5 mins read

Discover This Image Upscaler with ComfyUI – A Possible Alternative To Magnific AI

Ultra Upscale is a free alternative to the Magnific AI Upscaler, run it directly on your PC.

Image Upscalers in General

An image upscaler is a tool whose objective is to increase the resolution of an image without affecting its quality. This can be done using a variety of methods, including traditional upsampling algorithms and more recent AI-powered techniques.

Traditional upsampling algorithms typically work by simply interpolating new pixels between existing ones. This can sometimes lead to blurry or pixelated results, as the algorithm is not able to accurately predict the missing information.

AI-powered image upscaling has emerged as a more sophisticated and powerful approach to upscaling images. These algorithms are trained on a large dataset of high-resolution and low-resolution images, which allows them to learn the patterns and relationships between pixels. This allows them to basically create new pixels in a more realistic and detailed way, resulting in higher-quality upscaled images.

What would you need it for?

In general, it gives much more clarity to the images, making it visually more interesting. But some practical applications could be:

  1. Reviving Old Photos: Upscaling enhances old or low-quality photos, adding clarity and life by sharpening details and reducing noise.
  2. Large-Scale Prints: It enables the creation of large prints from smaller images, maintaining quality for impressive, large-format displays.
  3. Polishing Digital Art: Upscaling improves digital illustrations, giving them a professional, high-quality finish for prints or digital use.

About Magnific AI

Magnific AI Upscaler is an image enhancement and upscaling tool designed for improving the quality of photographs and digital images. It uses advanced AI algorithms to upscale images to higher resolutions while maintaining clarity and detail.

  1. Image Enhancement: It effectively sharpens details and reduces noise, resulting in clearer and more realistic images.
  2. High-Resolution Upscaling: The tool is capable of increasing image size without significant loss in quality, useful for large prints or high-resolution displays.
  3. Customizable Settings: Users have the ability to adjust the level of enhancement to suit their specific needs and preferences.

The only issue is that this tool is closed source and it’s not free. Whether you think this tool deserves your money, that depends on you and your use cases. However, we can try to obtain similar results by using Stable Diffusion and ComfyUI – if you have a good NVIDIA GPU card or a Mac with the latest chips, you art good to go.

Upscale with ComfyUI

A new tool, named Ultra Upscale, has been shared by the community. Using ComfyUI, we can simply load a workflow with all the nodes and default parameters ready to be used, creating our personal image upscaler. It can be downloaded on Civitai, and credits go to the author of that article.

If you need to setup the UI, you can have a look at this ComfyUI tutorial to get started.

Ultra Upscale workflow

If you see some red nodes, just use the ComfyUI manager to install the missing custom nodes. After that, choose an SD1.5 or SDXL model from the Load Checkpoint node, and a Lora in the Load LORA. I used ClystarClearXL in these examples.

You will also need some upscale models. You can see some suggestions in the box located in bottom-left corner; once you have at least one model, you can put it in the ComfyUI/models/upscale_models folder (NMKD Artisofject, Remacri).

The workflow will perform three main steps, creating an upscaled image each time. Just choose an image to upscale and leave the rest as it is. You can also specify a prompt to obtain different results, but I personally just leave this field empty. When ready, hit Queue Prompt button.

Depending on your specs, it might take a while to complete the whole workflow. On my NVIDIA 3060 with 12GB of VRAM, starting with an already big image of 1024×1024, it took around 12 minutes! Using a 512×512 image will take much less time.

Examples

You can use the last node to compare the original image and the upscaled images.

Right side is upscaled
Some upscale models or parameters might reveal tiles in the background

In order to obtain different results, you can tweak the parameters in the Self-Attention Guidance node, following the tips provided by the author: scale: 0 to 1 (beyond 1 causes artifacts); blur_sigma: 0 to 2 (beyond 1 can be seen blurred areas)

One last example, modifying some of the parameters and using the Remacri upscaler:

Scale: 0.3 – blur_sigma: 0.7

That’s it! The workflow is very simple to use, and thanks again to the creators that shared this. Definitely a good alternative to many paid upscalers such as Magnific AI, but they are still valuable for people that do not have a powerful enough PC to run Stable Diffusion.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.