Five Characters In Search Of An Illustrator

If you're going to carry on reading today's post, I'm afraid you're going to have to indulge me. There's no purpose to it other than for me to create yet one more self-indulgent reflection of myself I can gaze into, lovingly, which seems to be my main occupation these days. Also, it's mostly about AI so that's always welcome, isn't it? Anyway, consider that a trigger warning and act accordingly.

I have mentioned, repeatedly, that I've been spending a lot of time (A lot of time...) working on making artificially-generated music from various organically-generated sources. Most recently, and very definitely most successfully, the source has been a novel I tried to write in the mid-90s. 

It was (Still is, I guess...) called The Final Line and it revolves around a chosen family - not an expression in common use then, I don't think - of four youngish adults and a close friend or two as they deal with the typical concerns of any group of friends in their late twenties or early thirties - death and resurrection leading to demonic possession, mass murder, general mayhem and being hunted to extinction by the provisional wing of the Roman Catholic church. The usual everyday stuff we've all been through, I'm sure.

It's a bit like the plot from This Life grafted onto the characters from Twin Peaks but in an episode of the X-Files, I suppose. It was the nineties, after all. 

It's also the sort of thing that everyone in in my own friend group back then was doing. I don't mean we were all hanging out together, drinking and smoking in the day and going out murdering people at night, well not the last bit anyway. No, I mean the construction of extended narratives focusing on that kind  of behavior. I think at least half a dozen people in the Amateur Press Association I belonged to back then were doing something similar. Maybe more.

I hadn't read my story for a good, long while before I picked it up again a few months ago. I'd run out of songs I'd put down on cassette in the eighties and I was scratching around for something to use to keep my Suno addiction fed. God forbid I should actually have to sit down and write something original. I'm far too old for that.

Johnny Paradigm

It occurred to me I might be able to extract a paragraph here and there from something I'd already written, just to keep the hobby going a while longer. The obsession, I mean. Let's call, it what it is.

Being my own #1 fan, I'd always thought it was good but re-reading it after a long gap I was astonished just how good. As I've often aid, I had no idea at the time how I wrote it other than it involved what I like to think of as a fugue state, like I know what that means. There are many passages I can't imagine having written but apparently I did. I mean, there was no-one else there at the time so it must have been me, right?

I'm the worst person to assess it anyway, so I'll stop going on about it and get on to the point I was going to make. Two points, really. 

First, it turns out descriptive, poetic prose is a really good source for song lyrics. Barely needs tweaking. It took me a while to get into the swing of hacking it up thoroughly enough to make it flow properly but even from the start it was clear it was going to make for much better songs than I used to write, mainly because it forced me out of the usual, rigid, traditional structure I always thought songs demanded. Anyone would think I'd never actually listened to any.

This isn't a post about songwriting or using AI to make music, though, believe it or not. I'll do one of those soon I expect. Or maybe I won't.

The second point about the story was how much of it there was. I never finished the draft that I was embarrassed to call a "novel", which is probably why I'd always thought of it as being quite short but I've now transcribed it I find it comes to more than 35,000 words. 

I haven't done a final word count because I only have it in the seventy or so original pages, as it was published in the apa. I need to proof-read it, edit it and collate it into a single file before I'll have an exact count but it's definitely pushing the accepted boundary of a novella - forty thousand words - and it's not even finished.

I was dreading getting it onto the PC. That would be a lot of typing. Fortunately, this is the 21st century and no-one needs to type anything twice any more. I couldn't find a readable file for the original draft on any of my old floppy disks but it occurred to me there was probably some free website that would turn an image of text into a text file for me. And there was. There is.

There are loads of them in fact but the one I used, the best by far of the several I tried, was the no-frills and proud of it jpgtotext.com. If you ever need to convert a Jpeg into a text file, I thoroughly recommend this extremely simple option. It allows you to upload ten images per day for free, although one day it let me upload twice as many for no apparent reason. If you need more than that, the pricing is extremely cheap but I just did my free allotment every day and had it all done in a couple of weeks.

Cado Babe with Cathy

It produces a very accurate text file in seconds, even from some pretty dodgy thirty-year old photocopies of dot-matrix print-outs. None of the other websites I tried came anywhere close to doing as well. Most of them were next to useless, in fact. Some couldn't read the images at all and those that did garbled so much it would have been quicker to r-type the whole thing than correct the errors in the edit.

