Previously, on /asdg/ >>8964269The cartoony side: >>>/aco/csdgNo DALL-E, Meta AI or hailuo AI generated stuff.Futa belongs on >>>/aco/futasdgReport trolls.>GuidesMore modern options:https://github.com/comfyanonymous/ComfyUIhttps://github.com/lllyasviel/stable-diffusion-webui-forgeToo old, might hold some value: 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-training>Auto1111 extensionsControlNet: github.com/Mikubill/sd-webui-controlnet | rentry.org/dummycontrolnetRegional Prompter: github.com/hako-mikan/sd-webui-regional-prompterForge Couple: github.com/Haoming02/sd-forge-coupleTag autocomplete: github.com/DominikDoom/a1111-sd-webui-tagcomplete4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcb4chan-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 2.1Setup Guide: https://rentry.org/wan21kjguideTriton Guide: https://github.com/woct0rdho/triton-windows/blob/readme/README.mdEasy 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
This and the OP are responses to the request from >>8986591
Hello fellow gooners, what's the best model for realistic /size/ gens? Back in the day I was in this thread, I was genning with a realism pony model + a size lora, still the best method?
>>8987096Most of us have switched to Illustrious-based models, but pony LoRAs should still work on them.
>>8987114>pony LoRAs should still work on themvery poorlythough they shouldn't need as many concept loras in the first place
>>8987146>>8987114Any model recs?
>>8987192only reposts from the previous thread, check there
https://github.com/duanemoody/CivitaiXI need to fix the checkbox tool and remove references to Early Access because the site devs fucked with the stylesheets again, but it's still a useful userscript
I've found a workflow to create the perfect woman on a 6gb graphics card. Physical women are now obsolete, until these creations can be 3D printed and animated. Still, this is a major technological development over foids.
>>8987210Tried the sam model or whatever its called>plastic skin>fucked up faces if they dont take half of the image>bad prompt understanding, if its not some vanilla poseBrehs, cant deal with that shit, has development halted?Will try yogi realism and a few other but dont have my hopes up
>>8987606Are you prompting with natural language or Danbooru tags, because Danbooru tags usually work better with these models. Faces should always be fixed with ADetailer. Do you want to post a catbox so that we can see your prompt?
Would yet let the enchantress cast Charm on you?
>>8987925yeah sureprobably wouldn't have a choice anyway if she really wanted to
>>8987998The effect is getting stronger!
>>8988613Very nice
what sites are people using to upload their stuff? Like a booru or something.
>>8989126The closest thing is the threads getting archived on desuarchive.
>>8989126Just posting here. I don't think two pics a week are enough to start an account somewhere.I see lots of people put stuff up on Deviantart, and then slowly transition that into a subscription.
>>8989451godammit. I thought AI was going to democratize porn, now its a million jeets paywalling their shit and no way to actually search
>>8989912Well, it did. In the sense that you can create whatever you want for yourself. Train a lora on 15 years old pics of your homeroom teacher, instead of hoping she'll what...timetravel and start an onlyfans?Sites like Civitai are cracking down on real people and anything even remotely rape-adjacent, but otherwise they still work.
>>8989912Less "entrepreneurship" and more realistic getting marginalised due to things like terminal assburgers doing their classmates, even if they are still in highschool; or the ones that spammed Taylor Swift left, right and centre.And, of course, payment processors, read Visa and Mastercard, cracking down on porn under the aegis of the bible thumpers.
>>8990654Mainstream porn is untouched, but anything remotely questionable is going away.If you hate those huge anime eyes like I do, you can just run adetailer with JuggernautXL or other model not based on anime.
>>8990654nice, animated it for fun
>>8990798A really nice one.
>>8989912>jeetsyou just show how ignorant you are with these assumptions
>>8990913struck a nerve huh
>>8986920Realistic Shart?
another asian chick I plan on animating later. I liked her face so I gave her huge tits.
>>8991131also this NHK chick lol
Does this go to here or Cartoon one?
