Previously, on /asdg/ >>8674880The cartoony side: >>>/aco/csdgRe-post with subject and correct linking. Sorry.No DALL-E, Meta AI or hailuo AI generated stuff.No futa.>>>/aco/futasdgReport trolls.>GuidesGeneral: github.com/AUTOMATIC1111/stable-diffusion-webui/wiki | rentry.org/sdg-link | rentry.org/rentrysdHires Fix: rentry.org/hiresfixjan23Inpainting/Outpainting: rentry.org/drfar | rentry.org/inpainting-guide-SDAnimation: rentry.org/AnimAnonCreate LoRAs, TIs: rentry.org/59xed3 | rentry.org/simplified-embed-trainingList of models/LoRAs:rentry.org/acorepo>Auto1111 extensionsControlNet: github.com/Mikubill/sd-webui-controlnet | rentry.org/dummycontrolnetRegional Prompter: github.com/hako-mikan/sd-webui-regional-prompterTag autocomplete: github.com/DominikDoom/a1111-sd-webui-tagcomplete>Related>>>/b/degen>>>/h/hdg>>>/e/edg>>>/d/ddg>>>/g/sdg>>>/trash/sdg>>>/trash/slop>>>/vg/aids>>>/vt/vtai>>>/y/ydg
Not to be a prick, but its needed to post at least 6 images when making a thread or the chances of culling are high.
>>8682784I have been working on filling the gap. Hopefully, the mods will appreciate that I am not dumping whatever to meet the standard. There have been a few complaints about people just spamming crap in the thread. This freakin pic required so much, I should have given it up. Still on the fence if I saved it. I know why I didn't downscale this, but I am still annoyed with myself.
>>8682793I just post saved pics at random.That pic looks fine.
>>8682796thanks anon. Her proportions were really off too me when I started.
>>8682806A little caricature isn´t a bad thing.
freaking pussy juice keeps washing all the ink off the totem. >>8682825definitely. I have many ahegao loras/folders to prove it.
New question: Am I allowed to fill in a name when I post? I know some boards are against that.
>>8682929You can.Just do not spam or start role playing like the dude in last thread.
>>8682929please don't
>>8682938>>8682942Eh, I won't then.>>8682939What model is this?
>>8682950biglust mixed with stanleystriweekend
>>8682989Realistically he stopped doing sex arcade a long time ago, but the dudes done a pretty non trivial amount of guro adjacent art in the past and for a long time there was a long term series of text based stories alongside sex arcade images and they were definitely not friendly to the long term health of the women. It was indirect but pretty clear that most either don't survive the experience or are left permanently harmed, as would make sense given the fact that it's a giant interdimensional rape factory. Sabu's catalog is pretty deep into edge territory. I think he's just tired of SA at this point.
Ill probably do like 1 or 2 more of these things then delete it all in shame and uninstall forge again. It's unfortunate that it's so easy to create this shit, it's not healthy for it to be so easy to laser focus into a fetish and get effectively unlimited content. That fucks with your brain.
So I guess this is the place to post realistic images no matter what their content is. It's a little odd but /aco/ was always a dumping ground for anything not real or anime anyways.In any case, have some Street Fighter ladies.
>>8683273
>>8683275
>>8683148Yeah this is one hundo percent bad for me but here we are
>>8683273AI screwed up a bunch of categories and boards. I'll say it is a hard problem to solve. Seeing futa on /aco/ is a mind blow. Blessed thread that nobody has bitched about things needing to be better.
>>8683299I think /aco is just one of the politically weaker boards and lost the argument. Theres no reason this thread shouldnt be in /s or /hc other than the fact that those boards fought harder for it. /aco is sort of a trash bin for other boards at the moment. In theory this could all be in /b but thats the pedo containment zone for now.
>>8683299As long as I'm in the right place I'm not going to complain. /aco/ has had pretty much anything on it in terms of content since it's start since it's the only board that allows non-asian drawings/renders. >>8683311/s/ and /hr/ have always banned things like celeb photohops and forced that shit to go to /b/. AI is more similar to that conceptually than actual photographs, so it makes sense they wouldn't take it. The only other places it would make sense for this thread to live would be /b/ (which is infested with loli) or /trash/ (which is infested with furry). So really I'm thankful it's here and not in either of those boards.There probably should just be an /ai/ board to contain all the ai threads like what happened with /vt/, but most boards seem to tolerate the it so long as it stays in it's own thread anyway so probably not much incentive to do that.
