top of page

When Creativity Meets Technology: Exploring the Art of Coaxing a Character From an AI Artist

Updated: Dec 27, 2023

Leonardo AI is my go-to choice when I want to create a single image on a particular topic. After a few tries with various prompts, I usually get an image that satisfies my vision. I also found that if I use similar words, I can build up a set of images that have the same style. But what about characters? Wouldn't it be great if I could create a character that appears in many different images? That would provide an opportunity for me to create a story.


I wasn't getting anywhere on my own, so I found a blog post Generate the Same Character Consistently in Leonardo AI. There are two ways. One involves developing a prompt for the character and then using that prompt consistently. The other is to create a character look, and then use that image to prompt another, similar image. Both ways are easier said than done.


Prompts are an art in and of themselves (See Leonardo Prompt Guide). Besides being specific about what you want to see in a character, it is also important to include a negative prompt for the things you do not want to see. If you've used and AI artist, you've probably seen living things (people and animals) created with malformed hands and feet, two heads, and so on. Early on I saw a few iterations where my character, Bob Acorn, was old or had a beard. I wanted neither. I also wanted a well-formed character. My negative prompt was:

beard, deformed body, four hands, four heads, four limbs, four legs, low resolution, blur, ugly, old

I iterated on my basic character prompt until it was:

Bob Acorn a 4 foot tall forest character with a purple nose wearing a green jacket and a whimsical hat

Then I tried to get Bob to pose. Here he is reading a book:


Bob Acorn eats a bowl of cereal. Cute, but not the same Bob. Maybe it's his cousin?


Bob uses a cell phone. Rather, Bob's brother?


I made one attempt at using one of Bob's images to generated another, but I have yet to master that technique. No matter which technique, it takes a fair amount of time to develop a consistent looking character. Even then, getting the AI to pose the character is a challenge. I could not get Bob to assume a resting pose, close his eyes, or do laundry. When I asked to show him baking a cake, the cake was already finished, and by a different version of Bob.


Bob does like to eat and has no trouble handling an ice cream cone and another sweet treat. (I asked only for the ice cream. AI added a second treat.) Perhaps the fact I can't get Bob to do laundry or sleep says something about his personality. He's a work in progress. But first, I need to refine my prompts to get a more consistent character.



Iterations of a Character: From Start to Finish


16 views0 comments

Comments


bottom of page