SAGA: Stable Acceleration Guidance for Autoregressive Video Generation

2026-07-08

AI Summary

This work addresses the issue of temporal instability in autoregressive video generation, where reusing generated latent variables amplifies temporal errors, leading to flickering, jittering, and structural drift. The study introduces, for the first time, spectral analysis in the acceleration domain to this task and proposes a training-free stabilization mechanism. By detecting high-frequency perturbations through discrete latent acceleration signals, the method combines a Slepian projection–based spectral guidance objective in the acceleration domain with a structured autoregressive noise initialization strategy. This approach suppresses short-range temporal correlations while preserving long-range motion structure. Without modifying or retraining the backbone model, it consistently enhances temporal stability across multiple autoregressive diffusion models—e.g., improving Self-Forcing’s Temporal Quality from 97.30 to 97.91—while maintaining visual fidelity, as confirmed by human preference studies.

Abstract

Autoregressive video diffusion enables efficient streaming and long-horizon video generation, but repeatedly reusing generated latents as causal context can amplify temporal errors, resulting in flickering, motion jitter, and structural drift. In this paper, we investigate this failure mode from a spectral kinematic perspective and identify discrete latent acceleration as an effective signal for revealing unstable high-frequency temporal perturbations. To this end, we propose SAGA, a training-free \textbf{\textit{s}}table \textbf{\textit{a}}cceleration \textbf{\textit{g}}uidance approach for \textbf{\textit{a}}utoregressive video generation. SAGA integrates an acceleration domain spectral guidance objective based on finite-window Slepian projections with a structured autoregressive noise initialization strategy that suppresses short-range temporal correlations while preserving long-range motion structure. Without retraining or modifying the backbone, SAGA can be directly applied to existing chunk-wise autoregressive diffusion models, which is the prevalent setting for high-quality generation. Extensive experiments show that SAGA consistently improves temporal quality across multiple autoregressive diffusion models. On Self-Forcing, SAGA improves Temporal Quality from 97.30 to 97.91 and Image Quality from 69.60 to 70.51. Moreover, spectral analysis and human preference studies demonstrate that SAGA reduces temporal instability while maintaining visual fidelity.

Problem

• autoregressive video generation
• temporal instability
• flickering
• motion jitter
• structural drift

Innovation

• autoregressive video generation
• temporal stability
• spectral guidance
• acceleration signal
• training-free