Post Generative/AI Art You've Made

Admittedly, natural-looking people are hard to replicate in any artform.

Are you receiving this?◊ It's got More Teeth than the Osmond Family. It's got Corinthian eyes. And a mouth. And a mouth under that mouth. And a mouth in its nose. And oh god why did we even try to create it.

Edited by NotSoBadassLongcoat on Dec 24th 2022 at 3:25:19 PM

It asks for a save prompt. Dude, just... upload the images here or copy them to an image hosting site or something. Or even discord. <_>

Yeah, but it seems strangely better at doing anatomy when you DON'T ask for a portrait, it's pretty odd.

Will be interesting, given a year or two of people obsessively trying to fix the 'arm merging into side' problem and the like, what the underlying tech's going to be.

That link forces us to download it. Could you upload it to imgur or something after taking a picture and saving it? Other options could be posting it on Discord and linking to the url here. At least ones I know of. I know you can directly paste an image to Imgur now, so you don't even have to save said image on your computer.

Continuing to use waifu diffusion as a model.◊ It's definitely something that benefits very hard from more prompts rather than less.

does it work now

That's a lotta teeth.

That works. And it's wonderful.

(If it wasn't obvious from my other posts, my casual prompting tends to turn into "what horrifying things can I make")

Huh. Neat. Really cool design overall(that's to the teethy one).

The other girl just posted looks really nice.

Edited by Irene on Dec 24th 2022 at 8:34:27 AM

Heads half out of frame is strangely common.◊ And I've noticed a funny tendency for somewhat better anatomy◊ to have derpier faces and vice versa.

As for that last one, it became an input for this◊ (with a few tweaks to the prompt), which was used for this◊ one, and then finally this.◊ Using it as a sort of incremental refinement is kind of neat.

Of course, doing so by generating ten or so images at a time repeatedly... yeah, glad to be doing this locally with a pretty hefty GPU.

Tip: If the program your using has them, and you haven't already done this, look for the "negative prompts" section and input "cropped". That should hopefully help with the head problem.

It sadly didn't fix it all that much, I tried.

this one was I 2 I on Stable Diffusion, input was a selfie i took on the beach.

Ooooh, nice. I'm particularly curious what the actual inputs looked like (aside from the selfie); a stylistic transformation like that which doesn't dump the entire image is something I've got no idea about.

I decided to generate 50 images off the same prompt and input image. Made in a game, that input.◊

Edited by RainehDaze on Dec 24th 2022 at 3:50:30 PM

It seems to really like moving her hands around.

Yeah, I didn't specify anything about pose or hands, so I guess it kept rearranging those to fit.

Tried to see how close I could remake your basis. Stable Diffusion doesn't like mixing colors. The last word is "viewer". Second line is negatives.

Seems like I'd need like a whole paragraph to get it real good.

Edited by Malady on Dec 24th 2022 at 8:00:37 AM

one thing ive also noticed about Stable Diffusion is that it doesn't differentiate characters well; if you give it two character names, it creates one person with characteristics of both those characters, not a picture containing both of them.

the prompt for this one was "Hatsune Miku playing chess with 2B".

Ah, the magically appearing/disappearing leg.

Three results from my Lexperiment. I fed an old drawing of my RPG character into mage.space's Stable Diffusion engine, and started yanking the lever.

Several rewrites of the prompt, slider adjustments and negative prompt clarifications later, I ended up with a couple of reasonably looking renders and one that was too funny not to share.

- Such a weird prompt concept, that lets it do *that*~! Well, I guess the silhouette keeps it a girl?

Stable Diffusion, even if I tell it about pockets, doesn't guarantee pants. nnn

Edited by Malady on Dec 25th 2022 at 11:29:01 AM

First, the text prompt was "pale woman with dark hair, black tanktop, pierced lower lip, tattooed arms, olive cargo pants, background is industrial street at night" as an addition to a drawing. Second, I used inpainting not knowing what it was.

Somehow, I tried to make StableDiffusion draw nudes, and failed. The results were all horribly misshapen and having Barbie Doll Anatomy at best. Unless mage.space doesn't have enough reference for those in their dataset (as opposed to NovelAI, which unsurprisingly can and does render naked lolis, as I already saw on Pixiv).

You did turn off the NSFW filter, right? I got pantsless-es where their not-safe bits were all tattoo-ed.

I can't tell if my "tattooed, lip stud girl, hands at her waist, hairbun, waist-up" with anti-"cropped" is only giving me 2/4 images because the others are NSFW, or I'm taxing the system too much.

Yes, I did. With the NSFW content filter on, it wasn't even generating that. But notice that my text prompt expressly included pants in a specific cut and color.

Edited by NotSoBadassLongcoat on Dec 25th 2022 at 9:07:30 PM

I'd like to note that discussing how to make the best AI nudes on TV Tropes may not be the best idea.