Saved in:
Bibliographic Details
Main Authors: Cui, Hanshuai, Tang, Zhiqing, Yao, Zhi, Meng, Fanshuai, Jia, Weijia, Zhao, Wei
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.02979
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917382865289216
author Cui, Hanshuai
Tang, Zhiqing
Yao, Zhi
Meng, Fanshuai
Jia, Weijia
Zhao, Wei
author_facet Cui, Hanshuai
Tang, Zhiqing
Yao, Zhi
Meng, Fanshuai
Jia, Weijia
Zhao, Wei
contents Autoregressive (AR) video diffusion models enable long-form video generation but remain expensive due to repeated multi-step denoising. Existing training-free acceleration methods rely on binary cache-or-recompute decisions, overlooking intermediate cases where direct reuse is too coarse yet full recomputation is unnecessary. Moreover, asynchronous AR schedules assign different noise levels to co-generated frames, yet existing methods process the entire valid interval uniformly. To address these AR-specific inefficiencies, we present SCOPE, a training-free framework for efficient AR video diffusion. SCOPE introduces a tri-modal scheduler over cache, predict, and recompute, where prediction via noise-level Taylor extrapolation fills the gap between reuse and recomputation with explicit stability controls backed by error propagation analysis. It further introduces selective computation that restricts execution to the active frame interval. On MAGI-1 and SkyReels-V2, SCOPE achieves up to 4.73x speedup while maintaining quality comparable to the original output, outperforming all training-free baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2604_02979
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Not All Frames Deserve Full Computation: Accelerating Autoregressive Video Generation via Selective Computation and Predictive Extrapolation
Cui, Hanshuai
Tang, Zhiqing
Yao, Zhi
Meng, Fanshuai
Jia, Weijia
Zhao, Wei
Computer Vision and Pattern Recognition
Autoregressive (AR) video diffusion models enable long-form video generation but remain expensive due to repeated multi-step denoising. Existing training-free acceleration methods rely on binary cache-or-recompute decisions, overlooking intermediate cases where direct reuse is too coarse yet full recomputation is unnecessary. Moreover, asynchronous AR schedules assign different noise levels to co-generated frames, yet existing methods process the entire valid interval uniformly. To address these AR-specific inefficiencies, we present SCOPE, a training-free framework for efficient AR video diffusion. SCOPE introduces a tri-modal scheduler over cache, predict, and recompute, where prediction via noise-level Taylor extrapolation fills the gap between reuse and recomputation with explicit stability controls backed by error propagation analysis. It further introduces selective computation that restricts execution to the active frame interval. On MAGI-1 and SkyReels-V2, SCOPE achieves up to 4.73x speedup while maintaining quality comparable to the original output, outperforming all training-free baselines.
title Not All Frames Deserve Full Computation: Accelerating Autoregressive Video Generation via Selective Computation and Predictive Extrapolation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.02979