The Perils of Prompting and when I tried to follow AI art instructions on YouTube

A personal work of Thurwar and Staxxx from the novel, Chain-Gang All-Star by Nana Kwame Adjei-Brenyah.

This was a hard book to read. It very violent and deals with justice or injustice or injustice of the justice system. I don't know... It’s satire. When prisoners fight to the death for entertainment, you know it's going to be a tough book.

So I didn't want to do an action scene, because that wouldn't feel right. But the book ends (SPOILER ALERT) with Thurwar and Staxxx two lovers and team mates having to fight their last fight with each other.

Let's start with the character pose. I want to pose them in a sort of hugging pose (In the way athletics will hug their teammates after a goal or after winning a game, I mean, killing is a sport in this world). So that was kind of the idea I had as I browsed Pinterest and added image to my pureref board:

From looking at a lot photos of athletics (I was looking at emotions and how it's expressed in the body). The one on the left stood out for me of the football (soccer) players in black and yellow. It looked like they lost.

That was perfect. Because Thurwar and Staxxx are the loser here. The corporation wins. Damn. They always win…

So I tried to create something similar in blender, playing with the camera and the angles:

I kitbashed a few main elements, the two characters (using the Human Generator addon) and their weapons. The Jumbotron and the hovering camera thingy (HCU - lets audience get right in the action!) I added later in the process, but you can see it here.

So here's the output in Blender with how I imagined the lighting would be in the arena:

For the background, I went to AI prompting for that. AI can do a lot and can be used in many ways, but I think for most people when they talk about AI art, they're talking about prompting. Type something and BAM get an awesome image. In reality, it doesn't work like that. The worst part of prompting is how hard it is to control and make decisions.

I feel like artmaking is a process and a puzzle. If I prompt something BAM all or a lot the decisions are made for me. Which is kind of annoying because from my own experience, I would do things differently than what the AI choose. Also as an artist, I think every element in the picture needs to serve a purpose. But AI will just spew things out.. Very nice images, but not what I need.

Which is why I work this way, creating the composition in Blender first, I can make the decisions that are important to me and control the elements in the image. But sometimes, I'm happy to prompt because a decisions might not be as important. Like the background here:

I used Leonardo.ai using Leonardo Kino XL with photo real turn on. The prompt: "a futuristic gladiator arena, night, bright lights."

I thought anyone of these could work... But I went with the one with gladiator on the top left.

I photobashed Staxxx's hair and added the background and inputted this into Leonardo again.

This time is used Leonardo Lighting XL with an init strength between .6 and .7ish.

I would go through a lot generations changing the noise strength, but there would always be one image- BAM, OMG, that's Thuwar, that's really her. Or that's Staxxx! Or that's the freaking hammer! You know how I said AI make a lot of bad decisions. Sometime, they nail it. But never the whole image. The hammer might be right in one image while the hair is perfect in another. So I have to composite all the parts I want to use.

Then I out-painted the edges to get more space above the head. I thought a large jumbotron screen would fit there perfectly. Used Fooocus for the outpainting.

For the jumbotron and hovering camera, I went back into Blender. What's nice about this workflow is I can still go back into 3D to make adjustment or additions. Or sometimes change the camera angle or focal length, or lighting, if the composition is not working. Then, I render again, and blend it into the 2d image:

Darn, I forgot Staxxx's tattoos. Whoops. She has a X on her body for every kill in the arena (don’t try to count). Maybe I should go back into Blender and add them as a decal... That would be smart. Hmm or maybe I can be fancy and try inpainting. Decisions, decisions... I decided to try inpainting because I recently saw a youtube video that showed a cool trick by making CPDS image of the tattoo and inpainting it. I love trying new things.

So I tried that using Fooocus. And it didn't work. I rewatched the youtube video and tried again and again... Just didn't work for me:

So after wasting a lot of time, I did something wild... I drew them.

It's almost there, but the last 10% is always full of tiny things that bothers me. The more I look at it, the more I see things that I want to change, decisions, decisions... And things I forgot, like the glowing bands on the prisoner's wrists.

So after 20 layers of more tweaking I get to the point that I'm sick of tweaking and I guess that’s how I know I’m done.

I'm happy with this. It reminds me of the book, which makes me sad. Darn.

If you haven't read Chain-Gang All-Star by Nana Kwame Adjei-Brenyah, check it out. I don't think it's for everyone but I'm glad I read it.