MoGA: Mixture-of-Groups Attention for End-to-End Long Video Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jia, Weinan, Lu, Yuning, Huang, Mengqi, Wang, Hualiang, Huang, Binyuan, Chen, Nan, Liu, Mu, Jiang, Jidong, Mao, Zhendong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911224371871744
author Jia, Weinan
Lu, Yuning
Huang, Mengqi
Wang, Hualiang
Huang, Binyuan
Chen, Nan
Liu, Mu
Jiang, Jidong
Mao, Zhendong
author_facet Jia, Weinan
Lu, Yuning
Huang, Mengqi
Wang, Hualiang
Huang, Binyuan
Chen, Nan
Liu, Mu
Jiang, Jidong
Mao, Zhendong
contents Long video generation with Diffusion Transformers (DiTs) is bottlenecked by the quadratic scaling of full attention with sequence length. Since attention is highly redundant, outputs are dominated by a small subset of query-key pairs. Existing sparse methods rely on blockwise coarse estimation, whose accuracy-efficiency trade-offs are constrained by block size. This paper introduces Mixture-of-Groups Attention (MoGA), an efficient sparse attention that uses a lightweight, learnable token router to precisely match tokens without blockwise estimation. Through semantic-aware routing, MoGA enables effective long-range interactions. As a kernel-free method, MoGA integrates seamlessly with modern attention stacks, including FlashAttention and sequence parallelism. Building on MoGA, we develop an efficient long video generation model that end-to-end produces minute-level, multi-shot, 480p videos at 24 fps, with a context length of approximately 580k. Comprehensive experiments on various video generation tasks validate the effectiveness of our approach.
format Preprint
id arxiv_https___arxiv_org_abs_2510_18692
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MoGA: Mixture-of-Groups Attention for End-to-End Long Video Generation
Jia, Weinan
Lu, Yuning
Huang, Mengqi
Wang, Hualiang
Huang, Binyuan
Chen, Nan
Liu, Mu
Jiang, Jidong
Mao, Zhendong
Computer Vision and Pattern Recognition
Long video generation with Diffusion Transformers (DiTs) is bottlenecked by the quadratic scaling of full attention with sequence length. Since attention is highly redundant, outputs are dominated by a small subset of query-key pairs. Existing sparse methods rely on blockwise coarse estimation, whose accuracy-efficiency trade-offs are constrained by block size. This paper introduces Mixture-of-Groups Attention (MoGA), an efficient sparse attention that uses a lightweight, learnable token router to precisely match tokens without blockwise estimation. Through semantic-aware routing, MoGA enables effective long-range interactions. As a kernel-free method, MoGA integrates seamlessly with modern attention stacks, including FlashAttention and sequence parallelism. Building on MoGA, we develop an efficient long video generation model that end-to-end produces minute-level, multi-shot, 480p videos at 24 fps, with a context length of approximately 580k. Comprehensive experiments on various video generation tasks validate the effectiveness of our approach.
title MoGA: Mixture-of-Groups Attention for End-to-End Long Video Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.18692