What is upscaling in ComfyUI?

Upscaling in ComfyUI is the process of increasing an image or video resolution while preserving or enhancing visible detail. AI upscalers use trained models to add realistic detail rather than just stretching pixels.

What is the best upscaler for ComfyUI?

SeedVR2 is widely considered the highest-quality upscaler in ComfyUI for both images and videos. For fast, simple results, NMKD-Siax and Z-Image 4x upscalers are excellent basic options.

ComfyUI Image & Video Upscale Guide: Basic + SeedVR2 (2026)

Q: Why is my ComfyUI upscale running out of VRAM?

Use a smaller model variant (3B or GGUF) instead of the full 7B fp16 model. You can also reduce batch size, tile the image, or split videos into shorter clips.

Why Use Upscaling in ComfyUI?

Quick answer: Most AI models generate images at HD or SD resolution — not 4K. Upscaling lets you increase resolution while preserving or adding visible detail, without regenerating from scratch.

What upscaling does

Many AI models generate images at HD or SD sizes — for example 1920×1080 or 1280×720. They are not trained to produce clean 4K images. Upscaling lets you increase resolution while keeping or improving visible detail, using a dedicated upscaler model rather than simply stretching pixels.

When to use it

You generated an image that looks good but is too small for your use case.
You have a low-res video and want a cleaner, larger version.
You want to recover dark or low-detail areas without changing the overall composition.

💡 Tip: Some upscalers change facial detail or texture. If the result looks too soft or too sharp, try a different upscaler model before adjusting other settings.

Image upscale examples

ComfyUI image upscale before and after — SeedVR2 upscale result showing improved resolution and detail, example 1 — SeedVR2 upscale — example 1

ComfyUI image upscale before and after — SeedVR2 upscale result showing improved resolution and detail, example 2 — SeedVR2 upscale — example 2

Video upscale example

ComfyUI video upscale example — before and after using SeedVR2 video upscaler

Image Upscale Methods in ComfyUI

There are two main approaches covered in this guide: a basic upscaler for fast, simple results, and the SeedVR2 method — a popular, high-quality option that handles difficult images and dark areas better than simpler models.

SeedVR2 Image Upscale Workflow

SeedVR2 often gives better detail and more natural restoration of shadows and dark areas. It can outperform simpler upscalers on images with complex textures or faces.

Download SeedVR2 Image Upscale Workflow

Steps to run the SeedVR2 image workflow

Download the SeedVR2 workflow

Use the download button above to get the workflow JSON file.

Fix any red nodes

Red nodes indicate missing models or external nodes. See the guide on fixing red node issues in ComfyUI for a full walkthrough.

Install the required GitHub nodes

Install these two custom nodes via ComfyUI Manager: ComfyUI-SeedVR2_VideoUpscaler and ComfyUI-Inspire-Pack.

Load your image

Load your source image into the Load Image node in the workflow.

Choose a SeedVR2 model

Strong GPU (16GB+ VRAM): use 7B fp16. Limited VRAM: use 7B fp8 or the 3B model variant. The workflow will download the model automatically on the first run.

Run the workflow

Queue the workflow. The first run will download the selected model. Subsequent runs will be faster.

SeedVR2 upscale examples

SeedVR2 image upscale result in ComfyUI — detailed texture restoration, example 1 — SeedVR2 result — example 1

SeedVR2 image upscale result in ComfyUI — detailed texture restoration, example 2 — SeedVR2 result — example 2

Basic Image Upscaling in ComfyUI

The basic upscaler is the fastest way to get a larger image. It works well for 2–4x increases on images that don't require complex detail restoration.

Download Basic Upscale Workflow

Step 1 — Download and place the upscale model

Place your upscale model file in the correct ComfyUI folder:

ComfyUI/models/upscale_models

Download one of these recommended models and place it in the folder above:

4xNMKD-Siax (Civitai) — excellent all-round 4x upscaler.
Z-Image 4x Upscaler by Helaman (Civitai) — good for photorealistic output.

