ECCV 2026

FlexAM: Flexible Appearance–Motion Decomposition for Versatile Video Generation Control

Mingzhi Sheng^1,*Zekai Gu^2,*Peng Li² Cheng Lin³ Hao-Xiang Guo⁴ Ying-Cong Chen^1,2,†Yuan Liu^2,†

¹ HKUST (Guangzhou)² HKUST³ Macau University of Science and Technology⁴ Tsinghua University

* Equal contribution † Corresponding authors

Overview

Appearance and motion, controlled separately

Abstract

Effective and generalizable control in video generation remains a significant challenge. While many methods rely on ambiguous or task-specific signals, we argue that a fundamental disentanglement of "appearance" and "motion" provides a more robust and scalable pathway. We propose FlexAM, a unified framework built upon a novel 3D control signal. This signal represents video dynamics as a point cloud, introducing three key enhancements: multi-frequency positional encoding to distinguish fine-grained motion, depth-aware encoding, and a flexible control signal for balancing precision and generalization. This representation allows FlexAM to effectively disentangle appearance and motion, enabling a wide range of tasks including I2V/V2V editing, camera control, and spatial object editing. Extensive experiments demonstrate that FlexAM achieves superior performance across all evaluated tasks.

FlexAM teaser illustrating masked appearance editing, sparse guidance, multi-frequency encoding, and depth-aware encoding

Method

One representation, different controls

FlexAM pipeline — Appearance and motion are encoded separately, then injected into the video backbone through two control branches.

Masked appearance

Masked video provides a unified appearance condition for flexible I2V and V2V editing beyond first-frame conditioning.

3D-aware motion

An 18-channel motion video represents 3D point trajectories with identity, multi-frequency positional, and depth-aware encodings.

Sparse-to-dense control

Varying point densities enable FlexAM to balance control precision and generalization across different tasks.

Results

Versatile control with one model

Reference

Citation

@misc{sheng2026flexam,
  title={FlexAM: Flexible Appearance-Motion Decomposition for
         Versatile Video Generation Control},
  author={Sheng, Mingzhi and Gu, Zekai and Li, Peng and
          Lin, Cheng and Guo, Hao-Xiang and Chen, Ying-Cong and Liu, Yuan},
  year={2026},
  eprint={2602.13185},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}