UnityVideo: Unified Multi-Modal Multi-Task Learning for Enhancing World-Aware Video Generation

TL;DR

UnityVideo is a comprehensive multi-modal framework for world-aware video generation that integrates various data modalities and training paradigms to improve generalization, quality, and physical consistency in generated videos.

Contribution

It introduces a novel unified framework with dynamic noising and a modality switcher, along with a large-scale dataset, advancing multi-modal video generation capabilities.

Findings

Enhanced zero-shot generalization to unseen data

Improved video quality and consistency

Faster convergence through joint optimization

Abstract

Recent video generation models demonstrate impressive synthesis capabilities but remain limited by single-modality conditioning, constraining their holistic world understanding. This stems from insufficient cross-modal interaction and limited modal diversity for comprehensive world knowledge representation. To address these limitations, we introduce UnityVideo, a unified framework for world-aware video generation that jointly learns across multiple modalities (segmentation masks, human skeletons, DensePose, optical flow, and depth maps) and training paradigms. Our approach features two core components: (1) dynamic noising to unify heterogeneous training paradigms, and (2) a modality switcher with an in-context learner that enables unified processing via modular parameters and contextual learning. We contribute a large-scale unified dataset with 1.3M samples. Through joint optimization, UnityVideo accelerates convergence and significantly enhances zero-shot generalization to unseen data. We demonstrate that UnityVideo achieves superior video quality, consistency, and improved alignment with physical world constraints. Code and data can be found at: https://github.com/dvlab-research/UnityVideo