Thread 8901996

Previously, on /asdg/ >>8866074
The cartoony side: >>>/aco/csdg

No DALL-E, Meta AI or hailuo AI generated stuff.
No futa.
>>>/aco/futasdg
Report trolls.

>Guides
More modern options:
https://github.com/comfyanonymous/ComfyUI
https://github.com/lllyasviel/stable-diffusion-webui-forge
Too old, might hold some value: github.com/AUTOMATIC1111/stable-diffusion-webui/wiki | rentry.org/sdg-link | rentry.org/rentrysd
Hires Fix: rentry.org/hiresfixjan23
Inpainting/Outpainting: rentry.org/drfar | rentry.org/inpainting-guide-SD
Animation: rentry.org/AnimAnon
Create LoRAs, TIs: rentry.org/59xed3 | rentry.org/simplified-embed-training
>Auto1111 extensions
ControlNet: github.com/Mikubill/sd-webui-controlnet | rentry.org/dummycontrolnet
Regional Prompter: github.com/hako-mikan/sd-webui-regional-prompter
Tag autocomplete: github.com/DominikDoom/a1111-sd-webui-tagcomplete

4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb
4chan-x userscript from /h/ that lets you read generation metadata from images in-browser, and upload to catbox with one click when you want to share metadata.

>Wan Video 2.1
Setup Guide: https://rentry.org/wan21kjguide
Triton Guide: https://github.com/woct0rdho/triton-windows/blob/readme/README.md
Easy Sage Attention Install: https://github.com/woct0rdho/SageAttention/releases

>Related
>>>/b/degen
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/g/sdg
>>>/trash/sdg
>>>/trash/slop
>>>/vg/aids
>>>/vt/vtai

so as a guy that has had fuck all experience in terms of technical expertise outside of fucking about with MAGE/TensorArt, am I basically out of the game until 1) a new uncensored service debuts, 2) I buy a high end computer and learn how to run it all?

>>8902073
You can use civitai. Still uncensored, with some nuisances. But make the PC thing a priority.

I have no more ideas tybb

Hey I posted this in the other thread but thought this might get more views. I want to try making AI porn. Im interested in a realistic style making a bunch of softcore MILF pinup stuff. what program do you guys think would work well? Would it be this one?


Also, what sort of hardware do i need to use this? I have a shitty laptop right now so may end up just buying something else for this purpose. does the program/these types of programs have any inherent advantage running locally vs on their network or whatever? how long is the learning curve? like a month? 6 months to get it down 100%? longer?

im a lot more interested in the composition/ details of the images and trying to make it artistic at the same time. stuff like the background, types of shoes, hair style, etc. basically want to make aesthetically pleasing artistic stuff. i have some basic GIMP experience so can probably tweak and change things myself as well here and there.

>>8902004
i notice a lot of AI art has crisp detailed bodes but the faces often times are blurrier, do they just struggle with faces in general? It seems like he closer the face is the better job it ends up doing so maybe its a resolution related thing?

nice sweaty pits btw

>>8902114
1-The hardware:
You´ll need 16+Gb of RAM and as much VRAM as you can. These days the best is having a 12+ Gb GPU (if you are starting from scratch). You can do with less, but models are only get larger.
2- The software:
You´ll need a WebUI. These days the most recommended are Forge and ComfyUI. The latter appears to have more development, but the big downside is that´s node based and the learning curve is stepper than on Forge. Maybe a couple of weeks.
3-Composition:
That will depend on your prompting skills and what models you use.
And at any rate, there are add-ons for Krita for editing.
>>8902115
The problem is image resolution. The smaller the object it will be rendered with less detail. Irrelevant on a tit with only the nipple to care, ad on a face. The only solutions are large image sizes or focusing the prompt in the areas you want, in this case using cowboy shot, upper body or portrait; and pretty much forgetting about the knees and below unless you work the pose.

>>8902120
is it even worth bothering with a laptop or should i just build a PC for this? is this essentially the old PC gaming laptop argument? Im sure a gaming laptop or some higher end model can do it but idk how hot it would get if it took like an hour or something to fulfill a task. right now im limited on space so im unsure if i want to get a tower for this or not. not that im opposed to it eventually though.

is there a good resource you can point me to for a 101 course on this? i need to get familiar with all of the jargon and terminology. no idea what am model is or a lora but im guessing those are the same thing or related somehow?

would you suggest SD? it seems like there are like 100 ai people use now. but im guessing its like bitcoin where you have one main one and like 5 secondary ones and a bunch of weird ones that arent worth messing with? thanks for your help kind of excited.

>>8902133
nta but the only real requirement for AI gen is vram. Differences between GPUs with the same vram capacity are pretty minor.

Model (also called checkpoint) is the big 6+gb file. Loras are small optional addons that give it some more knowledge in a narrow area, or a stylistic bias, etc.

>would you suggest SD
>5 secondary ones and a bunch of weird ones
There's really only SD-XL and Flux right now, and old SD1.5 for people with crappy hardware. The rest you see are just people taking these and doing very small scale training on their own on top, to try and give it some nsfw knowledge.

>>8902114
>inherent advantage running locally vs on their network
You can try generating on Civitai or similar places for free. Base functionality is the same, if you get into it more it's missing a bunch of niche or advanced features.

>>8902148
so here is question i have that might help me understand how this works. Lets say you wanted to make something that looked like a character. would you describe the character and thats how it would be generated? Or would you have to download/create your own Lora which contains the information of that character? Do you train the AI with your own images? or Do you reverse engineer it by describing something in nauseating detail manually to get that character?


Lets say I was trying to make an image of gigachad, would i need to collects a bunch of gigachad memes for it to reference? Or would I make gigachad by making like 2 paragraphs of modifiers and keywords which creates something that resembles gigachad?