Some of those failures claimed to be "AI driven". This seems to me to be a very good example of one of the problems with what we're calling AI these days. Most of it isn't AI at all. I bet those weren't. People slap "AI" on the front of all kinds of apps and programs and utilities that would just have been called algorithms a couple of years ago because they think it's the way to get customers. Get them, maybe, but not keep them. For that, you need a service that works.

The number of songs you can get out of even thirty-five thousand words of prose is finite and I must be getting close to the limit now. I've done over fifty already so I really can't complain. I love almost all of them, too. I listen to them all the time, to the exclusion of just about anything else. How long that will go on will be interesting to find out.

It could be a while. At present, I'm making them so fast I forget the ones I did a while ago, so when I listen to playlists I've made I keep surprising myself. I also have a second, longer, completed novel from the same time that I will almost certainly start mining when I'm done with this one, so I'm good for a while yet. 

Just to make it even less likely I'll stop, Suno recently added a feature that's absolutely perfect for me. It lets you direct the AI to make new songs using only the songs on a given playlist as reference. That means you can control the outcome with a great deal of precision and end up with a whole lot of songs that sound like they were made by the same imaginary people. 

I'd probably never take the headphones off again if the damn thing worked. It's in beta and so far it works just well enough to let me hear how perfect the output would be if it didn't glitch and scratch and skip and pop constantly like a vinyl album someone dropped into a deep fat fryer then tried to play. God willing they get it fixed soon. It's so frustrating it makes me growl.

Working with The Final Line so extensively over the last few weeks, though, has made me keener than I was back when I wrote it to do something with it. What, I'm not sure. These days, it's incredibly easy to self-publish but I'd have to finish the thing for that and the main reason I didn't do that the first time was because I have no idea how it's supposed to end. I have a slightly better idea now though...

Buddy. Last name pending.

Other than actually publishing it, I could put it up on a website or a blog or turn it into an audio book with AI reading it. I did once try to read it myself. I have a couple of chapters on cassette. It's very hard work, though, reading that much prose aloud. No wonder people get AI to do it. 

There was a very revealing article at GamesIndustry.biz yesterday, quoting extensively from voice actor Jane Perry's keynote speech at Develop:Brighton 2025. I recommend reading the whole thing but I was struck by her observation that generative AI has already taken a big bite out of voice actors' potential employment because the profession isn't all about exciting movies and games. A lot of it is less  romantic and most likely quite tedious work on things like "audiobooks, narration, corporate videos, e-learning, localisation".

Leaving aside the enormous amount of very badly voice-acted "localization" I've suffered through in various F2P imports, something for which AI replacement could only come as a blessed relief, that's a pointer to the kind of work that won't just be taken by AI tomorrow but that's already being taken by it today. I believe in the realms of self-publishing, authors quite commonly read their own work if they want to sell an audio version, though, so using AI would only be taking work away from themselves.

It won't be any comfort to anyone who used to make a decent living voicing incredibly boring corporate videos but in this respect AI isn't so different from the endless march of technology through the ages. When I got my first job working for an insurance company back in the early 1980s, if I wanted to send a letter, which was a big part of my job, I had to read it into a Dictaphone, walk down the corridor, hand it in to the woman in charge of the twenty typists sitting in the typing pool then wait a day to get it back, typed up, occasionally with errors that needed a second stroll up the corridor. 

By the time I left that company two or three years later, I had a PC on my desk and I was typing my own letters and printing them out on the printer next to me. The typing pool was gone. Jobs, like games, have their time then leave. Or they would if no-one kept making petitions to stop the flow of entropy. 

One thing I won't be doing is getting ChatGPT to give me some ideas, as Jeromai has been doing with one of his old stories. He's been writing a fascinating account of how he's been getting on, which I recommend to anyone curious about the process.

I'm far too prissy about my own prose to let any AI get its chrome paws on it so I'm stuck with doing it myself. We'll see how that goes. I have no such compunctions about the illustrations, though. I'd like there to be some pictures and I can tell you no-one wants to see anything I'm ever going to draw so it's either pay someone or get an AI to do it.

