Giving the Computer Ideas

Dearest Rachel –

Admittedly, most of what I’ve been doing with the whole artificial intelligence artwork thing has been me making pictures of you. Some have been astonishingly realistic, to the point where, if I wasn’t the one creating them (and therefore knowing better), I might be able to implant false memories about you into my mind. Others are realistic apart from the knowledge that you would never wear such an outfit; either out of personal preference (like the fact that you would never wear blue jeans, or that neither of us would ever think of taking boudoir photographs of each other) or the ensemble’s impossibility (be it literally impossible to exist outside of the mind of, say, M. C. Escher, or just that we could never afford such a garment). And then, there are those that are clearly illustrations; they aren’t photorealistic at all, occasionally even depicting a scenario that couldn’t exist in real life. Some of these are fantastic, others (thanks to the still-obvious limitations of the medium) are horrific.

My point is, so much of what I’ve created has been of you, specifically; I probably wouldn’t be nearly so into this if I weren’t trying to recreate you, to a certain extent. But that’s not really something I can take and make much in the way of content for others to consume. Even the idea that’s been percolating in my head for a year or two about trying to create a manga about our relationship – and my attempt to recover from its sudden end (I’d even come up with a title: After Happily Ever) – would probably only be of interest to me and the folks who read this over your shoulder, if that (and it would require a better grasp of consistent character rendition than I’ve mastered thus far, at any rate). It’s not like I need to monetize this hobby, to be sure, but I’d like it to be something more than a solitary pursuit.

Well, there is one thing that comes to mind… and it’s not like I haven’t used AI to create something for it before. I think I’ve mentioned it before – although maybe only in passing – but I’ve set up a mini-studio with our favorite source for T-shirts online. Starting with that “Kiss Me, I’m Irish” shirt that I devised a year and a half ago, and moving on to a couple of others specifically for my various travels, I’ve also created several shirts that incorporate AI design either in part or entirely. So now, I thought I could walk you through how I do these sort of things.

For example, my weight loss journey has been fraught with various struggles, both in terms of diet and exercise. These are issues that are fairly common to everyone that’s tried to lose weight, I’d hazard, so making a joke off of this subject might prove to have a little more mass appeal than my usual material. I’ve been thinking of making a cartoon of a gladiator staring down at a bowl of salad, with a punchline of “Hail Caesar; we who are about to diet salute you” – a corruption of the greeting offered to the nobles in attendance at a typical gladiatorial circus, “we who are about to die salute you,” combined with the implication that it is a Caesar salad he is about to partake in as his training meal.

And that’s basically the instructions I’m giving the system, and letting it figure out how it wants to render this idea.

Now, there are several reasons as to why I’m using AI to do this. The first is fairly obvious; I’m not a particularly good artist. I’m better than some, but it doesn’t come easy, and I can never quite get a drawing to my satisfaction – which means that, had this been left to my own skill, this would never come to fruition.

The other reason isn’t quite as obvious at first, but makes a lot more sense when you look at the output the computer generates. It would take me hours to draw one cartoon (assuming I could draw one the way I envision it in my head), but it can generate one every couple of minutes. I basically gave the computer the idea, and let it produce various iterations of the concept:

And while I worked the booth in the morning, or toiled away in the gym during the afternoon, the computer kept grinding away at my description, coming up with various images, poses and expressions for this cartoon image. What you see here, in fact is barely the tip of the iceberg, as it gave me over three hundred different poses, body types, expressions and so forth.

In fairness, a lot of the results were rubbish. Sometimes, for instance, the helmet set on the table (and why would a guy have two helmets, one on his head, and another on the table, in any event?) had a head in it, which gave the image a weird creep factor. Others were just not useful for my purpose, as the computer leaned into the whole “Hail Caesar” bit, and had the guy tucking into his salad with great gusto. Great execution, but not the idea I was trying to convey. I wanted something like disgust, disappointment or just plain resignation.

Eventually, I settled on about a dozen or so finalists, which you can see here. Even here, I could see things that would need editing – for instance, the gladiator being highlighted would need that rat-tail edited out (and I have to admit to not being fond of his plate being full of what Daniel commented looked more like edamame than a typical salad. It would garner the same reaction from the gladiator, sure, but it loses the whole Caesar joke, as far as I was concerned).
Once the three of us settled on a finalist, I still had to do some editing, as none of these were absolutely perfect. I tried to turn the tomatoes into croutons, but to be honest, they look rather like olives this way. Still, they’re small enough in the overall picture that it shouldn’t matter all that much – I hope.
I also got rid of the background, as this would be replaced by the T-shirt color (whatever someone should choose), and the helmet on the table (since having two helmets would be redundant). It’s been pointed out to me that his gear is more that of a Roman army officer than a gladiator, but I don’t know how many people would know the difference, and this is more distinctively Roman, anyway, so I think I can live with this.
From there, I have to upload the finished design to the site, and add a blurb and several keywords for people to search for my image.
And this is the finished product. I think maroon looks pretty good, although I’ll admit that the black text doesn’t work as well against it as I’d hoped. Still, I’m pleased with the result, by and large; it’s certainly better than anything I could have done on my own.
And thus far, basically half my output has included AI-generated content, in whole or in part; three of which I’ve uploaded in the last couple of weeks.

Again, I don’t expect to be able to make a living from this; I’m just having fun with this, and if someone else likes what I come up with, all the better. It does give a dopamine rush every time I manage to publish a new idea, though, and makes me want to come up with something else. I’m thinking about maybe a rodeo clown hiding out in his barrel from an angry bull, with the pull quote of “This ain’t my first rodeo, but two rodeos ain’t enough to make me no expert nohow.” Whaddya think, honey?

Anyway, thanks for listening to me blather on about this; you might have been intrigued to hear about this back in the day, but compared to your afterlife, I don’t know. Still, if you get the chance, keep an eye on me, and wish me luck. I’m probably still going to need it.

Published by randy@letters-to-rachel.memorial

I am Rachel's husband. Was. I'm still trying to deal with it. I probably always will be.

4 thoughts on “Giving the Computer Ideas

Leave a comment