High-Quality Image Upscaling with SeedVR2 in ComfyUI

In the world of digital imagery, the demand for high-quality visuals is ever-increasing. This is where SeedVR2 Nightly Upscaler comes into play. It is a cutting-edge tool designed to enhance image resolution while preserving intricate details and textures that standard upscalers often overlook. Unlike traditional upscaling methods that may result in blurry or pixelated images, SeedVR2 utilizes advanced algorithms to ensure that your images maintain their sharpness and clarity, making it an invaluable asset for photographers, graphic designers, and content creators alike.

This tutorial will guide you through the process of installing and setting up SeedVR2 Nightly Upscaler within ComfyUI, a user-friendly interface for managing image upscaling tasks. We will cover everything from the initial installation requirements to practical usage, ensuring you can effectively leverage this powerful tool for your image enhancement needs. By the end of this guide, you will be equipped with the knowledge to upscale images with remarkable quality, transforming your visual projects into stunning works of art.

Before installing the SeedVR2 Nightly Upscaler, make sure your system meets the necessary requirements. You’ll need ComfyUI installed and running on a compatible GPU. For the best performance, a GPU equivalent to or more powerful than the RTX 4090 is recommended, ensuring smooth upscaling of high-resolution images without slowdowns or interruptions.

Requirement 1: ComfyUI Installation

For a detailed, step-by-step guide on installing and running ComfyUI with high-end GPUs on RunPod—including network volume setup and Sage Attention—check out the tutorial below:

Run ComfyUI with Sage Attention on RunPod (Network Volume Setup)

Requirement 2: SeedVR2 Model Download

When you select the SeedVR2 model within the ComfyUI node, it will be automatically downloaded in ../ComfyUI/models/SEEDVR2 folder directory. If you want an overview of all available models, you can find them here: SeedVR2 Models.

Keep in mind the different model sizes and precisions:

3B vs 7B parameters:

• 3B models are faster and lighter on VRAM, ideal if speed is your priority.

• 7B models offer higher quality, better detail, and more accurate upscaling results.

FP8 vs FP16 precision:

• FP16 uses more VRAM but provides higher quality outputs.

• FP8 is faster and lighter on memory, with slightly lower fidelity.

Recommended choices:

• If you prioritize quality and have enough VRAM (24+ GB): 7B FP16

• If you prioritize speed and lower VRAM usage: 3B FP8

This helps you pick the best model for your workflow depending on whether speed or image quality matters most.

We’ll also need to get the nightly SeedVR2 repository, which has the latest features not available in the stable release. Since the Node Manager in ComfyUI doesn’t let us select the nightly version, we’ll need to clone it manually. Don’t worry—we’ll go through this step by step together.

For this guide, we’ll use RunPod as our environment to install and run SeedVR2 Nightly Upscaler. Using RunPod ensures you have a powerful GPU setup ready for high-resolution upscaling. In the RunPod CLI, you can install the nightly version of SeedVR2 with the following steps:

To install the SeedVR2 Nightly Upscaler, you will need to use the terminal, as the Node Manager does not support the nightly branch. Follow these steps carefully to ensure a successful installation:

Step 1: Navigate to the custom_nodes folder

Open your terminal and enter the following command to navigate to the custom nodes directory:

bash
1
2cd /workspace/ComfyUI/custom_nodes

This folder is where you will be placing the SeedVR2 files.

Step 2: Clone the nightly branch of the repo

Next, you will clone the nightly branch of the SeedVR2 repository. Use the command below:

bash
1
2git clone --branch nightly --single-branch https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler.git

This command fetches the latest version of the upscaler, ensuring you have access to the newest features and improvements.

Step 3: Enter the cloned folder

Once the cloning process is complete, navigate into the newly created folder:

bash
1
2cd ComfyUI-SeedVR2_VideoUpscaler

Step 4: Activate your existing virtual environment

Before installing the necessary dependencies, activate your Python virtual environment with the following command:

bash
1
2source /workspace/venv/bin/activate

This step is crucial for ensuring that the dependencies are installed in the correct environment.

