A femdom thread with less of the gentle femdom, weird racebait cucking, and borderline-vanilla pegging that seems more popular these days, and more restraints, terrorizing, and all-around meanness. That said, it doesn't need to be wildly extreme or anything. Most femdom is fine, just going for more of "don't spam 20-page manga chapters of a girl basically having vanilla sex with a kind of shy and reluctant boy while acting bossy" kind of thread.Male, female, or futa subs/victims all accepted, as long as it's an anatomical female calling the shots! Domme-only pics that give BDSM vibes also okay.
Optional but much appreciated thread sidequest: give OP remedial lessons on proper femdom /d/gen-maxxing in 2026, specifically how to AI erotica. I've basically opted out of everything to do with AI for the last four years out of genuine disinterest, but my life and health have gone to shit this year with few realistic prospects of improvement, and wallowing in my fetishes is almost the only thing I have left now that seems worth doing. The AI porn boom looks like it could drastically improve that, but I'm showing up so late to the party, I don't even know where to start. Literally I just know that there are supposedly very good ERP sites now, but I don't know which ones are good/trustworthy/optimized for my interests, and I know that people can easily make porn media with text prompts somehow, but also don't know which programs for that are worth using or where to start. I'm stuck in nolife-mode fapping like it's the 2010s here, send help at once.Inb4 check the diffusion general, etc., not a word of that makes any sense to me. I need someone to explain me how to start on this stuff like I've been in a coma since 2021.
>>11589061Ideally you're going to want to have a decent PC with at least 12 GB VRAM for this stuff. If not, sorry, sucks to be you. Local models are best for this stuff because of a lack of censorship.>ERPLurk these threads: >>>/g/110016810Download and use koboldcpp, SillyTavern and Gemma 4 26B. Use chub.ai for cards.>imagesLurk these threads: >>>/g/110016925Use a Forge variant rather than ComfyUI because it's much easier and more intuitive. It's been a while since I've genned so I'm not sure what the anime SOTA model is. Probably Illustrious or Anima based I imagine. Lurk, read the archives.>I need someone to explain me how to start on this stuff like I've been in a coma since 2021.Honestly this would require like 10 wall of text posts so nobody is going to do this. Lurk the threads and read past threads on Desuarchive.
>>11589815Thanks for the info, I'll start reading up on that stuff this weekend.>Honestly this would require like 10 wall of text posts so nobody is going to do this. Lurk the threads and read past threads on Desuarchive.It's sad, because that wouldn't even have been seen as unusual on 4chan in years past, but you're definitely right about that. I appreciate someone responding at all, I honestly expected that post to go un(You)d to bump limit.>Ideally you're going to want to have a decent PC with at least 12 GB VRAM for this stuff. If not, sorry, sucks to be you. Local models are best for this stuff because of a lack of censorship.Yeah, I'm working with an eight-year-old 8 GB unit. It seems to be around lower limit of what's technically possible to use, but not the type of card that would be reasonable to found this hobby on.I've been looking into things on my own it since I made that post, and I guess my impression so far is:>imagesI now see why people just talk about Stable Diffusion, because it appears to be about the only thing around worth using. And even that you apparently need to jailbreak. The contrast between the general public moralfagging about an AI pornography epidemic and the absurd lengths to which these companies actually go to lock users out of it is kind of ridiculous.I found a site called Perchance that has pretty limited if any potential for actual erotica, but you can at least quickly generate images of cute girls and chat with rudimentary incarnations of their alleged personalities, so for now, I'm planning to just mess around and learn some basics on there, since it's free to use and doesn't even require an account.
