If training your own model with just one A6000 is possible, then eventually it will be possible with consumer graphics cards. Imagine putting in 10 hours of footage from your favorite anime or tv show and training images solely on that. We could write our own graphic novels and fanfictions if we wanted to!
What are some prompts to get a girl semi-naked rather than just naked? Exposed doesn't work that well, "exposed breasts" gives her gigantic milkers for some reason
Chase Ortiz
those cavemen bully themselves by discarding a useful tool as if it's some sort of crime against nature. let them scream and throw shit at the canvas like the apes they are
Is there a better model out there for nude women? It looks like the model hasn't seen many naked women. The parts are in the wrong places, out of shape, and the arms are in the wrong places, etc.
Matthew Phillips
Wrong quote, i mean for
Levi Powell
that one was Arthur Hughes and Alasdair McLellan and a bunch of generic prompts after it
if the jannies don't strip the metadata, all of my images should have their prompts in them for the png info tab. otherwise: retro anime, close up of girl standing in a field Negative prompt: large breasts, adult, mutation, mutated, (extra_limb), (ugly), animal, text, bad anatomy, ((((deformed)))) Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 4214474430, Size: 512x512
Can you draw a pose, and then have the AI finish the rest of the drawing and color it?
I would never have to spend any more hours trying clean up lines, paint, erase, forget that I'm painting on the wrong layer, redraw anatomy/proportions I messed up, etc. This software would be a dream come true in 5 years or so. Amazing.
Joseph King
is there any way to stop Voldy's from trying to display all the images in the browser? It gets so slow when I've made 100+ images and try to interrupt to change the prompt
Parker Jenkins
Yeah. I started to make the initial txt2img 512x832 or even lower height and then outpaint at the bottom to make it 512x1024 and then start doing modifications.
I got 99% grotesque multiboob nonsense from including Donato Giancola in my artist list, so whichever user recommended him to me yesterday, you were wrong. However, the 1% it got right, it got very right.
In the future, for sure. Also, there will be different SD branched models also for specialized things like poses, hands, feet, etc.
Hudson Wilson
my post vindicated again: specifically, > . . . the best prompts produce 99% garbage and 1% gold, versus a more belabored 'perfect' prompt producing 60% garbage and 35% meh and 5% ok-I-guess.
Adam Lee
upscaled? or did you set it that high of a resolution.