I think fanfiction falls into the “personal use” bucket so you could use AI freely to create covers, character art, etc. I don’t think anyone will give you trouble for that as long as you’re truthful and not trying to pass it off as your own art. I honestly see AI art as a form of fanart but just like there are people who claim that fanfiction is stealing, there are those who make that claim toward AI art.
Everything changes once money is involved.
If publishing commercially (expecting money), then you’d have some decisions to make.
The AI space is still unregulated but there’s been some backlash from the art community against it. Current AI models were sourced unethically so even though there’s no law against it, I’d rather avoid the potential backlash.
All art I’m creating is for my original stories, but I’m not publishing commercially just yet. I’ll cross that bridge when I get there. Maybe by then we’ll have ethical models. Maybe by then public backlash calms down. If not, I don’t know, maybe I’ll use the AI art as inspiration to create something of my own because I don’t know if I’ll ever have the budget to pay an artist (unless by some miracle I make money from this effort). I’d rather save that money for an editor.
I find that multi character prompts often backfire. AI struggles when you give it too many instructions.
My process for controlling the output of complex images is to do a bit of legwork. More work but the results are better so it’s a worthy investment.
See what I did here to create a book cover that featured two characters.
For any prompt, I would suggest to start with these:
The subject (blond boy holding x)
Style (photography, cartoon, oil painting, etc)
Modifiers to get the right mood (sad, happy, epic, muted, colors of the background).
The order you put the words in matter. Put the more important words in the front.
Also be aware that using a child in your artwork means you have to be extra careful with your prompt or else it will get blocked by censorship. I fear to imagine what people were trying to create for them to ban so many innocent words.
It took me forever to get her eyes right. I tried perhaps 10 or more different eyes. For some reason, she has a very unusual eye shape because none of the stock photos wanted to blend in.
But finally I found the right eyes. Believe it or not, Natalie Portman has the right eyes for my character’s face, but I modified them so much, you can’t tell they’re Natalie’s eyes.
Later, I added color to them because they’re supposed to be amber, not just a plain brown and it paid off. She looks great.
I also had to blur out some of the flyaway hair because upscaling really doesn’t know how to handle hair. lol. Thankfully, the background is blurred out so it all blends in.
And just to show you what I’m doing with these, I put them in the character profile within my Campfire project. It’s so cool to have a face to go with a character.
Random news. Found my wish-list ‘entry level’ Professional DSLR (Nikon D780) at a modest $3,000 AUS yesterday afternoon. Then I wake up to these videos in my YouTube feed…
Oh, and a while ago Shutter Stock (where I posted most of my travel photos) announced a while ago that the hosted images will be provided to Shutter Stock’s customers as AI art prompts…Yet they don’t allow AI images to be sold on Shutter Stock…
Guess photography will be a recreational hobby for now. Can’t quite justify thousand-dollar expenses with diminishing ROI.
Mid Journey v5.1 was released a day or two ago. A few subtle improvements in the overall image quality of my old prompts…A big surprise being my v4 prompts work with a few changes, and the images are more aligned with the prompts too…Also, I’ve only had one instance of text / watermark ghosting in my portrait experiments, so that issue is almost fixed. How long until we get AI image software with +4k rendering abilities (zoom in to count the character’s (in-focus) skin pours)?
Some of the other MJ v5.1 experiments had better image quality / detail than my Motorola phone. But the Edge / Fusion is apparently renowned for its rubbish camera.
I had more luck when I took out the scrubs from the prompt. This one captures Ciara’s vibe best. I’m not 100% loving the quality though. It could be better.
This watercolor preset came out pretty well as it put her in scrubs at last. I really wish it would stop trying to generate a stethoscope. I could edit it out, i guess. Ugh. It’s hard to recreate the watercolor texture though so that would be a challenge.
Later, I tried to create a ghost jar. I describe it in the story as a glass jar that you put a candle and a little crystal into. Should be an easy enough object to create, right? Well apparently not. AI has a serious problem understanding that a candle inside a jar isn’t the same as a candle jar.
So a lot of evolving was involved in the process to basically hand-hold the AI.
Summary
Dream wasn’t giving me what I wanted so I tried NightCafe and created a few nice ones. The challenge here was the candle’s flame and the effects of glass. Shapes need to morph inside glass and AI didn’t always get it.
This one is the closest to what I was looking for. It’s a little boring though. I’ll probably try evolving it again later.
I’ve read that the EU has extremely tight creator / author copyright protection laws, probably better than Australia’s laws…Anything a artist or author creates, even if its a bunch of draft sketches or random scribbling, they own it. End of story…The current / future legal duels should be interesting. Oh, apparently some AI-software are now creating their own training data sets from scratch (wonder who owns the copyright for that).
Anyway. I’m finding that MJ v5.1 has a more restrictive and inconsistent censor / banned prompt function than previous versions…So many false positives (e.g. polarized filter)…This is becoming annoying when trying to mix and experiment with older ‘kosher’ prompt sets. Also, the MJ users’ prompts and images are always public / easy to access. Successful prompts might become more valuable than the images they receive (secrets of the trade). A local installation of Stable Diffusion (or Daz-3D) remains an appealing yet expensive solution.
Some of the best Freyja portraits I've conjured yet
I had some luck with today’s prompts. I was creating characters from my fanfic and it put me in such a good mood, I’ve been grinning all day. I bet my coworkers are wondering what’s wrong with my gloomy self.
After some experimentation, I settled on using two presets: Ink and Watercolor and tried to create a nice set of both for each character.
Ink is awesome but each one comes out in such a way that I’d have to edit it later.
And eyes always need fixing. As usual.
I have so many good ones to choose from. Which ones should I show off? Okay, just a few.
I can’t believe I forgot Merlin’s scarf! See? That’s what happens when I let this epic story sit for too long. I forget.
Anyway. I was able to generate one with Merlin’s signature scarf and I fixed his eyes with actual eyes of Colin Morgan, lol, because I can. I think I used a Color Dodge layer? I tried every layer option and this one came out best. And of course, AI can create a perfect face, even with freckles, and make the hair so smooth, I want to comb it, but it cannot to save its life create round eyes, so I had to do some drawing in there to fix their shape.
Upon second look, I’m kinda annoyed with how even these freckles are. I’m going to have to something to make them more random. Ugh. Why does it never end?
UPDATE: I fixed the freckles, kept only a few of them, and I did some serious work on the eyes since their shape was really weird. And I fixed the pattern on the scarf since it was bothering me how it just ended.
I found a new platform that aims to become a go-to for all your AI needs. Leonardo.ai.
It’s got a free tier and paid tiers for more tokens and it promises that you can create your own models (though I don’t know if that requires a paid account).
I’m on the wait list right now. I’ll let you know how it works when I get access.
Looking at the images people share on the discord, quality looks good, even more so, it’s pretty consistent, has fewer AI flaws (eyes look good!) so that looks promising.