>>8682929It makes you look like a giant attention seeking faggotYou can adopt a filename pattern that sets you apart a bitThen at least you have plausible deniability
>>8683439if you don't like the attention seekers you encourage them to name trip. Makes them easier to filter. Although I suspect that you are just drama baiting. In either case you are doing it wrong.
>>8683427>There probably should just be an /ai/ boardsomeone should get one of those LLMs trained on greentext to simulate that board and see how that goes.
>>8683464Why waste time shitposting yourself when you can make a bot do it for you?
>>8683467>>8683469So what's the story behind these anyway? Did Lex run out of money and start trafficking metahumans to sell to other billionaires?
hahahahaha
>>8683590Essentially, yeah. I'm trying to avoid being too obvious with any sort of backstory. The pages are supposed to read similar to a high end product catalog. I have no idea who the reader would actually be.I decided to add some text to an image then just kinda kept going with more.
>>8683590Its basically the plot to the something unlimited porn game but a lot less vanilla. Also, what was the prompt for that one, its great.
Is there a mirror for acorepo? I'm trying to find a decent checkpoint for realism.
>>8683619>>8683625Yeah that's sort of what they reminded me of. I like them though, having the story on the side turns what would be another "1girl standing" image into something with some actual creativity involved.Here's a catbox for the original. I had to edit the hands and face, but the pose is pretty great.https://files.catbox.moe/znjpoy.png
>>8683643Dang my prompts are way too long. Mine are all like 5 times that length.
>>8683637What checkpoint is giving those legs?
>>8683646You just made me realize the knee is fucked up, goddamnit. I'll never not see that now. The checkpoint is realismIllustriousBy_v20FP16.
>>8683643What does override setting emphasis no norm do? I havent used overrides at all before.
>>8683643>>8683650Also 60 sampling steps seems like a lot.
Weirdly using those prompts mine come out cartoony, even with the checkpoint that emphasizes realistic images.
>>8683643What am I missing with these catbox links, it's just a png
>>8683644Lol, I have a few with more involved prompts but I kind of like letting the AI have some more "creativity". Sometimes you get things that you weren't expecting and can be better than what you had been going for.>>8683650No idea, it's just something that reforge adds for some reason. I usually remove those overrides and I don't experience any issues with genning.>>8683651It is, I use that to get the micro details on the skin, wrinkles, vellus hairs, pores. I could probably lower the step count a bit but it still only takes about 30s per image on my machine. On non-realism gens I usually use 30, but realism seems the benefit from higher step counts.>>8683656PonyRealism tends force a realistic style so I don't usually have to specify anything beyond some basic content and posing. I find that a lot of the work can be taken out of genning just by using the right model. >>8683668You extract the metadata stored in the image to see how it was genned.
>>8683668in webui and forge you can just drag it onto the "png info" tab to get the info.
>>8683656the prompt is neutral. Add a source_realistic or other realistic keywords>>8683650>>8683670Without digging in too much I believe it is for weight normalization. Short answer is forge will prioritize words at the beginning of the prompt. It seem reForge does mean normalization.
hey using stanley anon. I am finding it will just kick your saturation into overdrive and not much else. Pic related, it doesn't seem great for realism. Probably handles 2.5 like a champ. Your pic was good. This pic is not. Are you using it for initial gen or can you make it at least not color everything red some other way?
>blood appeared in a gen when I didn't ask for itOh
What would her father think if he knew his little girl was sneaking off during class to play with herself in the school restroom?
I chose to keep the blank toothless stare. Please don't judge me. >>8683835the random animal eyes (or half eyes) when using a fabric print style is the one that creeps me out
How to tell Forge or a1111 to not show the preprocessor images in the results?
any grok guides
>>8683872I remember back in high school a (female) friend saying the girl sitting next to her confessing getting wet during lit class, so whatever.>>8683915Just say hypno. And fuck when the actual element shows up when prompting clothing print, ie actual roses in rose print.>>8683997Check the archived threads. Some dude posts a lot of prompts.
