---
title: "FlashDecoder：实时潜在到像素流式视频解码器"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.14898v1"
permalink: "https://feed.astralor.com/items/itm_9f624e4fdce3348d3f3e7b78"
published: "2026-07-16T12:16:12.000Z"
score: 76
channel: "tech"
tags: ["FlashDecoder","Transformer","Wan2.1","Wan2.2","实时生成","视频解码"]
---

# FlashDecoder：实时潜在到像素流式视频解码器

## AI 点评

> 这篇论文最大的亮点是用纯 Transformer 替换了视频扩散模型中耗时的 3D 卷积解码器，实现了与卷积相当的画质和数倍的速度提升。对做视频生成、实时渲染和模型部署的工程师来说，这是个非常实用的优化方向。值得关注的是它的流式推理设计——不限视频长度、延迟恒定，这为直播级视频生成铺平了道路。

## 速览

FlashDecoder 是一种纯 Transformer 视频解码器，专为实时视频生成而设计。它逐帧将潜在表示解码为像素，每帧仅通过滚动 KV 缓存关注固定大小的历史窗口，从而在任意视频长度下保持恒定延迟和内存上限。在 Wan2.1 和 Wan2.2 潜在空间上，FlashDecoder 在 1080p 下达到 41.55dB PSNR，与卷积解码器相当，但解码速度提升 3.6-4.7 倍，内存减少 11 倍。结合架构感知优化，加速比可扩大至 12 倍。

## 关键点

- 逐帧流式解码，使用滚动 KV 缓存实现固定窗口注意力，解码延迟与视频长度无关。
- 在 1080p 下重建质量（PSNR 41.55dB）与卷积解码器（41.49dB）持平。
- 在单张 H100 GPU 上解码速度提升 3.6-4.7 倍，内存降低 11 倍。
- 架构感知推理优化可使加速比达到 12 倍。

## 全文

Real-time video generation demands fast decoding as much as fast denoising, yet current latent video diffusion models rely on 3D convolutional decoders that are slow and memory-intensive at high resolutions or for long video. We introduce FlashDecoder, a fast, memory-efficient pure-Transformer video decoder that decodes latents to pixels frame by frame. At each step, the current frame attends only to a fixed-size window of past frames through a rolling KV cache. The fixed temporal window keeps decoding fast and memory bounded regardless of video length, enabling constant-latency streaming. Because frames are processed sequentially, temporal causality is enforced without explicit attention masks, enabling training at resolutions up to 1080p and matching the reconstruction quality of convolutional decoders. On the Wan2.1 and Wan2.2 latent spaces, FlashDecoder matches each convolutional decoder in reconstruction quality (e.g., 41.55dB vs. 41.49dB PSNR at 1080p) while decoding 3.6x-4.7x faster with up to 11x less memory on a single H100 GPU. With architecture-aware inference optimizations, the speedup widens to 12x.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_9f624e4fdce3348d3f3e7b78