>greyscale image/ muscular man/ handsome face/ slicked back short hair/ short facial hair/ big jaw/ big chin/ thinking expression/ looking at viewer/ grey background/ etc

>>8902133
A high end/gaming laptop can do it, With 8Gb you should be able to use SDXL/Pony reasonably, and BigLust is the current "meta" for realistic generation.But remember those GPUs are a downrated version of their PC counterparts. Go with what your circumstances tell you.
Not really, there are sites like https://stable-diffusion-art.com/ but I just saw they started with the "members" claptrap.
https://rentry.org/ldg-lazy-getting-started-guide
https://rentry.org/sdg-link
May be of some use.
Model (or checkpoint) is the well, model, the WebUI uses to process your prompt. There are an ininty of them, but the important part is that there a few key ones, from which the rest derive, and dictate compatibilities. In a big generalization, the key ones are SD1.5 (stable difussion 1.5, the oldest, limited but can be used on a potato), SDXL/Pony(more sophisticated, better prompt interpretation, not intercompatibe), Illustrious (parallel to SDXL/Pony), Flux (can interpret natural language syntax, better have a lot of VRAM).
A LoRa is in short a patch or add-on for a checkpoint/model. For when you want a character/pose/clothes than the model doesn´t have or makes badly.
SD (stable diffusion) is a bit of a reminder of a bygone era. Their 1.5 model was the one that made local generation possible, starting Big Bang of content. But they themselves kept shooting their feet with claptrap like SD 2. SD2.1. SD3, and now going anti-NSFW.

>>8902156
Character:
Depends. Is something famous that has a lot of images on which the model was trained, say Wonder Woman or Rei? You´ll be able to just prompt.Something rare and obscure? Look for a LoRA.
You can train a LoRA yourself.
It can be by both methods, but just training a LoRA yourself will probably render more satisfying results (if any existent gigachad LoRA doesn´t)

>>8902167
so lets say i generated an image of a milf or something and i thought it looked decent but something needed to be changed, like i wanted a more defined nose, or wanted some light wrinkling around the eyes, or i wanted the eyebrows to be more penciled on, is there a way to use that image and to have it make those modifications in the same image that generated in it? or would you need a separate program to do that? i suppose another use case would be wanting to keep the same image, but rerendering the hair or the highheels or the lipstick, can you keep the same image but make modified variants of it? i kind of see myself doing that a lot to sculpt the image into what i want instead of wanting to regenerate the entire thing. salvaging what works and changing what doesnt sounds ideal. or adding something like longer nails, slightly more toned arms or legs, things of that nature. like i said earlier im a MILF coomer.

>>8902187
Yes. You can either inpaint or img2img (image to image) the wanted changes. No separate programs (unless you want to edit on editing software like krita or photoshop).
For modified variants you keep the seed of the image you liked and tinker with the prompt or img2img the changes.

>>8902162
>now going anti-NSFW
so when you see stuff which is just overtly porn are they using an older version of SD? or are they using something different than SD altogether? like i said earlier im more interested in softcore and pinup but im guessing anti NSFW means no nipslips or anything suggestive? i could probably make some hot milfs regardless but im sure the clothng styles would be pretty restricted in that case.

>>8902156
I doubt you'd be able to get anywhere close with a just textual description of the guy. Same as if you tried describing the meme to someone who hasn't seen it, they would imagine a different face that still has all the listed features.

If he was in the training data and properly labeled, you can just ask for him by name. Otherwise you need a lora.

>>8902190
is inpaint a tool within stable diffusion? i had the impression it was a completely different program but i have never used any of this stuff before. is SD a bit like blender where it has multiple tools inside one program?


i have seen some really interesting stuff where people take other peoples illustrations as a request but they put them through some sort of AI tool to make them photorealistic for them. its pretty cool. Is that inpaint or a similar tool?

>>8902193
okay i see, i was just asking that question to try and gain a grasp of how this stuff works. I wasnt planning on making a gigachad but then again since people only post the same like 20 maybe it would be useful to do kek.

Thanks for all the time so far guys i hope im not derailing the thread i only pop in here from time to time to see what you guys come up with, but it wasnt until today where i decided i wanted to try this myself.

>>8902191
Yes. People are still on SD1.5 either due to perfected workflows, not being able to use newer stuff or just not caring, but the last thing SAI made and people used en masse was SDXL, and that has an irrevocable license, so they can go pound sand.
By this point SD is a bit of a generic trademark, the currently used base models are SDXL/Pony, Illustrious, Flux and NAI.

>>8902191
Not like they can recall what has already been shared for free. Only their new models will be censored, and they're looking pretty bad anyways.

>>8902187
You have a couple of options. Reroll the whole image, reroll one area of it that you specify, draw/edit in something and have the AI clean it up, expand the edges of the image, etc. Depends on how much effort you want to put it.

>>8902194
What you use is a WebUI, that there are several of the out there. Stable diffusion was SAI´s product that started the local craze, more specific Stable Diffusion 1.5
Stable Diffusion is kind of an umbrella term today.

>>8902196
We're so dead anyways it's at least a breath of fresh air. Come drop a gen here once you make something decent.

>>8902201
if it isnt broken i guess theres no reason to upgrade. how do you obtain the older builds? i imagine this is like obtaining an old patch of any other software where if the creator no longer hosts it, there is the problem of finding a trustworthy website to download it so you dont get a virus or something on your computer DLing it from some random website?

SAI? as in paint tool SAI or whatever fro like a decade ago? they made this? thats crazy. im surprised they released this for free and never charged people for it. What was in it for them?

>>8902201
>SDXL/Pony, Illustrious, Flux and NAI
are these different builds of SD but branches other people have turned into something else? like they took the source code but changed it and made it their own project?

