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.
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.
What image AI are you using?
Stable diffusion with JuggernautXL.
System specs?