SURF：保留签名的快速视频生成
SURF: Signature-Retained Fast Video Generation
提出免训练的 noise reshifting：先在原分辨率做前几步去噪再切到低分辨率，避免高分辨率模型在低分辨率推理时的签名损失；配合预览-精修两阶段与窗口移位 Refiner，可作为插件套到任意视频扩散基座与加速方法上，做高分辨率视频生成的即插即用加速。
原文依据 we introduce noise reshifting, a training-free technique that mitigates this issue by conducting initial denoising steps on the original resolution and switching to low resolution in later steps
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论高分辨率视频生成的需求正在快速增长。然而，生成分辨率受到缓慢推理速度的严重制约。例如，Wan2.1 生成单个 720p 视频需要超过 50 分钟。尽管此前的工作从各个方面探索了视频生成的加速，但大多数都牺牲了原模型特有的签名（如布局、语义、运动）。在本工作中，我们提出 SURF，一个用于生成高分辨率视频的高效框架，同时最大限度地保留这些签名。具体而言，SURF 将视频生成分为两个阶段：首先，我们利用预训练模型在最优分辨率下进行推理，并对潜变量进行下采样，以快速生成低分辨率预览；然后，我们设计一个精修器（Refiner）对预览进行上采样。在预览阶段，我们发现直接在较低分辨率上推理一个（以更高分辨率训练的）模型会导致签名的严重损失。因此我们引入噪声重移（noise reshifting），这是一种免训练技术，通过在原始分辨率上执行初始去噪步骤、并在后续步骤切换到低分辨率来缓解该问题。在精修阶段，我们在预览与高分辨率目标之间建立映射关系，从而显著减少去噪步数。我们进一步整合移位窗口并精心设计训练范式，以获得强大且高效的精修器。通过这种方式，SURF 能够在高效生成高分辨率视频的同时，最大限度地接近给定预训练模型的签名。SURF 在概念上简单，可作为插件与各种基座模型和加速方法兼容。例如，它为生成 5 秒、16fps、720p 的 Wan 2.1 视频实现 12.5 倍加速，为生成 5 秒、24fps、720p 的 HunyuanVideo 视频实现 8.7 倍加速。
英文原文摘要 1873 字符
The demand for high-resolution video generation is growing rapidly. However, the generation resolution is severely constrained by slow inference speeds. For instance, Wan2.1 requires over 50 minutes to generate a single 720p video. While previous works explore accelerating video generation from various aspects, most of them compromise the distinctive signatures (e.g., layout, semantic, motion) of the original model. In this work, we propose **SURF**, an efficient framework for generating high-resolution videos, while maximally keeping the signatures. Specifically, SURF divides video generation into two stages: First, we leverage the pretrained model to infer at optimal resolution and downsample latent to generate low-resolution previews in fast speed; then we design a Refiner to upscale the preview. In the preview stage, we identify that directly inferring a model (trained with higher resolution) on lower resolution causes severe losses in signatures. So we introduce noise reshifting, a training-free technique that mitigates this issue by conducting initial denoising steps on the original resolution and switching to low resolution in later steps. In the refine stage, we establish a mapping relationship between the preview and the high-resolution target, which significantly reduces the denoising steps. We further integrate shifting windows and carefully design the training paradigm to get a powerful and efficient Refiner. In this way, SURF enables generating high-resolution videos efficiently while maximally closer to the signatures of the given pretrained model. SURF is conceptually simple and could serve as a plug-in that is compatible with various base model and acceleration methods. For example, it achieves 12.5x speedup for generating 5-second, 16fps, 720p Wan 2.1 videos and 8.7x speedup for generating 5-second, 24fps, 720p HunyuanVideo.