Me too just learning the ropes, working on my first sets of potential employees
. I have automatic1111 installed, and some quite realistic models - SD1.5, SDXL and Pony too, I'm experimenting with them, all could make good images, some better at nudes, same better at clothes. Yeah, EpicPhotoGasm is one of them.
First I generate some txt2img pictures, either with random generated faces or with some celebrities name in the description (or with Loras of these celebrities, but not too much success).
For SD1.5/SDXL, I try something like this for a young (18+) random girl: Positive: full body picture of a woman, standing, nude, high heels, gentle smile, white background, brown hair, small breasts, skinny
Negative: teeth, big breasts, teen
Or: Positive: full body picture of 1 women, 60 years old, standing, nude, thick, sagging breasts, small sagging breasts, high heels, photorealistic, Helen Mirren, blond hair, gentle smile
Negative: long legs, thin, teeth
Steps: 30, Sampler: DPM++ 2M Karras, CGF Scale: 7, 512x512 (upscaled to 1024x1024)
(that one was even made with EpicPhotoGasm - this model still gave me quite big tits, but at least natural looking ones, not fake basketballs)
(Bigger chance you get a standing girl form toes to head, if you have some kind of shoes decription in your prompt... otherwise SD tends to cut her somewhere over the knees.)
Let's say after some generations I got one good result I like: face, hair, body, posture. There could be some deformations (mostly fingers), so usually first I try to repair them with impaint.
Impaint is your friend.
When I finally have the good nude (the 01 picture), with using implant again I give her clothes. First panties (picture 02), then to this picture a bra (picture 03). Again, to picture 01 a swimsuit, and so on. It's quite a trial and error. Well, lots of trial and error. Less denoising don't changes much the body and the posture, but more denoising is more creative. There are lots of thing to experimenting. (For this one, original, fill or latent nothing? What about higher CFG Scale? Hm, maybe this other model can give me a playboy bunny dress, let's change to it.)
With higher CFG Scale and higher denoising you have better chance to get the color and clothes you want - and much higher chance to get an utterly deformed image, with a new face on the chest, a third hand, etc.
What is good in impaint, that you can leave out anything, that you don't want to change - and it won't change. Like the face, the hair. (Whell, shoulder-lenght or longer hair is still a bitch. And fingers too close to body.)
***
As they said, you have to remove the background. Either in a picture editor program (I use Gimp, it can work on the PNGs from SD), or automatic have (at least one) background-removing extension. Still have to learn them to use, so will see, witch is works better, giving smaller halo around the head, hair.
Some images I've made with SD and I really like, still has some little imperfections, so I'll try to repair them in Gimp; I have to use Gimp anyway, to resise and crop these 1024x1024 pictures to 512x800.