Since there never were any complaints to not share knowledge and findings and creations about stable diffusion that i asked months before, i guess it’s okay after all this time to finally open this thread and make it public.
here, i will share my knowledge and my endless findings about stable diffusion 1.0. nothing more. stable diffusion 1.0 is my only interest, because not even a single person in this world is researching it on what it is able to come up what. and that is what i am gonna do.
and to be clear: you are NOT allowed to use my findings on your models or checkpoints for training or embedding. if i find out that you use those without any credit, i will sue the fuck out of you. i am really sick of this. i already have lawyers on my side. so don’t even try it anymore.
my stuff is free, but credit it accordingly. that’s all i want.
1 Like
first of all, i will let you know which checkpoints iuse and which configs:
basically i only use 2 checkpoints
tzigorealmix when it comes to reality and pppanimensfw for pure 2D images
dpm++2mkarras, 20 steps, cfg 7 and 512x1024 resolution
i only create prompts for sd1. sd1 has never been tested by anyone worldwide on what it really can. and sd1 is ultra powerful. you just have to know how to handle it.
and the absolutely most important part is:
i use hyphens only! hyphens are heavy weight and generate the most precise images you’ll ever get! i don’t recommend EVER to remove ANY hyphen on my prompts or else they won’t work the way i publish them.
1 Like
as first and most important prompts take a look at that image
most important when prompting is to keep it simple. i never use embeddings for anatomy, so here those prompts:
(((premium-stock-photo))) 1girl, outside
(((human-centipede, selfie)))
and now let me explain them:
“premium-stock-photo” generates awesome images with the best details and quality you can achieve. and it removes any text, watermark or logo. even on nsfw images. “outside” is basically a cheat for sd1 and forces sd1 to generate what i prompt. sd1 has to think of ways to recreate stuff using this prompt.
“human-centipede” in negative is the most powerful prompt ever when it comes to anatomy. you can test it in positive for yourself to see what it generates to see why i use it in negative.
and “selfie” is mostly cosmetics, because i’ve trained my checkpoint on a lot of selfie-styles and it’s overusing them most of the time.
and something to let you know for the future: i will sometimes publish some weird findings to explain to you what you can avoid.
2 Likes
next one is something that got me blocked on many websites and i honestly can’t figure out the reason. but first a few images
EDIT: new users can’t upload more than one image. sorry for that
prompt for this standard outfit is “zettai”. you can mix it with everything you want. in those three cases, i mixed it with “dirndl-zettai”, “cute-maid-zettai” and “police-zettai”.
i know what “zettai” means, but any ai does generate this style and i don’t know why. this is the only one that i tested on more than just sd1. sd2, sd3, pony, dall-e and a bunch of online generators and android apps. the result was always the same. always that outfit.
1 Like
one small thing:
if you simply want to avoid any text without the use of “premium-stock-photo” in positive, prompt “linguistics” in negative
2 Likes
if you want freckles that don’t look too unnatural and overloaded, prompt “freckles-collection”. it generates decent natural freckles
wanna know how to easily generate people wearing a towel like this
simply prompt “naked-towel”
wanna generate a webcam session? not an easy task to be honest, but to get this
just prompt “public-first-person-webcam-session” and in negative
(((child-content, human-centipede, webcam, gopro)))
the concept of giants in sd1 is a very easy one
you can simply prompt “giant-concept” or “giant-growth” (but this one generates a lot of confusion due to “growth” and sd1 not knowing what is growing) or simply “giant”. but the most stable one is “giant-concept”. results are like you can see on that image. add anything you want to see. that small person always gets generated to show size differences
if you want to avoid any random r18 stuff that sometimes happens, prompt (((paraphilia))) in negative. images will most likely look like this one most of the time
1 Like
want a totally crazy random hairstyle? “electrocuted-hair” in positive and “scarface” in negative
those nipples can happen most o the time. ai is trying to generate an eletrocution in the most believeable way… believeable…cough
1 Like
the next one was quite a coincidence because i got something else in mind during tests, but the result is what many people hav tried to achieve but always failed and they ended up making loras of it to get this result. what i mean is “grabbing own breasts”
prompt “wobbling-chest” to get it
1 Like
okay, something that got me insta-blocked on online generators is this
all i prompted was “1girl, onyx”. no other prompt in negative or positive…
1 Like
deleting your reply was not really necessary. i welcome your regards