Step 2 — Load the workflow and run

Load the workflow JSON in ComfyUI

Drag and drop the downloaded JSON file onto the ComfyUI canvas, or use Load → Import Workflow.

Put your image into the Load Image node

Click the Load Image node and select your source image.

Select your model and run

In the Load Upscale Model node, select the model file you placed in the upscale_models folder. Then queue the workflow.

Basic upscale examples

ComfyUI basic upscaler result — image before upscaling showing lower resolution detail — Basic upscale — example 1

ComfyUI basic upscaler result — image after upscaling showing improved sharpness and resolution — Basic upscale — example 2

Video Upscale Workflow (SeedVR2 Video Upscaler)

Video upscaling uses the same SeedVR2 model but processes frames or clips through the upscaler. The result is a higher-resolution video with improved detail across every frame.

Download SeedVR2 Video Upscale Workflow

How to use video upscaling in ComfyUI

Download and load the SeedVR2 video workflow

Use the button above to download the workflow JSON, then load it in ComfyUI.

Load your video

Load your source video into the Load Video node in the workflow.

Select the SeedVR2 model and run

Choose your SeedVR2 model (3B for limited VRAM, 7B fp16 for best quality) and queue the workflow.

Video upscale example

SeedVR2 video upscale result in ComfyUI — before and after comparison

⚠️ Warning: Always check the input frame rate and aspect ratio before running. Mismatched settings can cause the output video to play at the wrong speed or be cropped incorrectly.

Common mistakes with video upscaling

Not checking input frame rate or aspect ratio before running — verify these first.
Running a high-resolution model with too little GPU memory — use the 3B model or split the video into shorter clips.
Upscaling a very long video in one pass — break it into 5–10 second clips to avoid VRAM errors and speed up the workflow.

Simple video upscale workflow

For a faster, less resource-intensive option, use the simple video upscale workflow below. This is a good starting point before moving to the full SeedVR2 workflow.

Download Simple Video Upscale Workflow

Tips and Troubleshooting

The most common issues with upscaling in ComfyUI are red nodes, VRAM errors, and slow video processing. Here is how to fix each one.

Red nodes — missing model or external node

Fix: Check the ComfyUI console for specific error messages. Install the missing custom node via ComfyUI Manager, or point the node to the correct model file path. See the missing nodes guide linked in the related tutorials below.

Out of VRAM error during upscaling

Fix: Use the 3B model or a GGUF variant instead of 7B fp16. You can also reduce the batch size or enable image tiling to process the image in smaller sections.

Video upscaling is very slow or crashes

Fix: Break the video into shorter clips (5–10 seconds each) and upscale them separately. Export frames, inspect a few upscaled frames first to confirm quality, then encode the full video.

Upscaled image looks too soft or oversharpened

Fix: Different upscaler models have different strengths. Try NMKD-Siax for balanced output, or switch to SeedVR2 for images with complex detail. For SeedVR2, try reducing the model strength slider slightly.

💡 Tip: For videos, always export a few frames first and inspect them before encoding the full video. This saves significant time if the settings need adjusting.

Frequently Asked Questions

Upscaling in ComfyUI is the process of increasing an image or video resolution while preserving or improving visible detail. AI upscalers use trained models to intelligently add realistic detail rather than just stretching pixels.

SeedVR2 is widely considered the highest-quality upscaler available in ComfyUI for both images and videos — especially for complex textures and dark areas. For fast, simple results, NMKD-Siax and Z-Image 4x are excellent basic options.

Place upscale model files in the ComfyUI/models/upscale_models folder. After adding the file, refresh your browser tab so ComfyUI picks it up in the Load Upscale Model dropdown.

Download the SeedVR2 video upscale workflow, load your video into the Load Video node, select the SeedVR2 model, and run the workflow. For large videos, split into shorter clips of 5–10 seconds to avoid VRAM errors.

Switch to a smaller model variant — the 3B model or a GGUF quantized version uses significantly less VRAM than the full 7B fp16 model. You can also reduce batch size or enable image tiling to process in smaller sections.