Skip to content
SPECIFICTION
Get listed

SeedVR2

github.com

Free plan WindowsmacOSLinuxSelf-hosted
Visit github.com
Spec score 8.3 #11 of 32 in AI Video Upscalers

What it is

SeedVR2 is an open-source model for restoring video at arbitrary resolutions, with a one-step diffusion-transformer approach that does not add a separate diffusion prior. Its official ComfyUI release can upscale both video and images and provides four nodes to control the pipeline. A standalone command-line interface supports batch and multi-GPU workflows. Inputs may be RGB or RGBA images and video; the documented model choices are 3B and 7B, with FP16, FP8, and quantized GGUF options. Model offloading, VAE tiling, and BlockSwap are available to manage memory during inference. The research describes adaptive window attention, which adjusts window size to output resolution, and adversarial post-training for high-resolution restoration. Standalone installation instructions cover Windows, macOS, and Linux, including Apple Silicon MPS support. Input height and width must be multiples of 32, and the demo crops resized inputs to meet that constraint. SeedVR2 is licensed under Apache 2.0, and released inference code and model weights are linked from the repository. The plan is free.

Spec sheet

Purpose
SeedVR2 is a one-step diffusion-transformer model for video restoration at any resolution.Read 10 Oct 2026 · github.com
Security
A release note says model loading was restricted to tensor-only deserialization to prevent code execution from malicious .pth files.Read 4 Oct 2026 · github.com
Support
The repository invites questions by email at [email protected].Read 10 Oct 2026 · github.com
License
The project is released under the Apache License 2.0.Read 10 Oct 2026 · github.com
Adaptive attention
Its adaptive window attention adjusts window size to fit output resolutions.Read 10 Oct 2026 · github.com
Authors’ affiliations
The listed affiliations are S-Lab at Nanyang Technological University and ByteDance Seed.Read 10 Oct 2026 · iceclear.github.io
ComfyUI integration
The release provides four ComfyUI nodes for controlling the upscaling pipeline.Read 4 Oct 2026 · github.com
Downloads
The repository links to released inference code and model weights.Read 10 Oct 2026 · github.com
High-resolution results
The project page describes efficient single-step restoration at 1080p with faithful details.Read 10 Oct 2026 · iceclear.github.io
Image and video upscaling
The official ComfyUI release supports upscaling both videos and images.Read 4 Oct 2026 · github.com
Inference
The model restores video in a single step without relying on an additional diffusion prior.Read 10 Oct 2026 · github.com
Input constraint
Input height and width must be multiples of 32; the demo crops resized inputs to meet that constraint.Read 10 Oct 2026 · iceclear.github.io
Integration
The maintainers state that a ComfyUI integration was released in November 2025.Read 10 Oct 2026 · github.com
Model options
The release documents 3B and 7B model variants, with FP16, FP8, and quantized GGUF options.Read 4 Oct 2026 · github.com
Performance controls
The release documents model offloading, VAE tiling, and BlockSwap to manage memory use during inference.Read 4 Oct 2026 · github.com
Platform support
Standalone installation instructions cover Windows, macOS, and Linux; Apple Silicon MPS support is also documented.Read 4 Oct 2026 · github.com
Research
The project page identifies the work as ICLR 2026.Read 10 Oct 2026 · iceclear.github.io
Research affiliation
The paper lists the authors’ affiliations as S-Lab at Nanyang Technological University and ByteDance Seed.Read 4 Oct 2026 · iceclear.github.io
Research method
The paper describes adaptive window attention and adversarial post-training for high-resolution video restoration.Read 4 Oct 2026 · iceclear.github.io
Standalone use
The release also provides a standalone command-line interface, including multi-GPU and batch processing workflows.Read 4 Oct 2026 · github.com
Supported formats
The upscaler supports RGB and RGBA video and image inputs.Read 4 Oct 2026 · github.com
Training
The project describes adversarial video restoration training against real data and a feature matching loss.Read 10 Oct 2026 · github.com

Plans and prices

SeedVR2 Free No paid plans or usage limits stated on the project pages Read 10 Oct 2026 · github.com

Every SeedVR2 plan and price →

Strong points and limits

Strong points

  • Upscales both video and images through ComfyUI.
  • Command-line workflows include batch and multi-GPU processing.
  • Offers 3B and 7B models with several precision options.
  • Documents memory controls including offloading and VAE tiling.
  • Apache 2.0 licensed with released code and weights.

Limits

  • Input height and width must be multiples of 32.
  • Standalone use requires installation on Windows, macOS, or Linux.
  • No paid plans or usage limits are stated.

Who it suits

SeedVR2 may suit people who want to restore or upscale video and images through ComfyUI or a command line. Its documented platform support and model controls are relevant to users managing local inference workflows.

Verdict

SeedVR2 combines single-step restoration with ComfyUI nodes and a standalone command-line option. Check the multiple-of-32 input constraint when preparing source media.

Where Specifiction places it

Alternatives to SeedVR2All alternatives →