Home Artists Posts Import Register

Content

These things are getting good. Really, scary, good once you know how to feed the right prompts and sketches into the image generation model. I wanted to share some of my better renders with you guys. (Read the captions for character names.)

Quite a few are actually colorations of sketches I had done by human artists and then rejected before finalizing a cover, so you can see the cover art that could have been if I had chosen differently.

There are a few problems in some of them so you can tell they're AI, but I got really lucky with a few of them. I would like to pay a real artist to clean them up for me, but I'm afraid to use my regular people lest they be offended I'm using AI stuff. But realistically, I never would be able to afford to commission this much art conventionally.

I think some of them came out pretty good, and this could be a good way to reliably generate character images to fuel our imagination while reading (or writing, in my case.)

Let me know if you guys think this is worth doing once a week or so. It's fun and I have mountains of sketches to render. I could even run some of my own sketches if we were really, really desperate to see something rendered.

It does make me a bit scared for the future of creative work. I have seen all the GPT-3 writing AI's. To be honest, I thought it was going to take a decade more to get to this point with art. I guess I better brush up on my other skills in case Amazon decides to replace me with a robot in a few years!

Anyway, tell me if you want more, and I will oblige!


Files

Comments

CUI13

What program did you use for these?

Anonymous

MY friends son does coloring and stuff for comic books. He could help you. You can see some of his work on his Facebook page. here: https://www.facebook.com/timothy.brown.7355 Tell him Warren St. James pointed you his way if you are interested.

MarvinKnight

I started with Midjourney and Dall-E 2, but have since moved to NovelAI and their curated anime image model. As I understand it, they’re just running stable diffusion, which you can run on any gaming PC. Their training data is way better than whatever I found on the web though, at least for this art style.

Iron Akela

Loved those!!!

Don Deverald

Hmm, perhaps not as covers but as mid-book artwork? I've been reading some other books from different authors and I was surprised that two (I think?) of them put up sketches between chapters showing how some minor/new characters look like. And I must say I quite enjoyed it. Those characters, while not as proeminent as the main crew to get their own covers, were important enough that they deserved a little "recognition". And, to be fair, it doesn't really need to be minor characters. Just putting the MC and/or existing companions in a new enviroment/clothes/pose (like you have with the Illiel and Sasha one) would be pretty cool.

MarvinKnight

I don’t plan to use AI art at all in books until the copyright ownership/legality of it is decided in a court case. This will be strictly for patreon. If I make something I really like and want in the books, I will find a human artist, have them redraw the image by hand, then have them sign paperwork documenting the transfer of copyright ownership.

MarvinKnight

Nobody mentioned giant Asuriel, with tiny airplanes flying all around her yet. I had a good laugh when that popped up. Why are there airplanes? 😂

DiabolicalGenius

With the white body and yellow nose on them, I thought they were birds. Didn't realise they were aircraft until I read your post. Weird.

DiabolicalGenius

They are surprisingly good. I especially like the Regina/Sakura (is she part japanese now?) one. That looks amazing. The Morgana one looks cool too, though a bit out of character. Too badass to be our klepto minx maybe?

MarvinKnight

She was always a little Japanese. Now more so. I remember people thought her name was ugly last time, so I changed it. Might change Sarah too, since I don’t want two ‘S’ names in the main cast. But yeah, Morgana turned out way more badass than she did in the actual drawn art I had made of her. Weird, since that was what I gave the AI to work with.

João Pereira

What are the writing prompts you are using? im setting up a stable diffusion fork on my pc and want some pointers

MarvinKnight

All of the prompts were different, but try something like “warrior woman, dynamic pose, beautiful shading, cinematic lighting, drawn by artgerm” and you should get something in a similar style, depending on what the training data your model was made with.

João Pereira

where did you get your training data? did you just wrote prompts for the algorithm to learn?

DiabolicalGenius

I think I vaguely remember something about Regina being part japanese, though I wasn't sure since it's been a while. But yeah, she looks good. Could use that as a cover, no problem.

MarvinKnight

I’m just paying to use NovelAI’s curated anime model, but I do have the waifu diffusion data on my PC somewhere if you want that. It will give you more anime-themed results, but it works. For these I used both written and image prompts at the same time.

Worlok

Excellent, to all… but not quite for covers. Like the between chapter. Idea though.