Harmony: Harmonizing Audio and Video Generation through Cross-Task Synergy

TL;DR

Harmony introduces a comprehensive framework that significantly improves synchronized audio-visual content generation by addressing core challenges in joint diffusion processes through innovative training, alignment modules, and guidance techniques.

Contribution

The paper presents a novel framework with cross-task training, a global-local alignment module, and a synchronization-enhanced guidance method to improve audio-visual synchronization in generative models.

Findings

Achieves state-of-the-art synchronization accuracy.

Outperforms existing methods in generation fidelity.

Effectively mitigates correspondence drift and improves temporal alignment.

Abstract

The synthesis of synchronized audio-visual content is a key challenge in generative AI, with open-source models facing challenges in robust audio-video alignment. Our analysis reveals that this issue is rooted in three fundamental challenges of the joint diffusion process: (1) Correspondence Drift, where concurrently evolving noisy latents impede stable learning of alignment; (2) inefficient global attention mechanisms that fail to capture fine-grained temporal cues; and (3) the intra-modal bias of conventional Classifier-Free Guidance (CFG), which enhances conditionality but not cross-modal synchronization. To overcome these challenges, we introduce Harmony, a novel framework that mechanistically enforces audio-visual synchronization. We first propose a Cross-Task Synergy training paradigm to mitigate drift by leveraging strong supervisory signals from audio-driven video and video-driven audio generation tasks. Then, we design a Global-Local Decoupled Interaction Module for efficient and precise temporal-style alignment. Finally, we present a novel Synchronization-Enhanced CFG (SyncCFG) that explicitly isolates and amplifies the alignment signal during inference. Extensive experiments demonstrate that Harmony establishes a new state-of-the-art, significantly outperforming existing methods in both generation fidelity and, critically, in achieving fine-grained audio-visual synchronization.