OmniForcing: Unleashing Real-time Joint Audio-Visual Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Su, Yaofeng, Li, Yuming, Xue, Zeyue, Huang, Jie, Fu, Siming, Li, Haoran, Li, Ying, Qian, Zezhong, Huang, Haoyang, Duan, Nan
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911511296868352
author Su, Yaofeng
Li, Yuming
Xue, Zeyue
Huang, Jie
Fu, Siming
Li, Haoran
Li, Ying
Qian, Zezhong
Huang, Haoyang
Duan, Nan
author_facet Su, Yaofeng
Li, Yuming
Xue, Zeyue
Huang, Jie
Fu, Siming
Li, Haoran
Li, Ying
Qian, Zezhong
Huang, Haoyang
Duan, Nan
contents Recent joint audio-visual diffusion models achieve remarkable generation quality but suffer from high latency due to their bidirectional attention dependencies, hindering real-time applications. We propose OmniForcing, the first framework to distill an offline, dual-stream bidirectional diffusion model into a high-fidelity streaming autoregressive generator. However, naively applying causal distillation to such dual-stream architectures triggers severe training instability, due to the extreme temporal asymmetry between modalities and the resulting token sparsity. We address the inherent information density gap by introducing an Asymmetric Block-Causal Alignment with a zero-truncation Global Prefix that prevents multi-modal synchronization drift. The gradient explosion caused by extreme audio token sparsity during the causal shift is further resolved through an Audio Sink Token mechanism equipped with an Identity RoPE constraint. Finally, a Joint Self-Forcing Distillation paradigm enables the model to dynamically self-correct cumulative cross-modal errors from exposure bias during long rollouts. Empowered by a modality-independent rolling KV-cache inference scheme, OmniForcing achieves state-of-the-art streaming generation at $\sim$25 FPS on a single GPU, maintaining multi-modal synchronization and visual quality on par with the bidirectional teacher.\textbf{Project Page:} \href{https://omniforcing.com}{https://omniforcing.com}
format Preprint
id arxiv_https___arxiv_org_abs_2603_11647
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle OmniForcing: Unleashing Real-time Joint Audio-Visual Generation
Su, Yaofeng
Li, Yuming
Xue, Zeyue
Huang, Jie
Fu, Siming
Li, Haoran
Li, Ying
Qian, Zezhong
Huang, Haoyang
Duan, Nan
Multimedia
Computer Vision and Pattern Recognition
Sound
Recent joint audio-visual diffusion models achieve remarkable generation quality but suffer from high latency due to their bidirectional attention dependencies, hindering real-time applications. We propose OmniForcing, the first framework to distill an offline, dual-stream bidirectional diffusion model into a high-fidelity streaming autoregressive generator. However, naively applying causal distillation to such dual-stream architectures triggers severe training instability, due to the extreme temporal asymmetry between modalities and the resulting token sparsity. We address the inherent information density gap by introducing an Asymmetric Block-Causal Alignment with a zero-truncation Global Prefix that prevents multi-modal synchronization drift. The gradient explosion caused by extreme audio token sparsity during the causal shift is further resolved through an Audio Sink Token mechanism equipped with an Identity RoPE constraint. Finally, a Joint Self-Forcing Distillation paradigm enables the model to dynamically self-correct cumulative cross-modal errors from exposure bias during long rollouts. Empowered by a modality-independent rolling KV-cache inference scheme, OmniForcing achieves state-of-the-art streaming generation at $\sim$25 FPS on a single GPU, maintaining multi-modal synchronization and visual quality on par with the bidirectional teacher.\textbf{Project Page:} \href{https://omniforcing.com}{https://omniforcing.com}
title OmniForcing: Unleashing Real-time Joint Audio-Visual Generation
topic Multimedia
Computer Vision and Pattern Recognition
Sound
url https://arxiv.org/abs/2603.11647