspace is constructed by multi-dimensional sketch styles (fidelity, simplicity, text guidance strength) and the timeline. With this hyperspace, stroke control points are parameterized as 4-variable Bernstein polynomials, ensuring smooth and differentiable style transitions. A multi-task, multi-stage optimization framework is designed to learn stroke hyperparameters accurately and efficiently. We further developed a web-based interactive interface that allows real-time style manipulation via editable curves. Experiments show the style controllability, high-quality, and user-friendliness of our method, which outperforms SOTA methods.

Framework

framework
Given an input video with tracked dense point trajectories, and a text prompt, we represent strokes with a four-variate Bernstein polynomial defined over a style hyperspace spanned by time 𝛼, fidelity 𝛽, simplicity 𝛾, and text guidance strength 𝛿 on the right. A progressive multi-stage optimization jointly learns the polynomial coefficients using five loss functions. The resulting representation enables real-time rendering of sketch animations with continuous, independent control of each style dimension.

Results

Citation

BibTeX

@article{zhu2026hypersketch,
  title={HyperSketch: Controllable Video Sketching in a Style Hyperspace},
  author={Zhu, Xinding and Yang, Xinye and Yang, Yingping and Li, Mengjian and Gao, Fei and Chen, Jiazhou},
  journal={arXiv preprint arXiv:2609.00919},
  year={2026}
}