📺

SeedVR2 3B - Video Upscaling

🛠

AI-Powered Video Upscaling & Quality Restoration

Most AI video restoration tools chew through many processing passes to clean up a single clip, which makes them slow and expensive to run at scale. SeedVR2-3B does it in a single pass — feed in blurry, compressed, or low-resolution footage and it reconstructs sharp, high-resolution detail without the repeated denoising. Built by ByteDance Seed.

📹 See it in action

image

What it does

SeedVR2-3B is a video restoration and enhancement model developed by ByteDance Seed. It takes damaged, blurry, compressed, or low-resolution video footage and restores it to sharp, high-quality output. What sets SeedVR2 apart from previous AI video restoration tools is its speed: where most comparable tools require many processing passes to produce a good result, SeedVR2 achieves the same or better quality in a single pass. This makes it significantly faster and more cost-effective to run at scale, without meaningful sacrifice in output quality.

Problem it solves

  • Old or degraded footage – Restores heavily compressed, blurry, noisy, or low-resolution videos to sharp, high-quality output
  • Slow restoration pipelines – Most AI video restoration tools are slow and computationally expensive. SeedVR2 dramatically reduces processing time by completing restoration in a single step instead of many
  • High-resolution output – Handles high-resolution video restoration without the visual inconsistencies that commonly appear in other tools when scaling up
  • Real-world content – Designed to work on real-world footage, not just clean lab test videos

Input/Output

  • Input:
    • A degraded, low-quality, blurry, compressed, or low-resolution video
  • Parameters:
    • Target resolution (dropdown, default 720p)
      • 720p — HD (1280 × 720). Fastest option
      • 1080p — Full HD (1920 × 1080). Common target for social / web / production preview
      • 1440p — QHD (2560 × 1440). Higher-detail restoration for larger displays
      • 2160p (4K) — Ultra HD (3840 × 2160). Highest-quality target; expect longer processing time at this setting
    • Random seed (-1 = random) — set a fixed integer to reproduce the same restoration across runs
  • Output: A restored, high-resolution MP4 video file
    • Visually sharp with reconstructed details and textures where they were lost
    • Smooth, coherent motion throughout
    • Resolution matches the selected Target resolution
    • image

Accuracy & Speed

  • Speed & Scale
Processing speed
Single-pass restoration (significantly faster than multi-step alternatives)
Infrastructure needed
Standard single GPU setup
Output quality
Comparable to or better than existing multi-step restoration tools

Model Variants

SeedVR2 is available in two sizes to suit different needs:

Version
Best For
SeedVR2-3B
Faster processing, lower resource requirements
SeedVR2-7B
Higher quality output for more demanding use cases

Known Limitations

  • May struggle with very heavily degraded footage or scenes with very fast, large movements
  • On video that is already high quality (e.g., lightly compressed AI-generated content), it can occasionally over-sharpen, adding unnecessary detail
  • These are described by the team as prototype models, and real-world performance may vary
  • Technical Details (for engineers)

Model Source

Compliance & Provenance

Provider
Open-source (ByteDance)
Provider type
Specialized
License
apache-2.0 (ByteDance)
EU AI Act risk class
Minimal Risk
Art. 50 transparency
Not applicable
Region availability
Available globally
Training data summary
Pending — provider has not yet published per Art. 53(d)

For more on how we classify models and mark outputs, see our AI Policy.

Related Video Upscaling Models

  • SparkVSR — Reference-Guided Upscaling for Short Clips
  • FlashVSR-v1.1 — Real-Time 4× Streaming Video Upscaling
  • SeedVR2 7B — Upscale & Restore Video up to 4K, highest-quality variant

New to CNAPS Studio? Start with the Quickstart Guide, or browse the full AI Model Hub.

Run it now