So one thing I need for my Substack is the cover image for each article. Now, if you read Frank Talk from your emails, you never see that. But it’s the image for each post you see on the Frank Talk main page and is the image that goes with each article when a link is posted on social media. For a while, I have just been using royalty-free stock images I can find online and settling on “good enough” as I don’t have much time or budget for that, but I’ve started now to play with AI.
My first attempts were with DALL-E, but those were never anywhere near what I was looking for. I’ve started to play with MidJourney, though, and it’s made some pretty impressive results. And at $10 a month for the basic plan, it’s well within my Substack budget. In fact, I’ve already gone through all my Short Stories and replaced the cover images for each (including now a unique image for each episode of Origin Story).
Now, I feel a bit guilty using AI — especially knowing illustrators like Ethan Nicolle — but this seems like a rather disposal use for it and a much better option than what I had been doing. My problem still is I’m not the most visually creative person, but this is an addictive area to play around with. In fact, I’ve already burned through my server hours for the first month learning the ropes.
Now, the important thing to remember with these AI things is to emphasize the word “artificial.” They work nothing like the human mind, using computational brute force to imitate it a bit. What would be really easy to describe to a human illustrator doesn’t quite make sense to these things. For instance, what I wanted for Glenda Dragonfire! is the title character confronting a lich, but any requests for both a girl and a lich in a scene ended up with a lich-like girl.
Or if I finally got a not undead-looking girl, there was no lich.
I ended up having to just have it create a scene of a woman confronting a lich and then used the MidJourney remix option for more specificity on the woman and rerolled until the hands didn’t look weird.
For Who Murdered the Dinosaurs, the image I always wanted was of a crime scene with a chalk outline of a dinosaur. I couldn’t get MidJourney to do any chalk outlines of dinosaurs that looked like crime scene ones, though. I instead tried for crime tape around dinosaur fossils, and ended up getting a decent result using its blend option giving it an image of a T. rex fossil and a crime scene to blend together.
This is fun! I think I might later replace images for other popular posts on this Substack (along with probably mostly using MidJourney for images going forward).
By the way, I hope new subscribers are using the archives; I have a lot of good old stuff, and most of it is evergreen — especially what’s in the Short Story, Fun Shorts, and Humor categories.
Now, the other AI I have been investigating is ChatGPT. ChatGPT is good at coming up with both computer code and writing — my two specialties — but I’m not too worried about it. For one, my writing specialty is humor, and that is where it fails because you can’t just copy patterns to make a good joke; humor involves the unique weirdness of the human brain that brute-force computation can’t quite imitate. Still, I’m curious to try and figure out some use for ChatGPT. There has to be some part of humor that brute force can help with (while a human will still need to ultimately make the joke).
I’m not one of those people who think AI can take over or is worried about the “singularity.” I’m actually pretty bearish on self-driving cars because, while 95% of driving is something a computer can easily do better, that last 5% is so complex I don’t think it can ever quite do it in its machine-learning method. And that’s the way it is with most things, whether it’s art or writing — the vast majority can be done well by a computer, but that last bit the computer can never quite do is critical.
That’s why I’m more interested in how computers can help us be better at what we already do. I think AI art can help actual artists at their job (who in the least would be better at prompting it for visually interesting things than I am). Maybe I can find a way to use ChatGPT to enhance my writing (I already use Grammarly AI to help with typos and some style). And AI can help reduce accidents for driving (while I’m not sure they can completely take over unless we completely restructure streets to accommodate it). Advances in AI should hopefully make things better. In the least, I’ll have some more unique images when I post links to my Substack on Twitter.
ChatGPT actually did really well over at the Babylon Bee in creating fake AOC articles. Like, 70% of them sound real. My favorite the generator came up with is "AOC Mistakes Ocean for Sky, Thinks She Can Fly".
Ha! I love it! Using AI to generate cover images for your articles is pure genius. It’s like the robots have finally taken over the creative department and are now designing the cover art for us. Who needs a graphic designer when you have AI on your side, right? I hope the AI also comes with a sense of humor because the next thing you know, your articles will have cartoon cats as cover art. 😂
(I had ChatGPT write my comment for me. I specified that it should be funny, and the result kind of proves your point...)