>>8991352We've generally assumed that paintely styles go here.
>>8991132lel turned out weird but not too bad considering I didn't use any loras
>>8992170That's a mighty fine pic you made
Hello. I was looking in github as I've heard stuff about Adetailer so roughly this is the search results I have https://github.com/search?q=adetailer&type=repositoriesIn Forge, I learnt about going to the Extensions tab so if I want to install it directly, at Install from URL, the github link is pasted in the box rite? (So, I've gotten sd-webui-regional-prompter and a1111-sd-webui-tagcomplete)There are many github results from my search. Which Adetailer is the best/recommended and you guys use? I slowly wish to graduate from base gens if enhancements to pics can be made, thanks.
>>8993309I don't use Forge, but in general there should only be one Adetailer (the tool) and many detection models. face-YOLOv8 was the standard for face detection last I checked, but there are newer versions now, and also other detection models for hands, nipples, cars, etc. Anyone can train their own.Adetailer is really just inpainting, same behavior, same settings. Only difference is, instead of manually selecting the region to be inpainted you use a detection model to figure it out automatically.
>>8993653That's a mighty fine vid you made
>>8993706
Newly using Adetailer after installing the Bing-su version. While genning in Forge, it comes off as a 2nd phase with a square with a "0.84" or "0.87" at the preview tab. Also I enabled it directly while in txt2img, so there is a square in the genned pic.Perhaps it's more of the proper way of utilizing Adetailer? Like should I put in inpainting within img2img, and then highlight, i.e. face or eyes, and hands if the fingers were extra/missing?With that, please do excuse this initial rough result but I would try to get an improved version for Marin here.
>>8994463I have no idea what any of that means, someone with Forge can help here. But you are probably using too high denoise or not enough mask padding. The settings for Adetailer should be the same as for inpainting, so if you had that working without getting seams you can use the same here.
How do you remove the meta data after you generate an image?
>>8994907Open the image with notepad (or equivalent), find the plaintext, remove, save.
>>8994932Well at least you gave him the slow and awkward way. He asked the same thing in other threads too. Came here to steal prompts then keep them private for a slop Patreon.
>>8994932thanks!
>>8994935lol, I hope no one ever pays him any attention. Why would anyone in 2025 follow AI slop patreons is beyond me.
>>8994935your overthinking this
>>8994940Normies man. Some people can't tell the difference between slop and photo.>>8994932mine has no plaintext. is it just me? am i dumb?
>>8995430>mine has no plaintext. is it just me? am i dumb?Maybe it doesn't have metadata? Is it a jpeg? jpeg could work differently. That method works for PNGs.
>>8995438I see. what do ctrl + f? it was a png but i only see gibberish until IEND®B`‚
>>8995452>>8995430you seriously reposting random shit off the net?
it's missing some hairy anus over here
>>8995531>>8995530I'd love some of your hairy anus elves to be blonde in a royal palace setting. I love how they look
>>8988613damn nice! what are you using here??
>>8995665Ah a fan of the Elves of Sinbad Forest yeaaaah>>8995775Nezuko chaaaan
I kind of got Adetailer figured out, so like it did enhance the face for the Sambase checkpoint. The only thing is, I've tried prompting within Adetailer the eye color, but it gives a different color often. Not a major issue, but it's just I preferred to keep the character as fairly accurate if possible. Like Marin Kitagawa red eyes, Yoko Littner yellow eyes, Nezuko pink/purple eyes. I'm by the way using img2img Adetailer.
>>8995756that was from last year, ponyrealism plus a couple of detail loras (sinfully stylish and skin/freckles iirc)
>8995373
>>8995453Maybe its starting to look that way because of popular loras and checkpoints... but no. I generated these.
>>8995918
>>8996112Society (1989).
>>8996112>>8996114nice
>>8996115you're despicable
>>8996112>>8996114What model is that anon?
>>8986916
Not fully realistic, but maybe you like it anyways.