>>8902203
can you still get the older model easily from the creators website or do you have to track that down? I have seen some AI art that is an obscene scale like really large. I have no interest in that but im curious what program does that since looking around it seems like SD makes relatively smaller images. like 1024 by 1024? ive seen some stuff that is really high resolution like 2000 pixels high maybe more maybe 2500?

>>8902204
So did SAI make it for free? you say product but im curious why they woul dmake something so powerful and release it for free if thats what they did. just curious what motivated them to do all that work if they didnt get paid.

>>8902205
i used to make a lot of art the old fashion way so this entire premise is interesting to me. ill definitely share some cool stuff i make and throw it your guys way at some point.

>>8902214
i shared this shitpost i made in the other thread. based on this image do i have potential to be one of you guys? or is it over for me? im curious what this would look like if you asked one of these ai tools to make this look photorealistic and how it would interpret this mess.

>>8902214
Not really. Just go to civitai. Although due to recent complications, you might have to look in other places.
Yes, these modes are refinements.
You can upscale or, having enough VRAM, just staring with bigger images. The benefit of newer checkpoints is that they are trained on larger images 1024x1024, giving out better image resolution.
SD 1.5 was made in collaboration. One of the collaborators leaked it and then there was light.

>>8902214
SDXL and Flux make 1megapixel images (1024x1024 square or similar). If you go higher from scratch it starts to break anatomy. But once you already have an image, you can upscale it as high as you want, it'll clean up details and even fill in new ones. Does take a long time though, hardware requirements increase with the square and online sites charge a premium.

>>8902222
there's no porn or nudity, try /g/

>>8902006
a VERY good friend

>>8902230
>there's no porn or nudity
i can draw a dick i guess i originally hand crafted that image as a /v/ reaction image. the original content matter was some guy having sex with a shemale on a couch.

>>8902230
i was looking around on deviant art just out of curiousity of what people are making and i couldnt help but notice there is a metric fuckton of users on there with paid content. Do you think they are just using an older SD version like you guys but upscaling it with those services you described? Some of those images were pointlessly large. where a boob took up the entire monitor.

>>8902233
Then you go to >>>/trash/sdg
Furry gets you a 3 day vacation in /aco/

>>8902236
oh i didnt know that, i dont even think of him as a furry i just think of him as a meme character my bad.

ive asked this before when ive stopped here on occasion but do you guys have a page where you share or upload your best work? it kind of seems like a shame that a lot of cool stuff has probably been made by people in these threads but it isnt viewable. that was why i went to deviantart because i was curious what stuff people were coming up with.

>>8902247
Not really, you can check Desuarchive for the old threads:
https://desuarchive.org/_/search/boards/aco.desu/subject/%2Fasdg%2F/
https://desuarchive.org/_/search/boards/aco.desu/subject/%2Fcsdg%2F/

>>8902247
You can check desuarchive.org for older threads. Some people post stuff on Civitai where you'll often find the metadata attached as well, see how it was made.

Most of the stuff on deviantart is mass-slopped with wildcard prompts, people make hundreds of images every single day and sell a subscription. Same on Patreon with cartoon/anime stuff.

>>8902257
what do you mean by mass slopped? and what do you mean by wildcard prompts? Do you mean they put in a prompt and they just generate like 100 images based on that prompt and then pick the 20 or so that look best?

is there a reliable way to make the face look similar between images? that seems to be a problem a lot of people have on DA wher ethey have a gallery or something but there isnt much consistency between the character from image to image. i suppose this is fine if you are just interested in a general theme but if its a gallery dedicated to a character or something you obviously wouldnt want the face changing much from image to image within one gallery once you got one that looked good.

>>8902295
They create a few separate prompts for the usual progression, posing, blowjob, cowgirl, missionary, boobjob, cumshot. Then run the same thing with a different girl every day.

>>8902508
Very nice

>>8902584
Thank ya! Trying to do more of her!

pnrsx

>>8902647
Hope it goes well. She looks great

>>8902736
I do love her as well and the tomboy variant I posted just above. Gotta cook them up in more poses/outfits. So, here's the newest one. May be loving the pink bimbo oni more than the other though, lol

>>8902767
Yeah, I would usually prefer the tomboy one but something about the pink one is really doing it for me too. Great stuff all around.

I kinda hate /b/ so I'll post this one here, just nobody special

>>8902222
A glorious shitpost, I hope you figure this out. It's pretty fun.

>>8902222
I noticed you had quads and figured you would like this.

>>8902958
And this is what happens when you run it through my Goon Machine

why is chroma so much better at making huge penises than huge boobs
are the captions for boob size just bad

>>8903051
chroma likes to drag ballsacks and dicks on girls because it was baked by a furfag degenerate

>>8903071
damn

>>8902009
Great. Model?

>>8902214
>SAI? as in paint tool SAI or whatever fro like a decade ago? they made this? thats crazy. im surprised they released this for free and never charged people for it. What was in it for them?
No, not Paint Tool SAI, we're talking about Stability AI, the company that kicked off the whole AI imagegen revolution in 2022 when they released the Stable Diffusion 1.4 checkpoint. 1.4 was a very primitive model and imagegen didn't start being taken seriously until 1.5. There's a company called NovelAI that runs a paid subscription service to generate anime images, and they produced their own checkpoint, NovelAI 1.0, by training with the contents of the Danbooru anime imageboard on top of SD 1.5. And this checkpoint got leaked to the public, which was a big deal for /h/. Notable checkpoints that were further refinements of NovelAI 1 include Based64 through 68.

SD 1.4 and 1.5 were trained to generate 512x768 images, but the quality could be improved through upscaling or HiRes Fix (which this is an example of). Stability AI then introduced Stable Diffusion XL, which could generate 1024x1024 (but in practice, most people use 1216x832 and continue to upscale from there). SDXL became the basis of Pony Diffusion XL by AstraliteHeart, which includes datasets from Danbooru, Derpibooru (the pony imageboards) and Rule34. There's a big ecosystem surrounding Pony, but it's controversial because AstraliteHeart hashed the names of all the artists in the training data, ostensibly because he didn't want users to replicate an artist's style exactly, but also because he runs a subscription service and he wants to charge money to unlock the names.

