Understanding Colors for AI Artists

Why you need to grasp RGB, CMYK and HSL colors

Erik Engheim

--

With AI Art, you write what you want and there is no need to know traditional artist stuff like color models, right? Wrong, if you really want to get good at AI Art, you need to actually learn the usual stuff that any artist has to master.

In this regard, the current article isn’t specific to AI artists, but to anyone who would like to get into photo editing or digital painting. But it is written from the perspective of someone who primarily do AI art.

Let me give some context for why an understanding of colors matter. Consider the title image of a futuristic female astronaut. It was generated with a text prompt containing text similar to the one you see below:

woman in shiny hard yellow moulded plastic futuristic armor, 
helmet, quantum effects, neon lights, space futurism

From this text, an AI art generator tool built around, for instance, the Stable Diffusion library, can generate a variety of images upon the same theme. If we add an image to guide the process, we will typically get an image which are smaller or larger variation upon the starting image depending on your settings. This approach is what we usually refer to as image to image.

Of all the similar images you get out, you rarely get one perfect image where everything is great. Maybe the suite worn by the astronaut looks superb in one image, but the face looks bad. Or perhaps one image has a good-looking face but the eye color or hair color is wrong. If there are hands in the image, they will likely go wrong. The left hand might be fine while the right hand is horribly mutated, or vice versa.

In other words, you need to be able to composite images, and combine the best elements from different images into one perfect image. Blending in parts from one image into another is a separate skill I will not cover here. Basically, it involves using masks to obscure parts of an image you don’t want included and let good parts from an image in a layer below shine through.

Blending in parts was the first skill I developed, but I quickly noticed that something looked off. The reason things look off when you composite different images in AI Art is because the…

--

--

Erik Engheim

Geek dad, living in Oslo, Norway with passion for UX, Julia programming, science, teaching, reading and writing.