>ERPThis seems WAY less impressive than I believed two days ago. At first I was overwhelmed by the hundreds of sites offering this service, but what I've gathered so far:>most of them try to cockblock you from doing much weird kinky shit with the models, and therefore are dumb and boring>basically all of them charge more than they're worth to use, and their free trials don't offer enough to let a noob like me see if it's even worth a dime or not>every single one of them is probably a scam anyway that lets you do like 10% of what you could make on your own if you're willing to learn a bit moreGiven my limited VRAM, I might try text-only ERP with the LLMs, but even that feels less than I'd hoped for from what I'm reading.>choice between being in a never-ending arms race against the censorship regime of anti-fun American models, or having an easier time with Chinese models that kind of suck at roleplaying>looks hard to get them to indulge my kinks either way, people sometimes have to fight a bit just to get them to go along with romance novel scenarios>truly besting censorship usually relies on pay-to-play third party stuff, not sure if $20 a month is worth it for a bot to talk dirty to me once in a while>you also need to develop your own style of prompts that get erotically-reluctant LLMs to do what you want, but the companies are deadset on stamping this out, so you'll have to revamp your workaround strategy every few months>chatbot memory for the RP scenarios reportedly not much better than "talk to your favorite anime girl" sites from c. 2017, users regularly have to break immersion to get the bot back on course; Claude is possibly an exception, but is also the biggest pain in the ass by far to jailbreakOverall, I don't think it's come as far as I thought, but it's probably a little early to be saying that when I haven't tried it out yet. I'm on day three of getting into this, so I'm likely wrong about at least 50% of everything I just said anyway.
I also think I just had the wrong idea about how widely-adopted the technology actually is at this point. I was thinking I was a total dinosaur already left in the dustbin of history, but actual consumer-level AI adoption in 2026 seems to be largely confined to various dedicated nerdy subcultures. After just the last couple days of reading up, I now believe that most people who brag about how they "use AI regularly" like they're a cosmopolitan citizen of world of tomorrow actuallyt just mean they ask AI dumb questions and to do Alexa/Playstore app-tier petty tasks for them a lot. The scale of adoption for people who actually understand how to do anything interesting with it (including making porn) seems more on the level of home internet-adoption around the time of Eternal September right now, so I don't think I'm quite as behind the times as I initially feared.
>>11590003>>11590004You'll be able to imagegen at standard Stable Diffusion resolutions with 8 GB VRAM but may struggle to gen at higher resolutions.You could give Gemma 4 12B a shot with 8 GB VRAM. Gemma 4 is, surprisingly, pretty damn uncensored. You won't be able to get good speed+context length out of the 26B with just 8 GB.
>>11589061>>11590067Also, re: femdom gens, be forewarned that doing gens with more than one character is Nightmare difficulty. 1girl is easy. More than that is hard as shit.Controlnet is your friend, stuff like Canny or Anytest. Feed Controlnet some selective lineart and it will spit out the characters you want doing the things you want after enough seed rerolls. You can generate said lineart by tracing photos, video stills or 3D renders. Koikatsu can be useful for this.
>>11590067I'll give both a try soon, then. I'm somewhat retardedly enamored of Perchance at the moment simply because I got it to work right away and am already having pretty good luck with keeping consistent character appearances and generating more softcore fetish content for myself. With my level of knowledge, it'll probably take me a day just to figure out how to even get Stable Diffusion or Gemma 4 running in the first place. From what I read, it looked like 512×512 was the best I could hope for from Stable Diffusion, but that shouldn't be much of a real limit if I just pump the pics I like into an upscaler.If I'm correct in my read of the Diffusion threads that you need some third-party Github shit to generate porn with it, then figuring out to marry that software to vanilla Stable Diffusion is probably going to take me a while as well. Does Gemma 4 12B not need that kind of jailbreaking to get going with?>>11590073>Controlnet is your friend, stuff like Canny or Anytest. Feed Controlnet some selective lineart and it will spit out the characters you want doing the things you want after enough seed rerolls. You can generate said lineart by tracing photos, video stills or 3D renders. Koikatsu can be useful for this.I'll have to take some time to learn what most of that even means, but nice to know Koikatsu's useful. I actually already have that, and also have been dusting off my old Illusion games the last couple of days as I begin this training arc. Sadly, Koikatsu is the one I've barely used; I'm better with AI Shoujo and MUCH better with Honey Select 1.
