It's not that AI doesn't take any learning or skill.
There's something to be said for having developed an artist's eye and understanding comp, this will improve the art. Or rather, it will allow you to select the best of a bad bunch and merge them more successfully. There's also something to be said for having learnt the prompting tricks and quirks of a program. It's not hard, but you do have to play around with it a bit. And with that, you mask and merge and iterate and get something servicable.
But this hasn't really been done yet. If you're going to do ai art, clean it up. Sketch-inpaint out the featherduster. Mask everything out and make it do a few iterations for a better hand. Clean up the bikini strings. And decide if you want goggles or cogs, run some iterations on that too.
And, imo get a model and/or lora that looks more visually interesting than the generic 'sakimichan' style. The B&W one you posted in the ai chanel with the paint splatter background is far more visually interesting imo, with far more interesting lighting bouncing off the clothes https://communities.win/c/ArtificialIntelligence/p/17rmOCIEgT/ . Yes it has it's problems too, a zipper on the skin, the whole shadowed belly area. But it's got a lot more going for it as art imo.
Beats bean style cal arts and modern art shit, but that's a low bar to pass.
Not OP, but I agree with what you said, so I did it. A single img2img to clean up the whole thing in general but keeping the anime style since I assume that was intentional. Then some iteration on the hands and one on the hat to make it less ridiculous.
Some notes of my own for OP: I don't know what you're using to generate, but your results don't seem great. Here are my first ten, totally unmodified results for "blonde steampunk girl at beach in bikini, anime style", and here are my first ten for the same thing, but "photorealistic". Get a model that looks good at https://civitai.com/ and read some of the prompts people use there. Also, AI still gets hands wrong a lot of the time. Especially if they're holding something like a gun. But it can be pretty good at faces and figures.
I think the way it works is that it broadly understands concepts like "gun" has "gun stock", "gun barrell", "grip", "magazine", etc, and it had a vague idea of where those parts belong in relation to each other, but it doesn't truly understand the concept. It's similar to how it clearly understands that hands have fingers, but it has no idea how many fingers should go in the finger space of a hand, or why you get extra arms sometimes, etc.
My biggest issue is that i'm only using the free version so I don't get 4 results to choose from.
Stop using midjourney, download stablediffusion https://rentry.org/voldy#-guide- , learn to use an actually customisable and editable ai art program. This is how you download and install SD 1.5 (xl is fairly easy to add after, ignore sd 2). It's 1.5 and XL you'll want.
The cogs kind of look like goggles, the black strap is circling around the hat like a headband for them. So they normally would be facing forwards, but fell sideways at some point. Cleaning them up into some kind of goggles might fit with the bikini aesthetic, someone clearly near an aquatic environment.
Terrible. The eyes are weirdly watery, the feathering smudge around her hair is ugly, she has a double phalange on her pinky finger, a bizarre purple nail on her ring finger, she has the teeth of Jaws from the James Bond films, whatever that telescope thing next to her is apparently vanishes once it gets past her arm, one of the dangly bits on the bikini bottom is disconnected so the last little bit I guess is just stapled to her leg.
The nicest thing I can say is that it makes me feel better about AI's chances of stealing our jobs.
A lot of the problems with this can be fixed using a home Stable Diffusion setup with specific plugins and tools and Photoshop (or any other drawing program that supports layers, so not MSPaint). It's virtually impossible to get it to draw something as complex as "blonde woman in a bikini wearing a steampunk top hat holding a gun with goggles on her hat", but you can (relatively) easily get it to draw "blonde woman in a bikini", then use sketching and inpainting to add "a steampunk top hat", "holding a gun", and "goggles on her hat" one at a time. There's an extension for the Automatic1111 SD frontend called ADetailer that's specifically meant for fixing faces and hands (it's very good with faces, dice rolling with hands) that would fix the crappy eyes and mouth on this one without much work on your part.
It's not that AI doesn't take any learning or skill.
There's something to be said for having developed an artist's eye and understanding comp, this will improve the art. Or rather, it will allow you to select the best of a bad bunch and merge them more successfully. There's also something to be said for having learnt the prompting tricks and quirks of a program. It's not hard, but you do have to play around with it a bit. And with that, you mask and merge and iterate and get something servicable.
But this hasn't really been done yet. If you're going to do ai art, clean it up. Sketch-inpaint out the featherduster. Mask everything out and make it do a few iterations for a better hand. Clean up the bikini strings. And decide if you want goggles or cogs, run some iterations on that too.