Step 5: Install dependencies

Finally, install the required dependencies by running:

bash
1
2pip install -r requirements.txt

This command will download and install all the necessary packages for the SeedVR2 Nightly Upscaler to function properly. By following these steps, you will have successfully installed the upscaler and be ready to configure it within ComfyUI.

Once you’ve installed the SeedVR2 Nightly Image Upscaler, the next step is to load its workflow in ComfyUI. This workflow simplifies the process, preloading all necessary nodes and connections so you can start upscaling images quickly and efficiently.

Step 1: Download the Workflow File

Start by obtaining the JSON workflow built specifically for SeedVR2 Nightly Image Upscaler. This workflow already includes all the necessary nodes for SeedVR2 image upscaling and basic processing. You can find the download link for the SeedVR2 workflow in the resources section of this tutorial.

👉 SeedVR2 Nightly Image Upscaler Workflow JSON

Step 2: Load Workflow in ComfyUI

After downloading the workflow file, open ComfyUI and drag the JSON file onto the canvas. This action will automatically load all the SeedVR2 nodes required for upscaling, along with pre-configured connections and default settings.

Step 3: Configure Node Settings

Settings Step 1: Upload Your Image

Start by uploading the image you want to upscale. This will serve as the input for SeedVR2 Image Upscale Workflow.

Settings Step 2: Configure BlockSwap Settings

SeedVR2 includes the BlockSwap feature, which helps manage VRAM usage and performance during upscaling. Configure it as follows:

💡 Tip: Adjust blocks_to_swap based on your GPU’s VRAM capacity and the target upscale resolution.

Settings Step 3: Configure SeedVR2 Extra Args Node:

In the SeedVR2 Extra Args node, adjust the following settings:

These settings help balance performance, memory usage, and output quality.

Settings Step 4: Configure SeedVR2 Upscaler Node:

In the SeedVR2 Upscaler node, adjust these key parameters:

Once all these settings are configured, you’re ready—simply click RUN to start upscaling your image!

When upscaling with SeedVR2, you may occasionally encounter Out of Memory (OOM) errors. To help reduce VRAM usage and prevent crashes, consider the following adjustments:

• Lower the new_resolution to a smaller pixel size.

• Increase blocks_to_swap to a value between 32 and 36 to reduce memory load.

• Enable Tile VAE to process the image in smaller sections.

• Reduce VAE Tile Size to a lower value, such as 512, to save VRAM.

• Enable Preserve VRAM to further optimize memory usage.

Applying one or more of these strategies will help SeedVR2 run more smoothly, allowing you to upscale images efficiently even on high-resolution workflows.

The model size and precision you choose in SeedVR2 have a direct impact on upscaling speed and final image quality. The table below shows a general comparison between the 3B FP8 (faster, lighter) and 7B FP16 (higher quality) models when upscaling to 4K and 8K on an RTX 4090, assuming the model is already loaded in.

These benchmarks give you a good idea of what to expect when choosing between speed and image quality for different projects.

Below is an example of an image that has been upscaled to 4K using the 7B FP16 model. Notice how the clothing fabric, hair strands, and eye details appear noticeably sharper and more refined than before. Even though the original was a 720×1280 AI-generated image, the results achieved with the SeedVR2 upscaler are impressive. To fully appreciate the enhanced clarity, make sure to select the highest quality settings on the YouTube video. Take a look at the difference below:

In conclusion, the SeedVR2 Nightly Image Upscaler offers a powerful solution for high-quality image upscaling, making it an essential tool for anyone looking to enhance their visual content. Throughout this tutorial, we have explored the installation process, configuration, and practical usage of the upscaler within ComfyUI. By following the steps outlined, you can now confidently upscale images while preserving their details and textures, resulting in stunning visuals that stand out.

As you experiment with the various settings and parameters, don’t hesitate to combine the SeedVR2 upscaler with other nodes in ComfyUI to further enhance your images. The flexibility and power of this tool allow for endless creative possibilities.

For those interested in taking it a step further, including video upscaling, check out this guide: How to Upscale Videos in ComfyUI.

Thanks for following this tutorial, and happy upscaling!