[Stable Diffusion] Prompt Sharing and Learning Thread

[Mr-Fox]

Does that sort of thing ever work for you guys???
Whenever I try to change the content of an existing AI generated image, the ai just slaps random shit into the result in the masked area if I set it to anything except Masked content: original.
If I set it to original, I just get the same stuff with more deformities even with denoising strength set to 0.75.

Spoiler: ss of my settings I guess

You don't have permission to view the spoiler content. Log in or register now.

Try change from "only masked" to "whole picture"

 

[HardcoreCuddler]

Try change from "only masked" to "whole picture"

it's even worse now Like the output doesn't follow the input in any way even if I set the denoising low (like < 0.3) and if I set it low enough, I just get the original image back with no change. I either get distored crap or the original XD
I also just tried filling with content from other parts of the image but that just makes a giant blob of the same color in the masked area. If I crank it to like 0.8 CFG it draws some lines through it but their incoherent.

 

[Mr-Fox]

it's even worse now Like the output doesn't follow the input in any way even if I set the denoising low (like < 0.3) and if I set it low enough, I just get the original image back with no change. I either get distored crap or the original XD

Make sure that you are not using static seed. Set it -1, otherwise it uses the same latent noise every time and of course you will not get a changed image or the result you are looking for. Just as with text2img, img2img and inpaint has it's own quirks that you need to learn by trial and error. In order to become proficient and get a good result you have to force yourself to use it often and experiment with it.

 

[HardcoreCuddler]

Make sure that you are not using static seed. Set it -1, otherwise it uses the same latent noise every time and of course you will not get a changed image or the result you are looking for. Just as with text2img, img2img and inpaint has it's own quirks that you need to learn by trial and error. In order to become proficient and get a good result you have to force yourself to use it often and experiment with it.

I never use the same seed, it's always at -1.
Buutt yea this isn't for me...I've already spend 30 min on just trying to get one single image out of it and I'm now going to go spend my time in some more productive way XD
Cheers

 

[Mr-Fox]

I never use the same seed, it's always at -1.
Buutt yea this isn't for me...I've already spend 30 min on just trying to get one single image out of it and I'm now going to go spend my time in some more productive way XD
Cheers

Hey I'm not a fan either in it's current form. But it does have it's uses. I think that's why ControlNet is even a thing because inpaint is so difficult to use and not predictable or consistent enough. If inpaint doesn't do it for you maybe ControlNet will.
I have to confess that I have not taken "the jump" myself yet but I have seen from videos and other's images that it is very useful.

 

[TitaniumDickDiamondBalls]

Bumped into this account on pixiv

You must be registered to see the links

. He gives detailed description of method he uses to create his works.

 

[miaouxtoo]

Where did you find them? Do you have the original png files? Those ones don't have the metadata you need to get an easy route to copying / cloning them
Interrogating the image though gives us this for the prompt that maybe you can use in a number of different Civitai models to see which fits best:

a sexy woman in a blue costume posing for a picture, hyper detailed and realistic, fantasy dungeon, plus size woman, doll, hyper realistic lighting, dark golden hair, featured on amiami, trending on artstartion, real life big mom, blue skintight closed longcoat, ellora

--
Refining that with a bit of trial and error (we can do this better if we keep going), we can have this:

Positive: (masterpiece:1.2), (best quality), (intricate details:1.2), (photorealism), wallpaper 8k, a beautiful woman, wearing sexy revealing detailed dress, hyper detailed, fantasy ornate castle window view, cleavage, wide hips, large breasts, sexy cloak, necklace ornament, (sheen to her skin), vibrant colors, (dynamic lighting:1.3), (vibrant cinematic lighting:1.3), (glowing lights and effects and lamps:1.2), lustrous golden hair, realistic skin, smiling, looking at viewer, facing front <lora:hipoly3DModelLora_v10:0.6>

Negative: poor quality, mutant hands, ugly, blurry, low quality, bad-artist, EasyNegative, facing front, boring, dull
--

That gives us something like the attachment - which we can see needs more lighting, and more colors... so you just keep adding to that in the prompt until you get something you need. If you have stable diffusion / automatic1111 - just throw the png into the png info tab and you'll get all the generation details.

I really like this ai artwork but i have no idea how to create anything like this in stable diffusion, pls help

 

[rayminator]

Don't know, it just generates penises for everyone xD Could not achieve a decent result.

keep on trying you get there i got it to work I know my preview wasn't great I did a fast test and those the ones that came up for me try adjusting the steps and the CFG scale and have Restore faces enabled, Variation seed, Variation strength

I just started to play around with AI myself too I don't know everything I have to play around with it to learn

try add this to the negative prompts

deformed penis, disfigured penis, extra penis

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

 

[Mr-Fox]

A couple more links about color contamination and the cutoff extension

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

 

[Nano999]

btw are ther any guide for promt formating?
It seems if you use "_" it will help in specifying the prompt, e.g.
blue_hair will make only hair blue, while "blue hair" will make blue hair, but might also blue cloth and something else

 

[StepSisWasNeverStuck]

Where did you find them? Do you have the original png files? Those ones don't have the metadata you need to get an easy route to copying / cloning them
Interrogating the image though gives us this for the prompt that maybe you can use in a number of different Civitai models to see which fits best:

a sexy woman in a blue costume posing for a picture, hyper detailed and realistic, fantasy dungeon, plus size woman, doll, hyper realistic lighting, dark golden hair, featured on amiami, trending on artstartion, real life big mom, blue skintight closed longcoat, ellora