>>8903179
SDXL is also the basis of BigASP, Lustify and their hybrid BigLust, which are currently the most popular checkpoints for realistic porn. A group in Korea trained their own SDXL-based anime checkpoint called Illustrious which doesn't obfuscate artists. There was a further refinement of Illustrious called NoobAI, and derivatives of NoobAI are now the most common checkpoints used on /h/.

In the meantime, Stability AI started being opposed to having their models used to generate porn, so they released Stable Diffusion 2 and 3, which were practically useless: SD3 was famously unable to generate a woman lying on grass without messing up the anatomy. Some employees of Stability broke away and founded a new company in Germany called Black Forest Labs, which released Flux, the current state-of-the-art local model. It doesn't actively censor porn, but it still didn't have much NSFW in its dataset, so it needs more training. The most promising of these efforts is Chroma, which is trying to finetune Flux with anime, furry, photo and classical art datasets. This is the model that was used for all the impressionistic gens in the last thread. Chroma doesn't seem to be able to identify anime artist styles by name, so /h/ is waiting to see if that'll improve when its training is supposed to be finished at epoch 50.

Also, the video generation models that came out of China -- Wan and Hailuo -- can be used to generate images if you just set them to produce one frame.

Does anyone have any idea how to make good prompts for a character leaning forward with a slack jawed/brainless expression? I'm trying to gen hypno stuff and have had pretty poor luck. I'm using a SDXL based model.

>>8903195
Perhaps you could add "half-closed eyes" to the prompt?

>>8903233
Thank you for the suggestion! 'Half-closed eyes' usually just gets me closed eyes but that's probably m model not being super good at parsing prompts. I've had some decent luck with 'drunk' 'emotionless expression' and 'neutral expression'. I'm actually more interested in the leaning forward/body pose than the expression at this point.

>>8902007
Stop putting the vienna sausages in the french press, Heather!

>>8903234
this is because "closed eyes" is more trained than "half closed eyes". Have you tried "stunned, entranced"?

>>8903244
Yes. "Stunned" and "Dazed" both get expressions of excited shock rather than sedation/mindlessness.

>>8903271
mind break, mind control, hypnotized?

Gave it a try myself with an illustrious-based model and "half-closed eyes, mind control, emotionless, open mouth". Though it's closer to what porn sluts think a seductive face looks like, than to hypnosis. Might be unavoidable dataset bias with nsfw models.

>>8903534
Just look around for a LoRA and avoid the headache.

>>8903681
Yeah I know. More curious about what the model can do than actually looking for a result.

>>8902958
>>8902965
brilliant looks like a jojo character

>>8903179
>NovelAI 1.0, by training with the contents of the Danbooru anime imageboard on top of SD 1.5
so is novel ai a plugin you use ontop of SD 1.5? or are they just 100% seperate programs?

which program are you guys running in here? ive been watching a few tutorial videos to get a general idea of the workflow but im not sure what tool to download. especially because people make it soud like 1.5 was either neutered, or replaced by the neutered new model (1.6?) so im not sure what i should use. what website do you get SD from? im sure there are 100 mirrors and hosts but i want it from the most reputable site so im not downloading anything weird or "altered".

>>8903194
>SDXL is also the basis of BigASP, Lustify and their hybrid BigLust

so in this example are those plugins you download and use on top of SDXL? or again are they separate programs? like would i download Noobai for example and use that as a plugin ontop of sdxl? or if you have noobai you do not need to download sdxl?

>video gen

its an interesting concept but i feel like for now images are more than enough. im sure in another year or two it will be just as good as image generation is, but for rapidly experimenting i feel like images are probably just easier to work with in most cases so for the time being i dont need to worry about that but i still appreciate the developmental context anyway. good to know.

>>8903688
The answer is both Anon. Loras are great, but so is being able to do it on the fly with just your throbbing brain. Have you tried familiarizing yourself with Danbooru tags? I know Novel was trained on it, so thats worth your time. Also consider reading the Thesaurus, Helpful to me.

>>8903709
The program yo run is a WebUI. There are several, Automatic 1111 (the first one, pretty much abandoned today), ComfyUI, Forge, SwarmUI, SD.Next.
The grand bazaar of AI generation is civitai.com
No those are refinement models of SDXL, so just download say, BigLust, put it in the right folder, make sure it shows up and start.

Imagine asking questions for two days without even trying it out

>>8903763
which ones are most trustworthy/ popular? i dont usually install much on my computers unless i know a little about it.

>>8903786
i literally started asking questions this time yesterday thats 24 hours. i have a shitty laptop right now that probably wont be able to run any of this but im trying to figure out what my options are in the future. need to buy a new computer anyway this one is an old laptop which is basically falling apart.

>>8903709
>so in this example are those plugins you download and use on top of SDXL? or again are they separate programs? like would i download Noobai for example and use that as a plugin ontop of sdxl? or if you have noobai you do not need to download sdxl?
They are not programs, they are model checkpoints. SDXL is a model checkpoint. Bigasp/noob/illustrious are also model checkpoints, they are versions of SDXL that have gone additional training on nsfw images. They are not plugins so if you were using illustrious you would have no need for sdxl as illustrious IS sdxl but finetuned. LORAs are like a "plugin", they're like a small model that you add-on to the main model. There is many LORAs for whatever character or concept you can think of. The model checkpoints require a huge amount of resources to train but a LORA can be trained by anyone with a decent graphics card.

>>8903794
The two most popular (and maintained) are Forge (UX similar to Automatic 1111) https://github.com/lllyasviel/stable-diffusion-webui-forge and ComfyUI (node based) https://github.com/comfyanonymous/ComfyUI

>>8904104
this is good, what model/upscaler?

