>>16027 >>16028 Any good guides/places to start? Completely new to this AI/Stable stuff

>>16076 https://rentry.org/voldy For the basics https://wiki.installgentoo.com/wiki/Stable_Diffusion https://sdtools.org/ Bits and pieces that may be useful https://civitai.com/ Where to go to get models TI's Lora's etc, protip make an account to see the NSFW stuff https://8chan.se/v/res/818677.html Lora's are good for whole picture effects, like making girls suck dick or everyone have bigger better tits or whatnot, they will however effect everyone in the pic. Textual Inversions are more facially focused for the most part so they can be put on any form but some people include body shots in TI training data so those come out looking more bodily like the real subject, though those weights can be modified in prompt Presuming we have a TI or called 'EmWatty' then (EmWatty:1) = Full effect Emma Watson (EmWatty:0.6) = 40% less Watsonified (EmWatty:1.3) = Hyper Emma Watonsification Loras work the same way but will often fuck a scene up at 100%, usually 0.6 to 0.2 is enough with some of them. Also the old threads: https://8chan.se/ais/ Have some useful info but installing the base program now is much easier all the Python shit is included in the package so don't go following some early vid for installation tips, it's currently all working properly, though updates a can still bring up the odd issue such as upscalers not working until their creator updates their code, might work one day and not after the latest 'git pull'. There's the shithole of reddit too but I only scan that to see if updates have caused any temporary conflicts or if there's anything insanely new I've not dug into. You need a Nvida card with 8Gb (ideally 12+Gb of Vram) to get shit done at a decent pace also you need to update your CUDA memory shit to 11.6 (I think) and install Visual Studio 14 to get the speed up. I've got a 3080 doing 30/32 steps of 'denoising' per pic and it's doing 8 pics every 33/35 seconds, which is 3 times faster than when I started so they're optimising well, it's quicker to make as many shots as your card can handle at once without running out of memory rather than doing one at a time, also you can always right click the Generate button and pick 'Generate forever' which is dangerous to the balls. Get ControlNet, Latent Coupling, Composable Lora, OpenPose and maybe Style Pile extensions (they're found in the UI under the Extensions tab) there's plenty of other useful ones but those are the most used ones for me anyway. This is pretty far down the line but you may need to edit your Launch.py file found in the webui folder in Notepad for shit to work without giving you faults or being slow as shit, the line that reads >commandline_args = os.environ.get('COMMANDLINE_ARGS', "") Should read: >commandline_args = os.environ.get('COMMANDLINE_ARGS', "--opt-sdp-attention --no-half-vae") The first makes the pics be generated more quickly the second prevents an annoying error that can fuck up long processes. The annoying thing about that is you need to install MINGW64 from github and every now and then you'll need to do a right click anywhere IN the webui folder and select 'GitBash Here' that brings up a command window where you type 'git pull' and it automatically updates. Thing is if you have a modified Launch.py you need to take those commands out to get the updates to apply (then re-enter the commands once it's done), I guess you could just keep a clean copy of Launch.py and swap it in and out, hardly difficult to do just just a pain in the ass. It can get a bit overwhelming/'Oh for fucksake work!' but it's well worth it to persist, it's completely replaced pictoral porn for me, I can make better than the real world can conjure up now, deffo have a look for videos on optimal settings but lean toward more recent ones as there's a lot of out of date advice out there.

>>16078 Dang yeah that seems like a lot lol Well, I'll probably try eventually thanks for the many resources!

>>16080 It's a bit of a pig to get going at full tilt but once you're there it's nice. I recently took 18 shonky looking poorly taken old pics of a girl I used to like off FB, tidied them up a bit in photoshop, upscaled them and then made a TI of her face, also made one of my bedroom. Now I have a 95% (occasionally 100%) accurate representation of her and can put her in my room, on my bed, pregnant, masturbating, I can make her happy or sad about the situation and can pose her in any position I like while I do it. I can even bring in a basketball team and let them have at her, and her twin and more, so fucking much more, thousands and thousands of pics of it. I've saved about 20k pics of all manner of porn since I started, and while I've probably made 10 times that amount that's still a hell of a lot better than hunting in the wild and the subject matter is astonishing if you get your prompt and checkpoints right. Seriously, SD is The Best Thing™, nothing but drugs and a REALLY good series or film has ever had my jaw drop so much and given that it's just gotten started it's gonna give drugs a run for their money, it's Pandoras Box for creating perverted materials, the potential is infinite already and it's growing daily, some results *may* horrify now and again and there's a lot of slightly fucked shots but the winners are so good it's beyond beyond amazing. I'm showing this to my friend who's visiting from Australia and the number of times I've made him go: "What the FUCK?!" and "JESUS look at that!" has been quite entertaining to see. I spent 3 grand on a new computer at EXACTLY the right moment and didn't realise what I was doing till I got going, I thought this was gonna be a gaming rig and I've barely played a thing in 7 months. Oh yeah get some good fans and maybe a directional vent on your PC to exhaust away from you too, the heat this fugger puts out while producing could put a man in a coma.

