Why you should upscale your images

Hello Intelligent Innovator.

I've got another FREE PDF for you. Midjourney has made a lot of progress and changes, so I’ve updated my Midjourney style templates. I’ve been posting examples of it in my recent weekly galleries. I’ll talk more about it at the end of this post and if you can’t wait, you can find it here.

Now, let’s get to today’s post - Upscaling.

It might seem like an insignificant extra step, but Upscaling your images can make a big difference in your AI workflows.

They keyword to explain why is NOISE.

Text to image AI works on diffusion models. These models are given a dataset of images. The images are scrambled with noise, until they are fully unrecognizable. To us at least. The goal of the diffusion model is to unscramble this noise and recreate the full image out of it.

Let’s say we an image of a cat. The model gradually applies noise to it, until it’s fully scrambled with it. It then attempts to unscramble that noise and recreate an image of a cat. It will remember this noise pattern every time it tries to recreate a cat. The more reference images it has and the more time it attempts, the better it gets.

You get the idea. I hope.

So what does this have to do with upscaling your images?

Well, if you use image to image in your workflow (either in Midjourney or Stable Diffusion), the resolution of that image plays a big role in the final result.

When noise is applied to the image, the more defined the details of the image are, the more accurate the AI can be in unscrambling them.

Let me show you with an example.

I have this base image I want to use as reference. The AI model will apply noise to this and then attempt to recreate the image.

I put it into ControlNET in Stable Diffusion with the canny model to show you the difference in details that the AI will read.

Here are the results with the same prompt and same seed:

And the best thing about upscaling - it takes very little time.

There are many websites and software which makes upscaling as easy as pressing a button.

Here are some options:

ImgUpscaler - free, browser based, with 5 credits per day.

TopazLabs - Professional grade workflow, with many features. Can be expensive for some (this is an affiliate link).

Upscale Media - Another free alternative. Doesn’t have many options, but does the job.

You can upscale directly in Stable Diffusion as well, but the higher you go, the more intense it gets for the GPU.

I would highly suggest running your images through and upscaler, if you intend to use them further in your creative process.

I hope you’ve enjoyed today’s post. It was quite dense with information. 😁

As mentioned at the top of this post, I’ve updated my Midjourney Style Templates. It’s free. Or not, if you wish to support my work. 💛

I’ve added examples for the latest Midjourney version along with --v 4, which some of the templates were created on.

I’ve also added 4 new styles:

• Watercolor Manga

• Dark Fantasy

• Magic University

• Mythology (v 5 only)

When I start creating, I often turn to these to explore my ideas in different styles. Along with the techniques I’ve included in my Brainstorming guide.

I hope you will enjoy playing around with different styles and make them your own.

You can stop by my website or social media and as always,

Keep creating and don't forget to have fun. ☀