Ms. Forcing: Efficient Streaming Video Generation with Multi-Scale Patchification and Attention

2026-07-23

AI Summary

This work addresses the challenges of real-time deployment in conventional streaming video generation methods, which suffer from nested autoregressive and denoising loops as well as spatial redundancy in multi-frame joint denoising. To overcome these limitations, the authors propose an efficient streaming video generation paradigm that dynamically adjusts spatial granularity via noise-aware multi-scale patching (MSP), employing coarse-grained image patches for high-noise frames. They further introduce a multi-scale self-attention (MSSA) mechanism to reduce computational overhead and integrate a same-noise-level DMD (H-DMD) strategy to enhance training-inference consistency. Built upon a fixed computation graph and a rolling-window denoising architecture, the method achieves 22.84 FPS on a single H200 GPU—39.6% faster than Rolling Forcing—and demonstrates significant improvements in both short- and long-form video generation quality on the VBench benchmark.

Abstract

Streaming video diffusion models have made substantial progress toward interactive and dynamic world simulation, but the nested autoregressive and denoising loops of conventional next-frame generation hinder real-time deployment. Recent rolling-window methods pipeline denoising across multiple consecutive frames at different noise levels, improving throughput and long-horizon stability. However, they tokenize every state at the same fine spatial granularity, leaving substantial noise-dependent redundancy in the joint denoising window. We propose Ms.Forcing, an efficient streaming video generation paradigm that adapts spatial granularity to each state's noise level. Its Multi-Scale Patchification (MSP) assigns coarser patches to noisier states, reducing the active-window token count by 45%, while Multi-Scale Self-Attention (MSSA) matches the density of visible non-sink keys and values to each query scale to further reduce attention cost. Because both schedules are fixed by window position, Ms.Forcing retains a static, hardware-friendly computation graph. We further introduce Homogeneous-Noise-Level DMD (H-DMD), which assembles each fake video from clean predictions sharing the same source noise level, thereby reducing the mismatch between DMD training sequences and inference-time rollouts. The multi-scale design helps offset the additional training cost of backpropagating through overlapping windows. We include both quantitative and qualitative experiments to show that Ms.Forcing reaches 22.84 FPS on a single H200 GPU, 39.6% faster than Rolling Forcing, while significantly improving VBench scores in both short video and long video generation setting.

Problem

• streaming video generation
• real-time deployment
• noise-dependent redundancy
• spatial granularity
• computational efficiency

Innovation

• Multi-Scale Patchification
• Multi-Scale Self-Attention
• Streaming Video Generation
• Homogeneous-Noise-Level DMD
• Efficient Diffusion Model