Every time I go into work I think, 'None of these poor cunts knows about Stable Diffusion, they haven't got a clue, I've got the keys to the kingdom ahahahahahaaaa!' and if I tried to explain it to them, minus a fagphone, their heads couldn't grasp the enormity of it, hell mine still hasn't and it's 7 months deep. You start out thinking it's a novelty, then you realise it's the answer to a ton of photshop issues, then you realise it has the capacity to regularly surprise you without too much input, then your brain catches electric fire as it dawns on you that this thing is eventually going to make the user a god of images (I'd have said that was a hyperbolic statement last year, not any more) and if you have a HD groaning with images of sloots (840k on my old PC) you can put them in SD too and the demon doors just keep opening and opening, you can do ALL this to ANYONE and it's only improving. Holy shit what a time to be alive!

me, an intellectual that learned how to lucid dream consistently just to get cucked by my waifus. your techfag toys are nothing!

>>16084 Hmm can you get cucked 3 thousand times a night? Cos SD can do that shit if you leave it producing for 7/8 hours.

>>16102 Also the cuker(s) could have a dick the size of your arm to throw into the bargain.

>>16084 >useless With 2 custom TIs, 2 community LoRAs and a bit of fiddling in Latent Couple I've got the hottest girl I used to go to school with and her daughter in nigh on photo-real BLACKED poses covered in jizz, might preg them for a while now too. Enjoy your dreams lad.

>>16078 >>16082 Any possible guidance (if you have any) with F222 (or something better) and inpainting to nudify pictures?

>>2638 >>2675

>>4671 >>10307

>>16188 Trying to nudify existing pics may be more hassle than it's worth, I was getting some minor success at first but it was a lot of going back and forth with the prompt while it generated nearly the same shot over and over, in the end the variable nature of txt2img won out for me. Now I generally just make a TI of my intended then throw her to the wolves in txt2img, that way you can do (nearly) anything them, easily change their body size, age, pose and such, it's so much more malleable if a lot less predictable, although now with ControlNet and Latent Couple you can make miracles with the right setting or pose. Most of the checkpoints (picrelated) I use are photrealistic or near enough, obviously ComicBabes and a few 'mix' ones aren't, not sure how much Inpaint functionality they have but I found F222 a bit shoddy looking in text2img at least.

>>16256 Yeah, i've been messing around with it, results have been greatly subpar at best. I'll definitely look into creating own TI/Loras then, i work with Stable Diffusion (automatic11111), hopefully should work fine. Out of all of those [aside from analogmadness], any you think of would make decent just regular selfies? Shit you'd just see on girlsgonewild reddit/late 2000's internet nudies Thanks again for all the help/info so far

forgot im a retard and have controlnet setup previously also

>>16261 Not sure off the top of my head, I tend toward the glossier ones aZovya being my current favourite but consistentFactor, nexGenSuperModels, edgeOfRealism and level4 have proven pretty good too, hardblend occasionally knocks TIs out of the park better than all the others too, best to just nab a slew of them and test them against each other with a few seeds. Pics related are decent TI training settings btw Make sure all your shots are face on (or near enough) they don't have to be amazingly clear just as long as you have about 15/20 you should be ok, even with 14 shoddy as hell ones I managed to get a girl near photoreal so it's very doable. You can include bodyshots but mostly you want headshots. Crop them to equal dimensions so they can be easily up/downscaled to 512x512 in the Extras tab. (I use 4x_Ultrasharp as the first upscaler and SwinIR_4x at 0.1 for the second) with GPFGAN set to 0.75) Pic1: The number of vectors per token relates to the number of training images you're using, around 20 images would justify 4 or 5 that usually all you need, otherwise it's roughly: 10-30pics 4-6 40-60pics 8-10 60-100pics 10-12 Initialization text really isn't needed. I did a faulty TI with 18 headshot pics and using 8 vectors, way too much 'weight' sow while it looked decent it was hard to get past just her face showing up in results, even forcing (full body:1.8) could barely get past her tits, if in doubt I'd err on the side of lower. Pic 2: BLIP option adds txt files describing the images, that's yer dataset. Pic 3: MAKE SURE you use the 1.5 checkpoint to train on, few times I've accidentally had the wrong one on and made nightmares in the training snapshots which can be found in: >webui/textual_inversion/images Examples of what it's figured out/mapped will show up every 30 steps in there. The results in there can be OTT, very cartoony/fever dream-ish, (I'll post some examples in my next post), you just need to parse them for the best looking ones and try out their embeddings: >webui/textual_inversion/embeddings drag and drop in >webui//embeddings and refresh in the UI. You can just set the Embedding rate to 0.004 and let it ride till it hits something you like but I use a gradual one that jumps the very early steps quickly then ramps down to sensible at 500. I've found most of the best TIs have hit at around 1500 to 3000 steps, though it can be earlier or 'a bit' later not seen any past 4000 that weren't overtrained. The number of training shots should be divisible by the Batch size, usually 1 or 2 is best (a 3080 can't hack more than 2 at once) For the Prompt Template don't use a default one: >custom_subject_filewords.txt Needs to be created and placed in: >\webui\textual_inversion_templates All it needs is: >photo of a [name], [filewords] That way it'll only try to make photos rather than renderings or paintings while training.