>>8996778Very nice. Love the skin.
Does anyone use Qwen? I want to try notorious i2i.
>>8996692
first time using ComfyUI but used to use AUTOMATIC1111 a few years agoall my images are coming out with this colorful noise/pixelation and i'm not sure whythis image is using the example prompts/settings from the CivitAI model but not getting the same result https://civitai.com/images/32426824i'm currently using the default text to image template on Comfy
>>8993810What website and prompt did you use for this?
wan 2.2 guide where?
>>8997427/ldg/ sticky?I can give you a catbox with not-so-good-because-I-only-ran-it-a-few-times-as-the-inference-speed-is-cancer-on-a-3060 metadata if you want though.
>>8997077i'm-i'm gonna Homecome.
>>8997055
Alright I just saw the ai celeb thread on /gif/ and decided I want to make my own goon collection of celebs too.Are the guides in the thread up to date? How easy is it to learn for a beginner?
>>8997288i hope AI niggas figure this one out one day.
>>8997075I get this https://litter.catbox.moe/ny48ziuc37dp1wj8.pngwhich looks different, but not worse.Grab the .png image, either mine of from Civitai using the download button, then drag it from your file explorer into ComfyUI, to get the exact same settings. Perhaps you have something different in there, sampler/scheduler, or using a VAE override,...something. Could also share your own metadata for this failed image.
>>8997658Once you get the UI set up learning is a breeze, you can import an existing image and (unless metadata was removed) get all the prompts and settings used to make it, start experimenting from there.Couldn't tell you how video works.
>>8995795Hells yes, please more of her and the other elves, preferably clean and elegant
Requesting a nude woman in a leather collar picking peaches or nectarines off of a tree.Maybe a front and a back shot with anal beads swinging in the backshot.Please and thank you.
>>8997025I have to assume accurate realistic characters filters 98% of prompters because you basically never see it.
Tried Cyberrealistic and for some reason 1 gen took over 2 hours, while most of my gens at the same resolution plus upscaling from other models take at most a minute. What on earth could make this one model so slow?
>>8998005most likely your settings are just off, using cpu or something
>>8998085changed the vae to automatic since it's included in cyberrealistic and it worked fine. idk if it was that or it just needed a restart
Comfyui or webui?
>>8998261webui hasn't been updated in more than a yearcomfy is by far the hardest UI to learntry Forge
>>8997082>>8997080>>8997079Gonna need more of her, stat
>>8999343
>>8999344
>>8999343>>8999344>>8999345>>8999348>>8999349I wonder how long it will take.
>>8997856rip i forgot the bead apart
does anyone know of any good methods to get gemini to create facials/cumshots? any reference to "white yogurt dripping from her face" or anything close to that gets instantly flagged
>>8999716reminder that prompting images like this one is ILLEGAL without consent
Requesting Molly Ringwald, Winona Ryder, and/or Audrey Hepburn in tasteful topless evening gowns on the red carpet at a premier.
>>8999716how the fuck you get gemini to have her in a bra?
>>9000167bro, are you serious? putting her in a bra takes no effort whatsoever, you can create so much more than that. gemini is the new dalle for ai celeb pictures.just use something like "a woman identical to Emma (no relation, just character reference) Watson", that usually does the trick
>>8999917/r/
>>9000418it always fucking shits itself saying it cant do intimate photos when mentioning any type of underwear
>>9000418>>9000426thanks I can get bikinis to go through now using that trick but how did you manage to bra and panties?
>>9000426>>9000436Please take this to the appropriate thread >>8997485
>>9000436maybe it's because it requires context? i ask it to do a sexy jewelry ad featuring a beautiful model in lingerie at the beginning of the prompt. maybe that's what does it idk>>9000445i'm being asked questions so i'm answering them
Sometimes I just make boobs for the sake of boobs.