>>8683872Her punishment is that she has to sit like this while daddy watches his show, and she isn't allowed to touch or play with it at all
So what's the secret to having two completely different women in one picture without all their characteristics merging
>>8684336This list is from least to highest effort and least effective to highest effective.1. Use Break Statement2. Use Regional tech - Forge couple is the best3. Comfy workflow with segementation/masking/inpainting4. Gen two characters, use optional segmentation, copy/paint one into the other, inpaint arms or whatever to merge the two. I normally stick with 1 unless I am very motivated
>>8684406>Forge couple is the bestNTA but I'll have to check that out, thanks. I tried installing the regional prompter plugin on Forge and it shits itself and doesn't work and I haven't given enough of a fuck to try to fix it since I never did get good at using it before I moved to Forge
looks like flux is definitely sfw or semi-nsfw.
>>8684406Break statement doesn't do that stuff, just read the SD manualUnless you are using it with regional prompter of course, in that case read up on how to use that
>>8682691i wish i wasent retarded and could get this shit to work
>>8684598What's your GPU?
>>8684613have an rtx 4080 just meant that it seems very overwhelming to get into all this shit with python stuff the github stuff just all off it i might just be rettarded tho lol
>>8684639my dude, it's literally a 1 click installer inside a self-contained folder.you just need to install Python and put the downloaded models in the right folder.
>>8683464>someone should get one of those LLMs trained on greentext to simulate that board and see how that goesLike this? https://glif.app/@001/glifs/clzymbou80006138uf1z6q7kl
>>8684545>just read the SD manualThere are 100s of manuals at this point so I can't know what you are referring to. I'll try not to assume you are just gaslighting, but provide some evidence to back up my claim. Pic related. Left is no breaks, right is with breaks. Prompt (with BREAKS):score_9,score_8_up,score_7_up,source_realistic,rating_explicit,2girl,hugging,embrace,kneeling,BREAK1girl,blonde hair,blue eyes,white sundress,huge ass,BREAK1girl red hair,green eyes,blue dress,huge breasts,I get results. No Breaks:blonde/white dress: 2 - 1 = 1 question person wanted different people. Negative points for both in white dressblonde/huge ass: 2red/blue dress: 1red/huge breasts: 3white dress/huge ass: too many white dressesblue dress/huge tits: too many white dressesBreaks:blonde/white dress: 2blonde/huge ass: 2red/blue dress: 2red/huge breasts: 1white dress/huge ass: 4blue dress/huge tits: 3.5But clearly I screwed up the testing as the images have different resolutions. I'll do another test shortly.
>>8684864epic AI posts to send to your facebook uncle with dementia
>>8684838got bored, distracted, not bothering with followup.Enjoy a forge couple pic>>8684879>facebook uncleis this a new daddy term?
After a full day of fucking around, it feels like the best results come from generating the initial image with a single character, then inpainting the second character afterwards. Whether or not this will still prove true as more complicated scenes and intertwined poses progress remains to be seen
if some footy could help me out with a foot inpainting model or something that would be most appreciated. I tried to gimp the feet into compliance, and I am still a work in progress>>8684921I would agree. If I am going that route I'll use a controlnet to keep characters roughly the same size/shape. Not sure why it, but it always seems if you warp their size the final results isn't as good. It is still a crapload of work.
>>8684838Literally read the fucking manual you retardhttps://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#Break-keywordYou are using this incredibly complicated piece of software without knowing how any of this works
>>8684956so it changes the tokenization boundaries through padding. Sounds like something very useful if you want to keep subjects aligned. I learned nothing new. Hopefully you did.
>>8684958You seem to be under the impression that cross attention doesn't work between chunksYour anecdotal evidence doesn't change technical factsRoll more seeds and you'll see that it still bleedsNo, I did not learn anything from your little post, but I learned a lot from reading the manuals of SD and its extensions, making 60 thousand gens myself and listening to people when I was still new
>>8684961>You seem to be under the impression that cross attention doesn't work between chunkswell aware. >Roll more seeds and you'll see that it still bleedsyup>making 60 thousand gens myself and listening to people when I was still newpuff up your chest more. Maybe I might get scared and run away. So far your grasp of logic has been minimal and the only reason I am replying at this point is that you are a no nothing bully. That was a list said the BREAK statement was the least effective. You seem to be under the impression that I think that BREAK statements will magically solve this issue. It helps and it isn't that effective, but it does do something. Stop putting words in my mouth and return to putting cocks in yours.