>>8903794
you can try it out if you give colab $10bux for the pay as you go plan, a "session" usually only uses like 30 cents of that.

>>8903709
>which program are you guys running in here? ive been watching a few tutorial videos to get a general idea of the workflow but im not sure what tool to download. especially because people make it soud like 1.5 was either neutered, or replaced by the neutered new model (1.6?) so im not sure what i should use.

1.5 isn't neutered so much as it takes more skill to use than the newer models. The 1.5-based models are good for generating 1girls, but not more complex scenes. and there is a tool for copying image structure (controlnet) that works better in 1.5 than in the later models.

The "base" checkpoint thing that people keep mentioning is mainly relevant for 1. what image sizes you can produce and 2. what LORAs work.

Excuse the slop, but thought of enquiring about nipple clamps. (Slop here was genned from img2img). I found 2 Loras and downloaded the first two from this result https://civitai.com/search/models?sortBy=models_v9&query=clover%20clamps
Subsequently I put the safetensors in the lora folder.
The intended nipple clamps still looks like this, or randomly more like nipple piercings or nipple accessories. But I'm quite new to Forge and didn't quite do anything else besides putting Loras into its folder. Any way to hopefully achieve having Forge gen the actual clover clamps like the civitai examples, any extras/steps needed? Any advice appreciated :)

>>8905116
>:)

>>8905116
Forge is built on top of automatic11, read the user guide
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#lora

>>8905116
Tried one of those loras (the Illustrious version) and didn´t quite get it to were I wanted, but it was the few attempts my Buzz budget could afford. Maybe putting the clamp trigger tokes behind a BREAK can help, ie BREAK, cloverclamps, clover nipple clamps, silver nipple clamps.

>>8905139
Sounds more like he didn't even activate the lora, just put it in the folder.

>>8905140
It´s a possibility. He first should check it if it shows up on the available LoRAs .

>>8902003
>>8905321
Requesting inflated belly (preg but with a hose anal insertion).

here's some pics I inpainted that I planned on img2viding later

>>8905461
wait no, I flux gen'd that one THEN inpainting it because flux sucks ass. picrel was inpainted from a screencap though

>>8905470
I'll stop spamming with this last one

>>8905496
fuck yeah

>>8905470
man, the wonders of technology

>>8905469
kek

It's like that Old Spice commercial

>>8905514
She the best. Don't know if I posted this one of her but was working on another. Didn't come out how I wanted so we settle.

>>8905461
>and they're fabulous
Having met her in her 20s, she was drop dead gorgeous in person in ways television couldn't capture.

>>8905602
she definitely is
glad to see more of her

>>8905116
what UI are you using? if you're in Forge or A1111 you have to go to the Loras tab and click on it to activate it

>>8905689
Oh, do you guys not write in the thing manually anymore?

>>8905721
You can do it both ways.

>>8905616
Nothing *TOO* crazy but another of her. Trying not to fall into just normal fuckpose pictures but who knows...I may. Just want to get varied with her.

>>8905803
I feel you
she makes it all look good and the fuckposes are definitely no exception

>>8905809
One more for the night

>>8905892
Holy shit
Thank you!

>>8905905
I got you. Though, sadly, the end for her for now.

>>8905936
It was a different time.

>>8905939
i'm surprised this era isn't more of a setting

>>8905951
Yeah, a lot that can be done with it

>>8905949
damn good one

>>8905137
>>8905139
>>8905140
>>8905689
Thanks for your all your replies. Yes it is Forge. I think I did go to the Lora tab. It does show both of the Loras, so I clicked them, and at the positive prompt and came out. Additionally I tried adding in keywords like nipple clamps and cloverclamps in there. Btw I went back each of the pages I got the Loras from and saw Illustrious and Flux. What are Illustrious and Flux? Coz it seems my gens kind of gave the same nipple effects even without the Loras.

>>8906519
Illustrious and Flux are other base model/checkpoints. They are NOT compatible between each other and SDXL/Pony.

>>8906573
Technically about as compatible with SDXL as Pony is, it's still the same architecture at least.

>>8906577
Things aren´t going to look very nice, thus the scary warning.
Should have said to check that the LoRA is made for the same base model being used, as Pony LoRas with Pony models.

Not that I'm complaining but shouldn't all the painting-based gens be in /csdg/?

I love inpainting so much, bros

>>8906986
Technically the thread is still called "adult" not "photorealistic" like I suggested multiple times. And they would seem equally out of place next to marge simpson as they do here. But yes.

>>8906986
>>8907204
I don't like the cartoony images so I'd prefer to post in the realism thread but if it's a problem I'll switch. I'm actually hoping someone could upscale a few of them with a photoreal model and CN tile and see what happens. Either way I won't be genning for about a week or two so here are some Catboxes. Adding that chroma quality lora at .5 strength doesn't hurt either.

>https://files.catbox.moe/9g517o.png
>https://files.catbox.moe/n0ucpm.png
>https://files.catbox.moe/q4vds9.png
>https://files.catbox.moe/t8ia6e.png

>>8906519
As >>8906573 said, Illustrious, "1.5", "XL", Pony, Flux, etc. are all Base Checkpoints. When you download your model, it usually says somewhere what base model it is. IIRC if you click on the lora info i circle it should be somewhere on there.

A LORA trained on a base model will generally work* on any model with that base. It will not work on models with other bases. If one Lora says illustrious and one says Flux, then you should only be using the one that is for your model's base model.

My unpopular opinion is that the vast majority of concept Loras degrade image quality to an unnecessary extent, especially in photoreal models... but post settings and proompt because you might be making a mistake there as well

>>8907557
She cute

>>8907714
Even better.