Cado Babe Under Glass

The idea of paying someone is intriguing. I've seen a few examples of illustrations various bloggers have commissioned and they've been pretty good, by which I mean they seem to realize the intentions of the commission quite accurately. I wouldn't rule that out for the future but for now, while I'm just in the playing around with ideas stage, AI is the inevitable answer to my curiosity about what my characters might look like, outside my head.

I am lucky enough to have one superb illustration already. Back when I was serializing The Final Line in the apa, one of the other members, professional comics artist and colorist Steve Whitaker, now deceased and very much missed, was so taken with one of the characters, he drew a picture of her and gave it to me. I've had it on my wall for thirty years.

This morning, I tried uploading that image as a seed so I could get an AI image generator come up with some variations. The results were not impressive. In fact, they were awful. I should say upfront that I was so annoyed with the first two that popped out, I gave up on the idea immediately, so it's very possible some more effort on my part would have seen some improvements but really I couldn't bear to go on.

Instead, I went back to the good old ways and did some text prompts instead. I wrote some character descriptions, used those as prompts, tweaked them a little as I went along and ended up with a set of images of the five core characters I'm fairly-to-very happy with. You can see those here today.

I won't go through the entire methodology in detail. I may do a separate post about it because I find it interesting even if no-one else does and it would be good to have a record. What I will say here is that I get by far the best results from the cheaper, supposedly less-sophisticated models than the expensive state-of-the-art super versions.

NightCafe is a website I often mention when I'm talking about AI and it's another I very much recommend. It has a generous free option, which is all I've ever needed, but it also regularly hands out free tokens for the upmarket AIs that otherwise require you to subscribe. Not that I recommend those, for the reason I just gave...

NightCafe has always handed out five tokens a day just for logging in but they recently gamified that into something any MMORPG player will immediately recognize as "a daily". Now, you still get a minimum five free tokens a day but you have to "create" something to get them and now you get an RNG roll to see how many tokens you've won. Five is guaranteed but you can get all sorts of numbers above that. And you get a fat reward for completing a "streak". I got 25 tokens for doing seven days in a row recently.

 Because I've been diligently collecting my free tokens every day for ages, I have almost three thousand in the bank. Again very generously, they don't expire, so there's no pressure to use them and I only bother with making AI pictures when I have a reason these days, not for fun like I used to. The fun in doing that wore off a long time ago.

Rachel Sunday and Sheba

I spent about two hours at NightCafe this morning. That's how long it took me to get satisfactory images of five characters. The time wasn't spent evenly across all of them. Looking at the images, all of which are handily stored indefinitely on my account, I see the first character took me just four tries, the second twenty-two, the third five, and the fourth and fifth both four goes each.

Guess which one took four times as long as any of the others? Yep. The one I already had a picture of, done by a human artist. Using that as a starter was a complete disaster but trying to get anything even a tenth as good by creative prompting wasn't much better.

And that's because I had an actual image with which to compare the results. Benchmarking the output of an image generator against a mental image is a lot more forgiving than comparing it to a real, high-quality picture.

Even leaving aside the personal connection, which honestly did color my reaction a little, the hand-drawn image just has something none of the AIs have. Life. It doesn't even look all that much like the character I had in my head back when I created her but ever since Steve handed me his idea of her, that's who she's always been.

Which isn't to say the AI versions of the characters are bad. Not at all, They're pretty good and I'm pleased with them. They're satisfying because I was able to nudge the AI into giving me something that closely resembled the image in my head, though. Steve gave me something he saw in the character and made me see it, too. AIs can't do that. Yet.

Which is why I don't think artists have much to worry about, aesthetically. They're always going to be preferable to and better than AI, at least until AI becomes truly sentient, at which point I would guess it will have better things to do than knock out commercial art samples and pictures of waifu by the gazillion.

At this point the post started to spiral into a general rant about AI and we've surely all had more than enough of those by now. I exercised my right as a human writer and deleted the rest of it. Another thing we can do that AIs can't. Exhibit judgment. 

I'm off to play with that potentially game-changing, practically infuriating new feature in Suno for the rest of the afternoon. Wish my blood-pressure luck!

PS. That's Sally Mandragora at the top, in case you were wondering.