>>11589061RE: image genning: Learn how to Gimp/Krita. You can photoshuup your source images for image-to-image and feed that to the AI. I don't know if I'm supposed to post this here, but here is something I genned during the summer.I only have time for a very short and brief explanation, but I genned Bea separately, then I genned Nessa separately, used Krita to erase their backgrounds to make it transparent, then copy/pasted both into the same image with a transparent background. Leave the BG transparent and save it with just the characters in the image. Use PNG exclusively to preserve the alpha transparency. Feed this photoshuupped image to yon image-to-image AI generator and describe via prompt exactly what you want to see and how you want the background to look like. For this, I explained to the AI that there should be a tropical paradise resort with palm trees in the BG and that behind Bea's back there should be that large wooden cross that Bea is strapped to with thick leather bonds. AI genned all the background stuff for me and this came out.So generate your characters one by one separately in the poses you want to see them in, copy paste to a blank image with a transparent BG and feed that to AI and let it create the scenery.
>>11590132>RE: image genning: Learn how to Gimp/Krita.Seconding this. Also learn how to inpaint, and how to use an image editor to guide your inpaints. Inpainting is THE difference between a flawless gen that looks like it could be human art, and a gen with 20-30 obvious AI flaws.
>>11590132>I don't know if I'm supposed to post this here, but here is something I genned during the summer.Nice pic. AI doesn't violate /d/ rules as long as it's done in anime-style like that, so if you've done more like it on-theme for the thread, feel free to share!>>11590132>>11590207Appreciate the guidance. Right now, my main obstacle is actually freeing up enough drive space to get Stable Diffusion installed, which will probably take me a few days at least. I could try to learn some basics in Krita in the meantime.Actually, before I start sorting through my drive, do either of you know how workable Stable Diffuson is on an AMD graphics cards? Not only am already about at the threshold of VRAM, but most of what I see online highly recommends having an Nvidia product to be doing this at all, although some people seem to think AMD works fine.
>>11590004Just use F-list like a normal pervert
>>11590862I'm too shy and not bi enough for that.
>>11590623>basics in KritaPay particular attention to learning how layers, transparency, clone tools and the eraser work. These are my most used tools, though certainly not the only ones. Layers and the eraser are important, because you can use them to take two generated images and merge their best bits together. Got a bum hand in one, otherwise acceptable gen? See if you could transplant a better looking one from another gen where the character is doing the same pose. Just as long as both images are the same px*px size, you can just layer them on top of each other and see if you could reveal a healthier limb from the bottom layer by erasing away the twisted monstrosity on the top layer. A tip for Gimp: In Gimp alpha transparency isn't usually enabled by default for the layer, it needs to be enabled separately from the layer-window. In Krita, you can just use Krita's eraser to start creating transparent areas right away.>freeing up enough drive spaceYeh, you're gonna want terabyte class hard drives in the future. Loras eat space and generations build up clutter and shit really fast. Don't hesitate to nuke things that do not work.>AMDNo personal experience, but reports state that at least for video AMD is less supported. Yeah, it's all about the NVidia cards, unfortunately. Why do you think that company is making a killing right now? Their product sells like hotcakes 'cause they're the only game in town. For image gens you should probably be fine, but I don't have an AMD setup to test with so can't say anything concrete.Here's an alternative one of the same kind with Nessa+Bea: a secret meeting in a hotel room for some "pain tolerance training". My favorite aspect of this one: Nessa's smug face. She knows and Bea knows that pretty soon this is gonna suck for Bea.
>>11590984I really might just be out of my league hardware-wise trying to do this. Ah well, I'll probably still see how far I can get. At least I'll learn some things on the way to my dead end. I guess for now I can try learning some of the stuff you're talking about in Krita/Gimp on the lower-end AI shit I'm definitely able to pump out on generative websites.Your pics really make me wish I could post some of my old Honey Select dioramas they remind me of on here, but the style of the models probably isn't allowed on /d/.