I fucking love this technology. But sometimes, I can't get it to obey simple directives like FACING FORWARD / STRAIGHT ON view. This is the ONE good gen out of like 20 ... do you guys see anything wrong with my prompt?

https://files catbox moe/95wl9s.png

>>8907877
that is a hell of a pic

>>8907886
Thanks! I have ... a lot... more ... haha. Like .. i'm past 5,700 generated images now ..

>>8907891
I can see why. I'd have a hell of a time stopping if I was making things like that.

>>8907893
It's addicting being able to change anything ... age... hair stlye .. position ...

>>8907897
Fuck that one is killer.

>>8907902
Thanks ;)

>>8907906
I'm going to stop annoyingly spamming you with (you)s, but I really enjoy your work!

>>8907912
Oh you can keep doing it. Eventually I'll drop off. I like the conversation. If you want to find me again, my discord is jjhm437

>>8907930
Ok. I just didn't want to wear out my welcome. The oiled look is 10/10.

>>8907714
So deliciously early 90´s

Where do you guys get your real people LoRAs now that civitai banned them?

>>8908372
Tensor.art

Bit if a pain to find what you want between 90% of models being flux, unavailable to download, or asking for money, but they're there to be got. Get 'em while you can- build a strategic LORA reserve.

>>8908074

Used a Yelena Belova/Flourence Pough model and put her in The Temple of Doom.

I'm just starting to use ComfyUI and I don't know how to increase the length of videos on Wan2.1 in ComfyUI

guessing the overlap between this thread and gacha games is minimal, but oh well

>>8907877
Need an asian like this

>>8906573
>>8906597
>>8907383
Made sense. All along I was only using Pony. So checkpoints is like basically the skin and look the gen would produce. The Loras I got are meant separately for either Flux or Illustrious. Alternatively I found one Illustrious checkpoint that I'm interested in, and have been experimenting for a while. Meanwhile I found Pony loras but seemingly they are tweezer clamps that I would experiment later.

Not so perfect, but I think this came out not too bad, for me that is. An img2img from an existing DID art but I had to play around with the denoising (0.7) as Pony is different and Illustrious MMMix is new to me. Hope attached pic is suitable for /asdg/ as it seems quite bright and not that entirely photorealistic. A bit tricky to get the chains/strings on a clover clamps like their real life counterparts, but anyway I'm satisfied the main "head"/actual clamps can properly pinch at the victim.

Gotta say you guys in /asdg/ have been really helpful with fairly prompt replies so it really helps those who are keen to learn and improve here. So thanks a lot guys. How helpful is /csdg/ in case I switch to cartoons? Is there any relevant space to discuss Novel AI as I casually gen pics from there, and got decent DID artist styles and characters genned but unsure which thread is relevant for those?

>>8909520
No problem.
/asdg/ is still a general thread, /csdg/ is the specialised one (long story).
Probably not much trouble discussing NAI here or in /csdg/.

>>8909545
>Novel AI
Oh alright. I thought it is strictly Stable Diffusion threads and I understood it as local genning only. I did post NAI pics before in /aco/ female bondage threads but a few fellows were unwelcoming of what they term "ai slop" but Novel AI is able replicate those well known DID artists and with proper prompts produce really hot ones easily compared to my experience so far with Forge. Also once, I posted an NAI pic in /csdg/ but no response from them, so I thought it was a wrong thread lol.

>>8909579
It is for SD (local or on-line), but there´s nothing for NAI and, really, we do not need another AI thread. Slop is just a general derogatory term for AI, chiefly due to the onslaught of low effort 1girl pics that absolutely flooded the site.
About responses, it happens.

>>8909545
>/asdg/ is still a general thread, /csdg/ is the specialised one
I thought it was the opposite. This seems to be for the very specific photorealism niche, while /csdg/ takes everything from actual cartoons through videogame models and uncanny valley 3d to borderline anime.

Maybe it's just me but Illustrious MMMix checkpoint seem less photorealistic compared to Pony? But then I really like it. Pony often gave me like improper fingers, weird limbs, etc and fairly often wonky with ropes. But then what I'm genning probably may not fall into the /csdg/ category, unless I tinker further somewhere in the prompts later.

(Still trying out, so denoise increased to 0.9 but with tonnes of positive prompts to maintain as much of the original pic from img2img. But the clamps are clearer and improved)

>>8909665
There´s a continuum between realistic and cartoon, and you have to check and compare to see what fits your needs.
Watch it with denoise, the higher, the more the image changes, maybe try inpainting (with a mask) to at least localise any issues.

>>8909742
thanks, I hate it

that said whenever I try she looks even worse

>>8909780
How can one even hate on Poppy?

>>8909798
only in realism

>>8909742
Suggestion: pussy from behind.

>>8909839
Here, sir.

>>8908424
Amen, and thanks!

>>8910146
who's gazmandros and why is he shilling his patreon/deviantart here

>>8910157
Just like and subscribe you doublenigger

what's the best way to upscale a WAN generated 480 video to lets say 720p?

>>8910318
pirated topaz

hello, i’m not sure if this is the right place to ask, but does anyone know how this guy i found on instagram named “stabledai” makes his videos? i’m new to stable diffusion, so i’m especially curious how he made the jiggle physics of the characters so smooth

>>8911037
go on ..

>>8911251
wan video guide in the OP

>>8911266
fuck that's nice

>>8911270
Glad you like it.

>>8911284
oof I like your style

>>8911288
Thnak you. The last two pics are clearly different from the previous ones, but I like experimenting with different artstyles.

>>8905116
>>8909520
>>8909665

>>8911298
yeah I guess taste would have been the better word
big girls in tiny bikinis

>>8911306
Based taste, indeed.

I accidentally created a realistic Aeon Flux.

>>8911305
>"Those ropes do nothing"
"What?"
>"Those ropes, on her tail, what are they supposed to stop her from doing?"
"... I thought it was sexy."
>"Fair enough."

