<aside> 🛠

Real-Time Diffusion-Based Streaming Video Super-Resolution

Diffusion-based video upscalers produce beautiful results, but they are usually far too slow to use on real footage — many denoising passes per frame add up fast. FlashVSR-v1.1 collapses that into a single step and streams frames as it goes, so diffusion-quality upscaling becomes practical at scale, even at ultra-high resolutions. It is the first diffusion-based one-step streaming VSR framework, built by researchers at Tsinghua University / OpenImagingLab.

🏆 Up to ~17 FPS at 768×1408 on a single A100 — and up to ~12× faster than prior one-step diffusion VSR models

</aside>

📹 See it in action

What it does

FlashVSR-v1.1 is a diffusion-based video super-resolution (VSR) model that takes low-resolution video and reconstructs it at significantly higher resolution (optimized for 4× upscaling). What sets FlashVSR apart from other diffusion-based VSR tools is its speed: instead of running many denoising passes per frame, it produces results in a single step and processes frames in a streaming fashion. This makes diffusion-quality video super-resolution practical at scale, even at ultra-high output resolutions.

Problem it solves

• Low-resolution video sources – Upscales legacy, web-compressed, or low-bitrate footage to sharp, high-resolution output (4× recommended)
• Slow diffusion VSR pipelines – Most diffusion-based super-resolution tools are far too slow for practical use on video. FlashVSR completes super-resolution in a single inference step, achieving up to ~12× speedup over prior one-step diffusion VSR models
• Streaming / real-time use cases – Built as a streaming framework, enabling near real-time processing rather than batch-only workflows
• Ultra-high resolutions – Scales reliably to very large output resolutions where competing tools tend to break down or introduce artifacts
• Stability across aspect ratios and content types – v1.1 specifically targets improved stability and fidelity over the initial v1 release

Input/Output

• Input:
  • A low-resolution video (optimized for 4× super-resolution input)
• Output: A high-resolution MP4 video file
  • Sharp, detailed reconstruction with preserved fine structures and textures
  • Temporally coherent motion across frames, processed in a streaming pass
• Parameters:
  • Scale — fixed at 4× (this build uses the FlashVSR Tiny pipeline optimized for 4× super-resolution, so the scale is not user-selectable)
  • Random seed (-1 = random, default 1) — set a fixed integer to reproduce the same result across runs
  • Advanced Options:
    • LCSA top-k ratio (range 1.0 – 4.0, default 2) — how aggressively the model's Locality-Constrained Sparse Attention selects tokens per window. Higher = more tokens attended to (potentially better quality, more compute); lower = sparser attention (faster, may lose some fine detail)
    • LCSA window (range 5 – 15, odd values only, default 11) — the spatial window size used by LCSA. Larger windows capture more surrounding context per token (better structural coherence for large objects and long motion); smaller windows are faster and focus on local detail. Only odd numbers are valid so the window has a clean center pixel
  • Input resolution — up to 720 × 720 (per the studio node's Tiny pipeline)

A sample workflow from community

Accuracy & Speed