>>8684962You were spreading the typical beginner's mistake that the BREAK keyword does any separation (it's mentioned in too many tutorials by mistake)I made the exact type of post like you 6 months agoKeep it up and you'll make really good gens soon, but learn to take simple advice betterNo one cares about egos here, we're all anonDon't take shit personal
>>8684976>Keep it up and you'll make really good gens soonI have been. >No one cares about egos here, we're all anonClearly you do. Mister 60K>Don't take shit personalYou insulted me over and over trying to bully me into your incorrect opinion. Now that your personal attacks aren't working you are trying to back out. Get fucked. >You were spreading the typical beginner's mistake that the BREAK keyword does any separation (it's mentioned in too many tutorials by mistake)There is that I am a beginner bs again. Padding is separation. Tokenization is a thing. Take your holy war against bad tutorials somewhere else.
>>8685005love me some good Taila. Sorry if it isn't and I am just hoping too hard.
>>8685007In effect, it is.It was a cartoon-with-real-checkpoint test.
A couple more test Talias.
Rules>No Futa>Woman helping male rape a victim>Woman is cheering onPreferences:>Creampie/breeding victim>POV perspective>Crying victimDisclaimer: This is a fantasy. I do not endorse doing this in real life to anyone.
>>8685126i don't get it. not disgusted or anything - to me this is just unfunny cartoon, sexualised, but devoid of anything erotic.but im actually curious where one would see the appeal ?would photorealism make it better or worse ? does the male need to be a faceless stand-in ? is the xray bubble necessary ?
>>8685138breeding kink like most kinks is based on dominance. The immediate dominance is obvious. There is also the idea of building an army people just like you who would obviously follow you. Like most of this stuff it falls apart quickly when you think about it too much. Generally, people wouldn't like absent fathers and you would be building an army of enemies. >would photorealism make it better or worseworse. >does the male need to be a faceless stand-inyes. Kink falls apart if it isn't your army. Watching some dude rape a bunch of chicks isn't sexy, unless you are really messed up.> is the xray bubble necessaryI would be very interested if anon posted a realistic version of this this and included the workflow. I have no idea how I would approach it.
>>8684022need moar office sluts
Now have my gen time down to under a minute for a 5 second mp4 on my 8GB 3050 using ComfyUI and Hunyuan plus LORA, It's like having a genie granting perverted wishes except it's not a lamp i am rubbing
>>8685192Was playing around with something that would fit that theme earlier.
>>8685352nice
>>8683451>>8683452>>8684270>>8684536moar with penis
>>8684536>>8685511
Has anyone looked into using different clip_L for image gen? I am going to try Nepotism_V5. I am still trying to figure out how to evaluate between it making a difference and just being more random.
>>8682785more
>>8686180>>8686181wft is wrong with you
>>8686186>>8686102 asked for more, more was posted. That´s what I saved from that set. There were worse things.
>>8686186>Stop liking what I don't like!This thread is not your personal gooncave, skip what you don't like, or post more of what you do and let people enjoy things.I personally find the style iteresting
>>8686197could you replace the nipples with valves, like in an air mattress or floating device? ^^
>>8686500>That´s what I saved from that set.If I had inpainting/img2img capabilities, they would have been used
>>8686180>>8686181>>8686197thank you, saved.
Once the prey ceases struggling, the Paraguayan Cockvine probes for body cavities. The intrusion is not usually fatal, and the plant will generally release its grip after three to five hours.
Spending hours upon hours tweaking a single image just so that I can look at it later and get horny is probably normal
retard noob here. trying to train a lora to be as realistic-looking as possible. am i correct in thinking that i need to train the lora with a mode like DreamLike-photo real or Realistic Vision instead of just stable diffusion 1.5? then i can generate images with biglust as the model and get realistic nudes?