Pics related are training images, the early ones can get a lot of features but you're best to let it run to see if it refines and hits something good later on, at 1980 there she's nearly spot on (eyes are usually iffy in them) and that would put out nearly perfect results when viewed with another checkpoint, the 2550 one might be a tiny bit overdone but would sill put out good results. Just filter out the crappy ones and vie the good ones against each other till a winner emerges and don't forget to change your checkpoint once you're done training. >>16279 >webui/textual_inversion/images Should read >webui/textual_inversion/DATE/NAME/images

Shots made with the 2550 one.

>>16279 Thanks for this, I actually tried this last night, gotten an embedding that was similar but pretty nice quality. Would you be able to show off an example or two when creating faces (like how much of the face is in the frame?) Otherwise, after some trial and errors gotten some crazy quality nudes with the face mostly in tact. It's pretty insane. Preciated again

>>16296 You mean examples of my dataset shots? I don't have the one above to hand but pics related are ones I'm currently using in an attempt to do Louise Nurding. You can get away with slight turns but make sure the face on ones are the majority. Glad you're getting going with it, it's a fucking marvellous piece of kit.

Make sure their hair isn't jaggy round the edges, old ass jpegs might have a boxy artifact effect that can get carried over in upscaling even with facial restoration on the top of the head isn't smoothed out like the rest of the face is so they can end up looking like they've got dandruff or some weird headband on in generated shots. Quick way to get pics right is crop them roughly then hit Image/Canvas Size in Photoshop, anchor the image middle centre and equate the dimensions and save, it'll put white bars at the side but that's preferable to image clutter, cut out the hair/head if you're getting weird effects otherwise it's pretty good at getting it. >>16297 Now I look that last one should be rotated so her eyes are more horizontal.

>>16297 Oh no, the pictures you just posted are perfect. Previously was just asking for reference face photos to get an idea of how much should be in the frame. >>16299 Yeah, I kinda made a mistake of not clearing enough in editing (and i didnt upscale, shouldve done that) You talked abit about cropping/white bars, would you recommend going extra mile and just painting the entire background white so that only thing in the picture is her face?

>>16301 I generally go by: 'If their hair is easily cut out then do it'. If it's going to be a pain in the ass to do then don't bother. A little messiness with hair is usually expected anyway, I did one last night with half of the shots semi-cut out and she has a few wild strands in generated pics just like reality so it can lend a little authenticity. Try to eliminate any hair decorations/accessories though unless you want weird shit on her head.

>>16307 Roger thanks for the advice, I'll probably try again sometime and upscale the photos as well. Preciate all the help

Lmao

White people are evil. Especially white men. I mean, they didn't kill as many people as the Chinese, but no one is mad at China because they keep their people in line. Best thing to do if you're a white male is to turn gay so you can't reproduce and at least can get some scraps from the pity table that is the far left... Or anyone else who pretend their lives suck because they're a color, sexual orientation, gender and species they don't think they're supposed to be.

>>16633 Best thing you could do is dilate

hjk

>>2113 His dick is so cute omg(⁄ ⁄•⁄ω⁄•⁄ ⁄)⁄ I wanna put it in my mouth until it starts twitching like its nuzzling me(/ω＼) or have him ram it inside my boypussy owo hehe ~ bbc-sama is happy to see me!!(＾ワ＾) and the most adorable thing ever is when his sperm comes out!(๑•̀ㅁ•́๑) but when bbc-kun and boypussy-chan meet and theyre blushing and all like "uwaaa~!" (ﾉ´ヮ´)ﾉ: ･ﾟhehehe~bbc-kun is so adorable (●´Д｀●)・

>>16978 >>386

>>17187 >>2675

>>16281 Once you have the TIs created, what sampler, model and prompts do you use to generate these so that there's actually ? It's honestly amazing what you've managed to make here

>>17330 actually interracial fucking/cumshots*

can you enable thread creation without files? i would like to share some links but because i use tor i can not create a thread here because i can not upload files. or just create a thread titled "bbc cuck pov roleplay" or something like that and i will post my links in there. that shit is my drug and i hope others have some similar videos they like.