Missed Part 2?

This article is part of the "Making Of" series about "Urlan, Cosmic Cat" and features some of the weirdest, funniest and/or more interesting unused images generated in the rollercoaster process of creating this incredible AI-illustrated graphic novel.

You can check out a free preview of the actual work, grab it on Amazon or watch the Launch Trailer.

Heads up:
this article contains some spoilers!

When we had all the panels (or so we thought), now we knew we were facing the hard part. First, we downloaded them all. We had about 500 images. Then:

1) We divided them in two folders: “discarded” and “maybes”;

2) We created eleven folders (simple Windows folders): one for each scene we had in mind, and a “generic but good” folder with details like close-ups and such;

3) We went through the “maybes,” and each of them was put into the relevant folder; in this phase, we also made some basic cropping, but not much: we thought it better to leave it to the layout phase, when we would be able to see the panels on the page and decide how to crop them the best way;

4) We went through each folder, and gave a “speaking” title to each image; it didn’t have to be fancy, just explain what was happening;

5) We went through it all again and selected the best panels, that we would use in the final work;

6) We were ready for the following phase: laying out the comic in Clip Studio Paint!

… or not. Because we realized that some images we needed were missing. So we went back to the ChatGPT+Dall-E 3 dynamic duo and asked for some additional panels. This was tricky, because our conversation had become so huge and data-heavy that the ChatGPT interface just kept getting stuck. We had to put in a prompt – it got stuck. We refreshed the page, tried again – it gave us an error message. Reload, try again. Regenerate the answer. And so on a few times, until an answer came out, often truncated or with errors. Trying to open a new conversation and feed info about the story was a bloodbath.

The images, deprived of the context of the conversation, were just off. We got a couple of good panels as, in the meantime, ChatGPT had developed the feature to upload images to be used as source, so we uploaded some jpg layout examples. The rest were unusable so we had to stick with the old agonizing conversation.

Also, these extra images needed to be pretty coherent with what we had in mind because they had to fill narrative holes. We needed something precise, and the AI had moments when it seemed to have a hard time being consistent even with the characters and style – it spouted out CGI-like images, full-color panels, panels with human characters instead of cats and demons…

Laying out the comic was challenging. When you work with a human artist, they usually give you a finished page, not single panels. So, layout is basically in their hands. You just have to think about the book design if you’re producing it. This time, we had single panels, all done in different ratios – we weren’t strict about ratios with our ChatGPT prompts – and also Dall-E 3 doesn’t really listen if you try.

We had to throw a bunch of panels on the digital canvas – much like throwing some photos on a table – move them around, make them fit. Fortunately, CSP’s “Frame Border” tool is a breeze and creates a masking layer where you can operate on the panel easily and effectively. We started with a simple “square” layout in mind, but in the end decided to have some fun with it and make the story more dynamic and modern, also allowing us to cut off so many unwanted pieces of panels.

Finally, we got back to the dinosaur conversation for some final panels we really really needed and this was a hard fight with the AI, but we came out with what we were looking for.

Lettering was not particularly different than on a normal comic, technically, though most of the dialogue text was emergent as well. We didn’t have a pre-existing script, which the artist was aware of while drawing. So it was a little tricky as the panels were very full and rich, and with little dead space. We went for CSP awesome “balloon pen” tool, which allows to manually draw the balloon on your touchscreen or drawing tablet. This allowed for an organic look to our balloons, which was useful to make the best of the space we had, but also was extremely poetic – side to side, something so human as a hand-drawn balloon and something so inhuman as AI art. Yeah.

In the end, the comic was done, and we were excited with it. It was cool. Time to answer a few interesting questions…

Had it been easier than working with a human artist?

Honestly, no. Working with a human artist is easier, as you can count on common sense, and you don’t need to specify everything because so many things are obvious once you have discussed the concept and shared all the info about the project. An AI artist doesn’t think like that, so many things need to be reiterated, and many things just won’t work. You have to try and think differently, be flexible, improvise. A good human artist will draw exactly what you have in mind, probably better than you have in mind. An AI artist will not draw exactly what you had in mind. Weirdly, an AI artist feels more idiosyncratic than a human pro.

Had it been faster than a human artist?

Yes: getting this level of detail from a human artist will take significantly more time. In a couple of weeks, working the odd one-or-two hours a day on the project, we had the complete art. Of course, there was a lot of editing to do in post-production, much more than would take with a human artist.

Had it been funnier than working with a human artist?

Well, it was different. Working with a human artist is a fantastic experience when you’re tuned. Working with the AI can be frustrating, yet it’s exhilarating, and it really felt like a dialogue with an “entity”; it gave a great sensation of creative freedom, as you don’t have to negotiate with the artist’s own vision – something which often is the case with human artist. And reasonably so! Of course, negotiations with the AI happened as well, but for some reason the final result feels very much “ours”.

Had it been cheaper than working with a human artist?

Well, definitely yes: a human artist will get 50% of the royalties – which is right: writer and artist contribute about equally to a project – or about 75$-100$ per page (which is fair, and maybe a little more for images so intricate and detailed like those in Urlan), while the AI artist contents itself with a humble monthly fee of 20$.

In the end, we think Urlan is a very interesting and useful experiment, a cool showcase of the current level of AI, and an exciting experience. We’ll definitely do that again soon!

(Actually, we already have a few ideas for an Urlan sequel…)

In the next episode: Part 4 - A Gallery of Anomalies!

Urlan, Cosmic Cat

A groundbreaking AI-human collaboration in graphic storytelling. With yellow eyes.