Re-Illustrating WarStrike With An A.I.

This job market isn't as easy to find employment in as the media lets on. Not by a long shot. But I've managed to establish myself as a graphic/programming freelancer with a couple of good clients.

Fortunately, because I'm now a full-time contractor, there's no limits on what I can create and sell outside of work. So I've cracked open WarStrike's rulebook, and am currently working on upgrading my often so-so stock art collages with much better images generated by an AI. Which has been an interesting experience.

Here's a sample:

My previous placeholders.

Their replacements.

And another sample:

Before

After

The process has been interesting. I'm using mostly Midjourney AI, whose capabilities keep improving at a tremendous rate. You give it a text prompt, and it returns an image. Like this one:

/imagine two soldiers studying a map, sci-fi, dark, dystopian, backlit, downward angle, 16k resolution, 3d rendered, ultra-detailed, --ar 5:7

You then get back a choice between 4 quick images:

You then have some choices:

1. Go back and modify the text prompt, which will give you a new 4-image set that's completely different from the previous one.
2. Roll the dice again, and let the AI come up with 4 new images to show you, based on your original prompt. Which will also give you 4 new images that don't look like the previous set.
3. Tell the AI to create variations on one or more of the 4 images.
4. Tell the AI to create a higher resolution version of one or more of the 4 images.

Unfortunately, you can't do an additive prompt. Like give the guy on the left a rocket launcher. It doesn't work that way. At least not yet.

So in the above case, let's decide to go with image 4 (bottom right) and have the AI create a higher resolution version:

A couple of minutes later...

And... Now it looks like they're walking through a foggy cave. Not exactly the image I was hoping for in order to illustrate the "Organizing A Battle" Chapter.

At this point, I can go back and try another prompt, which I had done already 3-4 times. So I decide to try hitting the "Remaster button". Which attempts to use your current general composition, but clean up the image and make it look more specific. This is the button that will either make your image shine, or make it look like a 10 year old painted it in art class.

In this case, it shined:

I can use this.

Now remember that red head above, from the "How To Read These Rules" page? It took a total of 33+ permutations, complete restarts, and remasterings before I got an image where she was both looking at the camera, and had an emotional expression that would work for that page's context. But I had to go with a result that didn't make her look like the half-robotic cyborg that I originally wanted.

Here's some of the permutations:

Meh...

Only permutations of the bottom-left and bottom-right ones worked at all.

A weak contender.

Meh...

The rough details in this one work, butt he facial expression and composition don't.

This is the one, cleaned up a bit, that I wanted to use.

Commence The Repeated Pressing of The "Remaster" Button...

Sigh... not looking at the camera.

Hi there Scully! Don't want to mix universes, but that's some cool-ish headgear.

Wrong emotion, not looking at the camera.

Beautiful, but still not looking at the camera.

Um... Creepy...

Looking at the camera, perfect hair, but wrong emotion.

Am I boring you?

Look. At. The. Camera!
(also ear farts)

Finally!

From the first red head prompt to the final image took about 1.5 hours.

So is this art? To a certain extent I think so. Since the AI just spits out randomized permutations, you have to have a certain mixture of stubbornness about your goal, a good idea of what it is that you need in your final image, flexibility in the outcome, and an appreciation for what separates appealing images from unappealing ones.

Being able to paint over the result, to tweak it a bit more to your liking, doesn't hurt either. Though all of these images are unmodified Midjourney output.

Thoughts?