--
Refining that with a bit of trial and error (we can do this better if we keep going), we can have this:

Positive: (masterpiece:1.2), (best quality), (intricate details:1.2), (photorealism), wallpaper 8k, a beautiful woman, wearing sexy revealing detailed dress, hyper detailed, fantasy ornate castle window view, cleavage, wide hips, large breasts, sexy cloak, necklace ornament, (sheen to her skin), vibrant colors, (dynamic lighting:1.3), (vibrant cinematic lighting:1.3), (glowing lights and effects and lamps:1.2), lustrous golden hair, realistic skin, smiling, looking at viewer, facing front <lora:hipoly3DModelLora_v10:0.6>

Negative: poor quality, mutant hands, ugly, blurry, low quality, bad-artist, EasyNegative, facing front, boring, dull
--

That gives us something like the attachment - which we can see needs more lighting, and more colors... so you just keep adding to that in the prompt until you get something you need. If you have stable diffusion / automatic1111 - just throw the png into the png info tab and you'll get all the generation details.


View attachment 2531032 View attachment 2531040

omg maiouxtoo you completely recreated the image from scratch without any generation data, that is really amazing and I really appreciate your work.

but if possible can you please create more images(10) like this so that I can get better understanding of this ai artwork.

please add these things to make the ai art look more better:-

-huge boobs
-wide hips
-small and slim waist
-different clothing styles
-different hairstyle
-pose
-more well defined perfect female figure
-little bit muscular
-more beautiful faces
-different backgrounds
-jewellery in accessories

here are some more images for reference (these are the photos I got from Instagram so there is no meta/generation data)



i would really appreciate your work, thank you and have a good day.

 

[MLocke]

omg maiouxtoo you completely recreated the image from scratch without any generation data, that is really amazing and I really appreciate your work.

I'm pretty sure the Perfect World model on Civit is what you are looking for.

 

[Nano999]

Any tips on a "guy kissing (licking) a belly of a woman" prompt?
I get a casual kiss no mater how hard I try xD

 

[miaouxtoo]

Use an anime / hentai model first - descriptors and lewdness on those is turned up to the max. You may have to run hundreds of them to get what you want.
Once you have what you want in rough generations, then apply the real model over it using controlnet if necessary.

Alternatively, find a real world porn image - use that in controlnet and good luck.

Any tips on a "guy kissing (licking) a belly of a woman" prompt?
I get a casual kiss no mater how hard I try xD

 

[miaouxtoo]

This is the learning thread - not the 'please make me more fap material thread'! Lol

That being said, maybe I'll have a go from curiosity in the next few days. HOWEVER, I gave you all the ways to make it ... and you are in this thread. I won't make all of them because I don't have a GPU, it takes me a few mins to get the high res ones.

See inline details

but if possible can you please create more images(10) like this so that I can get better understanding of this ai artwork.

please add these things to make the ai art look more better:-

-huge boobs -> "huge breasts"
-wide hips -- this is there already
-small and slim waist --> "slim waist"
-different clothing styles --> wildcard extension required
-different hairstyle --> wildcard extension for hairstyles
-pose --> "dynamic pose" / "action pose"
-more well defined perfect female figure -- This is just your other prompts.
-little bit muscular --> "fit" ? "gym body" ?
-more beautiful faces -- this depends on the model. The images you pasted are not all the same prompt.
-different backgrounds -- again, just prompt for the backgrounds and run multiple generations
-jewellery in accessories -- this one is scaling.

here are some more images for reference (these are the photos I got from Instagram so there is no meta/generation data)

if you want me to help more, find out where he's posted .png files with info.

 

[miaouxtoo]

btw are ther any guide for promt formating?
It seems if you use "_" it will help in specifying the prompt, e.g.
blue_hair will make only hair blue, while "blue hair" will make blue hair, but might also blue cloth and something else

I haven't read about that. It should be the opposite to what you claim I believe.

blue hair = broad prompt
"blue hair" = more phrase / exact.

I think what might be happening is that depending on the location of blue hair in your prompt, it'll change the priority, and maybe take blue and put it elsewhere. If you haven't given it extra guidance, it'll just assume a color you want is blue.
Maybe more specificity for the color of the clothes?

 

[TitaniumDickDiamondBalls]

Trying to figure out which model was used to emulate this FSN anime style. Any help/idea ? All files have metadata wiped out :/.

 

[miaouxtoo]

I'm going to guess it has a Misato (?? from Evangelion?) Lora applied - as to which model, it looks kind of cell shaded. Do you know when it was posted? Otherwise I'd guess Anything3 or 4 or Orange ... it doesn't have much 3d effect to it.

Trying to figure out which model was used to emulate this FSN anime style. Any help/idea ? All files have metadata wiped out :/.

 

[TitaniumDickDiamondBalls]

I'm going to guess it has a Misato (?? from Evangelion?) Lora applied - as to which model, it looks kind of cell shaded. Do you know when it was posted? Otherwise I'd guess Anything3 or 4 or Orange ... it doesn't have much 3d effect to it.

Lora is some sort of Tohsaka Rin one, but there are like 3 of them atm. It was posted today. May give you their pixiv account if it makes guessing easier ? Anything3....didn't check that model out.

 

[StepSisWasNeverStuck]

I'm pretty sure the Perfect World model on Civit is what you are looking for.

It is more advance then just perfect world.
Fine tuning the image is the real challenge, and i don't have experience like you guys so i am asking you guys to please help.