SeedVR2

SeedVR2

Upscales images and video in ComfyUI with repackaged SeedVR2 3B and 7B files from ByteDance Seed, in several precision formats. For ComfyUI users who run upscaling on their own machine.

Made by Comfy-Org on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: SeedVR2 here is a set of repackaged model files for ComfyUI, taken from ByteDance Seed's SeedVR2 3B and 7B models. The files come in several precision formats, and the 7B model also has a sharp variant. Comfy-Org's workflows use them to upscale both images and video.

Who it suits: ComfyUI users who want to sharpen up low-resolution pictures or clips on their own computer. Workflow builders can also start from the ready-made upscale templates.

Industry | Who uses it | How they use it | Example
Video | Editors | Upscale a clip inside a ComfyUI graph | Open the 3B int8 video upscale workflow and run a clip
Photography | Photographers | Enlarge a picture with the 7B model | Run the 7B int8 image upscale workflow on one photo
Creative work | Digital artists | Compare the 3B and 7B models on the same image | Run both image workflows and compare the results
Tooling | Workflow builders | Build graphs on the repackaged files | Swap one precision format for another in a template

You can skip it if you do not use ComfyUI or have no pictures or video to upscale.

Where to use it:

• ComfyUI, the node-based image tool, on Windows, macOS or Linux
• Your own computer, using files fetched with the Hugging Face tool
• Three ready-made upscale workflow templates for ComfyUI

Use case | Who | What they get back
Video upscaling | Editors | A larger, sharper version of a clip
Image upscaling with 3B | ComfyUI users | A larger version of a picture
Image upscaling with 7B | ComfyUI users | A larger version of a picture from the bigger model

How to set it up: You need ComfyUI and Python for the download tool.

• ComfyUI installed on your computer
• Python, to run the Hugging Face tool
• Disk space for the files you pick, since the card gives no sizes

1. Install ComfyUI if you do not have it.
2. Install the Hugging Face hub tool and download the repository.
3. Move the model files into the diffusion models folder inside the ComfyUI models folder.
4. Move one VAE file into the vae folder. The two VAE files hold the same model, so either works.
5. Open one of the three upscale workflow templates in ComfyUI.

pip install -U huggingface_hub
hf download Comfy-Org/SeedVR2

How to use it: With the files in place, open an upscale workflow in ComfyUI, load your picture or clip and run it. The card shows a before and after picture for each of the three workflows. It gives no commands beyond the download.

Tool | What it does
Video Upscale: SeedVR2 3B Int8 | A ComfyUI workflow that upscales a video with the 3B model
Image Upscale: SeedVR2 7B Int8 | A ComfyUI workflow that upscales an image with the 7B model
Image Upscale: SeedVR2 3B Int8 | A ComfyUI workflow that upscales an image with the 3B model

Example: You place the 3B int8 model file and a VAE in their folders, then open the 3B video upscale workflow. You load a short clip and run the graph. A larger version of the clip comes back, like the before and after pair on the card.

For newcomers and experts: ComfyUI is a tool where you connect blocks to run an image model on your own computer. Upscaling makes a small picture or clip bigger. A first small step is to pick the 3B image workflow, since it needs the smaller model.

The card lists the formats fp16, fp8 e4m3fn, int8, mxfp8 and nvfp4 for each model size, and the 7B model has the same set again in a sharp variant. Lower precision usually takes less memory, though the card does not compare quality or size. The files were converted with a script from Comfy-Org's model tools.

Download the files from the repository
Place the model and VAE in the ComfyUI folders
Open an upscale workflow template
Run it on your picture or clip

Good to know:

• The workflow templates all use int8 files. The other formats are there if you build your own graph.
• The card gives no hardware needs, file sizes or quality comparison between formats.
• The original models are SeedVR2 3B and 7B from ByteDance Seed.
• License: apache-2.0.
• Cost to run: none named, you use your own hardware.
• Data it reaches: files downloaded from Hugging Face, plus the pictures and clips you upscale on your own machine.
• Not for: the card names no limits.

On Hugging Face, 13 November 2026

• Library: diffusion-single-file
• Downloads in the last 30 days: 814,130
• Likes: 154
• License: apache-2.0
• Based on: ByteDance-Seed/SeedVR2-3B
• Fine-tunes: 1
• Tags: diffusion-single-file, comfyui

Links

• Model: https://huggingface.co/Comfy-Org/SeedVR2
• Files: https://huggingface.co/Comfy-Org/SeedVR2/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Comfy-Org/SeedVR2

Then follow the usage in its model card. Nothing in this kit runs by itself.