And, imo get a model and/or lora that looks more visually interesting than the generic 'sakimichan' style. The B&W one you posted in the ai chanel with the paint splatter background is far more visually interesting imo, with far more interesting lighting bouncing off the clothes https://communities.win/c/ArtificialIntelligence/p/17rmOCIEgT/ . Yes it has it's problems too, a zipper on the skin, the whole shadowed belly area. But it's got a lot more going for it as art imo.
Beats bean style cal arts and modern art shit, but that's a low bar to pass.
Not OP, but I agree with what you said, so I did it. A single img2img to clean up the whole thing in general but keeping the anime style since I assume that was intentional. Then some iteration on the hands and one on the hat to make it less ridiculous.
Then did a second img2img to make a photorealistic version.
Some notes of my own for OP: I don't know what you're using to generate, but your results don't seem great. Here are my first ten, totally unmodified results for "blonde steampunk girl at beach in bikini, anime style", and here are my first ten for the same thing, but "photorealistic". Get a model that looks good at https://civitai.com/ and read some of the prompts people use there. Also, AI still gets hands wrong a lot of the time. Especially if they're holding something like a gun. But it can be pretty good at faces and figures.
I feel like AI physically can't understand guns because it's referring to so many drawings where the artists don't understand guns either.
Then you see it make a gun and it just baffles the shit out of you.
I think the way it works is that it broadly understands concepts like "gun" has "gun stock", "gun barrell", "grip", "magazine", etc, and it had a vague idea of where those parts belong in relation to each other, but it doesn't truly understand the concept. It's similar to how it clearly understands that hands have fingers, but it has no idea how many fingers should go in the finger space of a hand, or why you get extra arms sometimes, etc.
What image AI are you using?
Stable diffusion with JuggernautXL.
System specs?
I'm using a 3090, but that is definitely not a requirement. Other system specs don't matter much.
Almost bought a 3080, when they raised the prices again (lol) and the announcement of the super came.
Beautiful work. (compared to what I've been able to do at least)
So those are goggles lmao. I wondered what those round things on her hat were supposed to be.
Stop using midjourney, download stablediffusion https://rentry.org/voldy#-guide- , learn to use an actually customisable and editable ai art program. This is how you download and install SD 1.5 (xl is fairly easy to add after, ignore sd 2). It's 1.5 and XL you'll want.
Lazy, cursed garbage. Why is this even here?
Right hand belongs to an elite from halo.
WORT WORT WORT
If you want to make AI art, go actually learn stable diffusion, research workflows for auto1111 and comfyui, including img2img and inpainting.
Most importantly, get good enough at art that you can spot glaring errors, push it in the right direction by sketching, and touch things up.
She kind of looks like the chick from the new Game of Thrones.
Naw she’s almost identical to Kate Upton sans freckles/ marks
Rhenyra, the younger, non tranny version.
Diffusion I'm guessing, which version?
She's 1/16th orc for the teeth, make it canon.
And nevermind the feather duster on the right, what purpose do those two massive things to the left serve?!... I mean really, why do hats need cogs.
The cogs kind of look like goggles, the black strap is circling around the hat like a headband for them. So they normally would be facing forwards, but fell sideways at some point. Cleaning them up into some kind of goggles might fit with the bikini aesthetic, someone clearly near an aquatic environment.
Adventitious feather-duster for purposes that spring easily to mind in the context of the whole piece.
Also, there's a certain down-to-earth charm about imperfect teeth when there's so little else to complain about.
Terrible. The eyes are weirdly watery, the feathering smudge around her hair is ugly, she has a double phalange on her pinky finger, a bizarre purple nail on her ring finger, she has the teeth of Jaws from the James Bond films, whatever that telescope thing next to her is apparently vanishes once it gets past her arm, one of the dangly bits on the bikini bottom is disconnected so the last little bit I guess is just stapled to her leg.
The nicest thing I can say is that it makes me feel better about AI's chances of stealing our jobs.
A lot of the problems with this can be fixed using a home Stable Diffusion setup with specific plugins and tools and Photoshop (or any other drawing program that supports layers, so not MSPaint). It's virtually impossible to get it to draw something as complex as "blonde woman in a bikini wearing a steampunk top hat holding a gun with goggles on her hat", but you can (relatively) easily get it to draw "blonde woman in a bikini", then use sketching and inpainting to add "a steampunk top hat", "holding a gun", and "goggles on her hat" one at a time. There's an extension for the Automatic1111 SD frontend called ADetailer that's specifically meant for fixing faces and hands (it's very good with faces, dice rolling with hands) that would fix the crappy eyes and mouth on this one without much work on your part.
Don't worry about the AI. Use it as a foundation. Learn to draw what is pleasing to the eye, then keep going from there.
Think of it as training wheels or something.