>>8905471
Hey trekkie anon, can SD make decent men? I... would really like a Spock nude... tasteful, and as he looks in tos. Do you think its possible? Can anyone here try it?

T. Autistic femanon

>>8911746
Why Spock? Like, what about an emotionless xeno gets you hot? No judgement, I'm an autistic gooner too. In other news, I'm beginning to think trying to scam normies out of cash in exchange for low effort porn won't work. I'll keep you guys posted.

>>8911746
You're that Spock fangirl who was in the DALL-E 3 threads last year, aren't you?

>>8912053
Different Anon here ,
can you do some nude males with leather collars walking around town, in and about a fancy mansion, by a pool, et cetera ?

I will give you bonus points if you can gen two naked-other-than-leather-collars-and-sandals slave boys running while a mature man on horse back is throwing a net at them.

Inspired by a dom/sub scenario that was inspired by the original Planet of the Apes movie.

>>8905045
Simple, casual, public nudity.....

I like ! More please!

Maybe a nerd girl , nude in a college lecture hall? in a college library?

>>8912175
Sorry, that's too gay for me.

>>8912215
kek based

i love how retards here expect you to do commissions for free

>>8912221
it's because some lonely anon on /b/ comes on every day offering to do free commissions for loras and images

>>8912221
to be fair I'd do a chargeback if I paid money for 98% of the slop in this thread

people in the more active threads usually welcome requests to challenge their abilities

>>8912345
Weiss with tits? That can't be right...
Cool style tho.

>>8912406
Is she supposed to have small tits? Sorry, I don't know the canon appearance of the characters of this show.

>>8912420
She's also like 14, but w/e didn't stop all the other fan artists.

>>8912441
Fuck I'm feeling guilty for genning her now...

>tensor deleting now also deleting celebrity Loras
Did we settle on proper civit alternative that's not in danger of bowing to credit card jews yet?

>>8912502
No.
And that´s pretty much impossible unless torrents are involved, and even then, it still would need either a big fool with very big pockets or many people wiling to go the crypto way.

>>8912535
/hdg/ maintained their own repository of Pony loras for a year. Just a rentry where each lora had a name, a catbox/mega link and optionally an image sample.

The downside is only one person has the keys to add/edit stuff. But it's super easy to archive the rentry locally, and no way to takedown all the loras at once.

>>8912535
So no chance on having something as organized and managable as civit again? I've tried using huggingface and other sites that rehost deleted models and it's absolute ass without any screenshots or prompts and trigger words. Just seeing what others were able to do with a model was worth a lot.

>>8912573
It would have to stay non-profit, just a repository and nothing else. HuggingFace could work if it required a description and a sample image with metadata to go with every upload.

Finally checked out SDXL for realistic porn after mostly fucking around with realistic Pony models. Is it just me or is SDXL actually dog shit at generating realistic porn?
No matter what model, I always get the same kind of dumpy MILF-ish body type. Then the moment I prompt anything about the tits or ass (even just the words "breasts" and "ass") the girl just gains weight. Breasts lot of times look flat, saggy and fatty, big asses always look like they're made entirely of fat and 0 muscle.
Body type loras do the same, breast sliders just make her fatter. It's like the whole thing only knows how to do one specific type of mombod.
Using BigLust1.7.

Pic related, quick example of the difference between not prompting breasts and just putting in "big breasts". Had a bigger effect on her weight than the breasts themselves and it just gets worse the harder I try to force breast size.

>>8912624
You wanted realism, you got it. Jokes aside, you didn't provide metadata so we can only guess at what you did wrong. Look at the image gallery for your Biglust version on Civitai, you can see other people's generation metadata and plenty have big tits without bodyfat.

>>8912624
Apart from what >>8912630 said I´d recommend either using a slider LoRA or adding tokens like skinny, svelte, hourglass figure.

>>8912682
wrong thread and fuck off with your reposts

>>8912630
>>8912640
Aight, trying to upload with meta data now.
Unrelated but now suddenly the background/lighting is fucked up and looks over generated. Same with her smile, she suddenly has this super exaggerated smile from just prompting "flirty, smile".

>>8912722
And with "large breasts" added.

>>8912722
4chan strips metadata you dingus

>>8912725
Used the catbox script in OP. Should show a green icon next to the file name if you have it.

>>8912726
oops, egg on my face

>>8912722
Try with a low weight on smile, like (smile:0.7) and start hitting the negative prompt with fat, chubby and any synonym you can think of and don´t be afraid to go ([token]:1.3).

>>8912751
>start hitting the negative prompt with fat, chubby and any synonym you can think of and don´t be afraid to go ([token]:1.3)
I did that before but I stripped the neg prompts clean for the two examples since I keep reading about shizo negatives (or any at all) messing with generation. Will try with messing around with weights though.

>>8912724
I meant to reproduce this pic exactly, but ended up too lazy to download the dmd2 lora or the girl. Also Euler A cause fuck using meme samplers. Same prompt and seed though, on biglust 1.7. https://litter.catbox.moe/mshj4kxtns6cuwtl.png
Then i added (gigantic breasts) instead, which blew her up https://litter.catbox.moe/74klrc9dfz651c9e.png
and then added skinny, which shrunk her back down https://litter.catbox.moe/bez6skyj2m2unevg.png
If you're seeing different behavior, it's likely bias from either of those loras. Either way it seems pretty unstable across seeds. Feels like the model wants natural language prompts, not comma separated tags like anime models do. Not my thing, maybe someone else can help.

>>8912751
"pro" tip, try using positives when possible instead of negatives. They're usually much stronger.

Sometimes I legit wonder how a celeb would respond to being diffused. Shrug, cool, angry, batshit...

