DISPLAY: Directable Human-Object Interaction Video Generation via Sparse Motion Guidance and Multi-Task Auxiliary

TL;DR

DISPLAY is a novel framework for controllable human-object interaction video generation that uses sparse motion guidance and multi-task training to produce high-fidelity, flexible videos with minimal control signals.

Contribution

We introduce a lightweight guidance method and multi-task training strategy that enable controllable and physically consistent HOI video generation with sparse inputs.

Findings

Achieves high-fidelity HOI video generation across diverse tasks.

Enables intuitive user control with minimal guidance signals.

Improves object robustness with Object-Stressed Attention.

Abstract

Human-centric video generation has advanced rapidly, yet existing methods struggle to produce controllable and physically consistent Human-Object Interaction (HOI) videos. Existing works rely on dense control signals, template videos, or carefully crafted text prompts, which limit flexibility and generalization to novel objects. We introduce a framework, namely DISPLAY, guided by Sparse Motion Guidance, composed only of wrist joint coordinates and a shape-agnostic object bounding box. This lightweight guidance alleviates the imbalance between human and object representations and enables intuitive user control. To enhance fidelity under such sparse conditions, we propose an Object-Stressed Attention mechanism that improves object robustness. To address the scarcity of high-quality HOI data, we further develop a Multi-Task Auxiliary Training strategy with a dedicated data curation pipeline, allowing the model to benefit from both reliable HOI samples and auxiliary tasks. Comprehensive experiments show that our method achieves high-fidelity, controllable HOI generation across diverse tasks. The project page can be found at \href{https://mumuwei.github.io/DISPLAY/}.