>>8686833model*
>>8686342Bro, that shit belongs in /d/
>>8686342it would be terrible if an anon judged somebody for their preferences. Especially if they liked something personally.
>>8686833You are correct in that you shouldn't use a vanilla model for trainingThe resulting LoRA will work best with the checkpoint that was used for trainingBut BigLust is an XL model, so your 1.5 LoRA won't work with that
>>8686905thank you. so i should use biglust for training the lora since that's what i'll be using to generate images? i picked biglust because that's the one that looks most realistic to me. maybe there are better ones?
>>8686930It's a great one for what you are trying to doTraining an XL model is a lot more resource intensive though
>>8686839Anything with realistic humans goes here. That's how every other board has decided with their AI threads and the only exception to that is the realistic futa thread. Bolt-ons and fetish wear aren't futa, so they go here. You don't like it, then start your own niche content AI thread and watch it die when you're the only one that posts in it.
>>8686977>Images depicting bestiality or scat are not to be uploaded. Unless that is a human penis.Just calm the fuck down, give out your (yous), let others give out their (nos) and stop being a little bitch. This entire thread is full of violations.
any tips for getting an "arms wrapping around breasts" sort of pose? i figured out the prompt/keywords for it at one point on a whim and then completely forgot how i did it shortly after. driving me crazy
Any advice for minimizing background/body bleedthrough when training a face lora? I've tried blocking out the background but it makes the lora want to give me blank backgrounds if I set the strength high enough to get the face right. I've tried inpainting the face using an earlier version of the lora over various other bodies and backgrounds but it degrades the face too much in the resulting lora. I have a decent collection of photos but they were all taken in the same house/backyard and obviously all have the same body along with the face. Should I just only use close up pics of the face? My concern is that it'll make the lora want to generate portraits rather than full body pics.The person I'm trying to lora is aware of the end goal so if it's a matter of more pics I can do that. The goal is a lora that gets the face right but lets the prompting generate fantasy bodies to go with it. They want an easy way to make pics of themselves as their D&D character [spoiler] and the payment is letting me use the lora for personal things [/spoiler] so I'd very much like to get it right. I'm running a local kohya with SDXL and 16 gigs of vram if that matters.
>>8687009>Quoting rules you don't understandJesus, you really are a newfag. Bestiality means actual animals, not tentacles. /xeno/ has been going on this board since it started. The current OP image of the freaking drawthread is tentacles. There is a goddamn /plant/ thread right now and there have been tentacle threads in the past that have lasted for nearly a month.The jannies are perfectly fine with all of this shit. You're just another sperg that thinks their taste is the only taste that matters. If you think there are actual violations in the thread then report them. Otherwise shut up and gen.
>>8687046"covering breasts, crossed arms" seem to work pretty well. Unless you were looking for "self hug" but that one seems a bit more random.>>8687070Training is way beyond what anyone her is going to be able to help with. You're probably better off asking for help from the >>>/g/sdg thread.
>>8687072whatever Mr 60K. Nobody cares about you opinion or you being the 4chan police. You stick with 2.5D and not realistic because of a skill issue. You terraforming this thread isn't going to change it.
>>8687094>covering breasts, crossed armsawesome yeah those seem to work very nicely. was getting close results with "breast hug" but adding those keywords as well made it really consistent. and adding other things like underboob adds some good control to it as well. thanks dawg
>>8686977>>8686342>>8686839>>8687072>>8687121I think you're all retarded, if that helps
>Original file was too big to post>>8687046Funny i was also trying to get that pose. Couldn't seem to find any specific phrase that convinced the AI to cover the nipples do I just had to gen the post over and over until she actually did it
>>8687165I think that's a fair assessment
>>8687046Did you lose those gens? Just look at the prompts, it's all in the PNG meta data (drag file to PNG Info tab)>>8687070You can inpaint the background a bit to look more "generic"Also change the background a bit and keep both images in the training setTraining works like a giant Venn diagramTo reduce bleed during genning, lower the LoRA weight if likeness doesn't suffer and always specify a background by promptIf you leave it blank, it will always fall back to something it knows from training>>8687094There are many people with training experience hereIt's exactly the place