First time trying out text2img. Was previously using img2img so far in Forge, more lately using Illustrious rather than Pony as I'm satisfied with how it combined with the clover clamps Lora for Illustrious.
Having a break from bondage pic gens, so these few are text2img but I discovered, to make the clover clamps visually better is to increase sampling. Also like Novel AI, the prompt can recognize a character name and Illustrious can make i.e. this anime girl from a certain series.

>>8913074
>"Laughing like the clamps are nothing"

>>8912053
Bit gay m8.

>>8907906
>>8907897
Damn these are good, can you share the metadata for them?

>>8912053
>>8913427
Different anon, and definitely gay. However it gave me a wacky idea with my casual Forge run. Sorry Trekkie Anon, but I tried running this through img2img and this did not look much like Spock. But I just thought for fun of adding clover clamps....to the guy....

Shit. This is a one off slop and I want to return to genning girls with clover clamps instead after this.

>>8913906
nta but metadata is in there
if you have the catbox extension from the OP, just right-click the little download button next to the filename

4chan prunes png file metadata, but many tools instead save it in the transparency channel of the image itself

>>8913912
Oh shit I'm a dummy, thanks a lot

>>8913427
I can't think of any way to make an image of a solo nude male appealing to straight women without also making it appeal to gay men.

>>8914389
>She's with me

What is the best way to make this kind of images (forget that it is a video), using a realistic illustrious model or using a refiner from an anime model?, im obsessed with that body textures quality

>>8914902
I don't think you can get skin texture that detailed from an anime model.

>>8914998
chroma core

>>8914902
>realistic illustrious model
NoobAI. There's no reason to stay on illustrious if you're based on anime already, you're missing out on a ton of character and concept knowledge. Now, even though these are somewhat capable of realism normally, as soon as you dip into their anime knowledge it tends to push them away from realism and into uncanny valley. Left pic.

>using a refiner from an anime model
This has its own set of issue with color bias, lighting, character proportions, head to body ratio, huge eyes, etc. Imo it's usually obvious it was a "drawing" underneath. Maybe from a very realism-adjacent style. The advantage here is you don't lose any knowledge compared to the more realistic anime ones. They tend to sacrifice some in the process of retraining their base style.

I go from a realistic noob model to a proper realistic one, and use img2img instead of refiner to allow use of controlnet. Right pic. Main downside is those models struggle with even basic anime features, like purple eyes or drill hair, and will try hard to remove them from your image. Controlnet helps with some of that, or inpainting those features separately afterwards.

>>8914998
This, and even for proper realistic ones it depends on whether it was trained on airbrushed insta thots, brazzers, or a more reasonable dataset.

>>8915297
which is a good noob ai model to make realistic anime chars?, thanks

>>8915345
Left pic was SamBase. Not sure I'd call it good, people make new ones every day so it's hard to keep track of where the quality is currently.

>>8914169
Welp after jumping through a thousand hoops to get SD working on my 6750xt for a while it just stopped working(again) and I've caved. Buying an Nvida card. Also I need it for Lightroom because there's just a glitch with AMD cards that hasn't been fixed in years apparently cus Adobe is Adobe.

How much faster is generation with team green?

>>8907714
Do you have any metadata for this one? Can't quite put my finger on what makes it look so good vs others to me.

>page 9
Bump

I kinda feel bondage gens are probably niche here and appreciate the discussion and help I got earlier in the thread. Like I'm probably the only guy who crossed over from the female bondage thread but casually genning my own stuff.

Btw on its own, strappado or armbinder poses don't get genned, so the boxtie (forearms crossed and behind the back) are the common position that still comes out despite the prompts, so I had to add in Illustrious loras for that.

Thought of posting something here if the thread fell down to page 9 and someone needed to bump >>8916803

>>8916863
The pose situation is related to model training. The more images of [x] in the dataset, the more likely that [x] is gong to come up.

>>8917062
She got them Wonderbread knees

What's the strat for inpainting lingerie? Every time I use a a single solid color, it just looks cg instead of realistic and lacy or whatever.

>>8917916
do you just inpaint or also prompt the lingerie, e.g. "red lace panties" ?

>>8917452
>wonderbread
haha not too sure what is that. But its a tag I learnt from danbooru, so I typed 'sitting_on_knees' or danbooru specifies it as 'seiza' making the checkpoint/model sit this way.

Btw a lot of txt2img gens I've tried shows her full body, and unsure how some DA AI genners can sometimes focus on the model being nearer or like focus on their upper body. I don't like some undesirable gens that zooms out the model. (Still Forge btw)

Hope its not too boring of me posting here (bondage stuff), but out of curiosity there is a tag for a girl from a very viral and well known recent hit movie. Maybe she's too new perhaps and I needed to type some tags manually that she's known for, i.e. clothes and the hair "dutch_braided_ponytail" but however this is the best ponytail that can be genned that I tried.

>>8918275
Better look for a LoRA.
It´s highly unlikely any model has that if it´s something that just appeared.
Maybe combining ponytail and braid tokens something similar can be obtained
https://danbooru.donmai.us/wiki_pages/tag_group%3Ahair_styles

>>8909798
tristana better

>>8918270
Her knees look like the end of a loaf of wonderbread. To answer your question, have you tried adding shot names to your gens? Try Cowboy Shot to get a pic from the knees up etc.

>>8918275
>>8918496
Thanks for the suggestions.
>Lora
Stumbled upon Loras for her and also her Huntrix groupmates.
>loaf
LMFAO.

>>8918292
>>8918496
Thanks for the suggestions.
>Lora
Stumbled upon Loras for her and also her Huntrix groupmates.
>loaf
LMFAO.

>>8912053
Dude you should know the answer to that by now

>>8918186
I use a realistic pony checkpoint and include something like "red lacy bra" and it just ends up a mostly solid color instead.

Kinda new here, why exactly are all 3 AI threads segregated?