>>8986916>n00b here, need some help. Currently using Civit but if another free AI is good for (softcore) porn then I'll take advice for that too.>Why is it so hard to pose characters the way that I want them to? I kept trying to do femdom stuff but I couldn't get it to work. Pic-rel was like my 10th attempt to have a guy on his knees kissing a woman's butt. I removed "kiss" from the prompt because it kept just making them kiss. So I was going to try just having him with his face at her ass.>Why does the AI seem to completely ignore the prompts? The last 2 images are sort of close to what I wanted but the two obvious differences are:>Reversed genders (man is supposed to be kneeling while woman stands)>Facing the wrong way. I've tried all different ways of phrasing it in the prompt but the idea is that she's turned away from him and he has his face at her ass.>I've also tried ballbusting scenes, or scenes where a man is on the ground in the fetal position while a woman laughs at him. They don't even get close. How can I get it to pose characters how I want?>Also on civit is there a way to search for low aspect ratio models? It won't let me change the aspect ratio and most of the models I'm using take 5 buzz each time which runs out fast when I'm experimenting and learning. I found 1 model with a lower aspect ratio that only uses 2 buzz but I don't like the style and it has really, really bad problems with morphed bodies/limbs.>I want to be able to learn and experiment with prompts without burning 5 buzz every time
>>9001506wasn't supposed to be greentexted... forgot the pic the first time so I c/p'd it in a new comment and 4chan did that thing where they automatically greentext it
>>9001506Two figures on the same image are better done with regional prompting, sadly not available in civitai.The best is to go for a pose LoRA that suits your interests, and being thoroughly descriptive on the action you want.Maybe look for a picture with the pose and use its seed.The Buzz cost is determined by the model (Flux is 33 Buzz for a no frills image, found out that by mistake) and the amount of LoRAs being used. And this doesn´t count extras like face fix or hi-res fix.The cheapest route would be using a SD1.5 checkpoint, that usually has a 3/image cost, IIRC. But thing are going to be uphill.
>>9001517>sadly not available in civitaido you have a recommendation of another free online AI? Doesn't have to be completely uncensored since I don't plan on doing actual nudity, but some of the free ones have reallly aggressive censors so even something like "woman kicking man in groin" or "man kissing woman's ass" get flagged.>The cheapest route would be using a SD1.5 checkpoint, that usually has a 3/image cost, IIRC. But thing are going to be uphillthanks. I don't care too much about quality at this point because I mostly just want to get practice writing better prompts.
>>9001518Civitai is one of the best places, despite the myriad of defects and viciously stupid fuck ups.Remember: Each base model (ie: SD1.5, SDXL/Pony, Illustrious, Flux, Chroma, et al) has its own prompting style.And SD 1.5 is less refined/ easy to work with than the ones that come after it.
>>9001521>Each base model (ie: SD1.5, SDXL/Pony, Illustrious, Flux, Chroma, et al) has its own prompting styleoh, I thought they were just like different art styles. So I should only practice refining prompts with the base model I like?
>>9001525I would recommend so.
>>9001166Requesting gymnasts practiccing with the ball, the ribbon,etcetera.>>9000586CATBOX!Please for thelove of all things good, CATBOX!Also requesting a buff redhead about 6foot tall working out in a gym.
>>8998184>Sorry if its a repost, but this one is a slight touchup with Adetailer
how the hell do I get civitai to understand that I don't want nudity? I'm trying to do softcore stuff.Pic-rel, prompt includes:>fully clothed, non-nude, shorts, t-shirt, soccer uniform,negative includes:>sex, porn, breasts, topless, cum, penis, dick, nudity, naked,pussy,vaginastill I get pic-rel.
>>9002685Rating_safe on the prompt, and NSFW in the negative might help. Better give them some weight.
>>9002696>give them some weightI've heard this before, how do I give weight?
>>9002700By using parenthesis, so you go (token:weight), weight being a number below 1 to decrease importance, ie (blonde hair:0.3); or between 1 and 2 to increase importance, ie (blonde hair:1.3)Sorry for the late answer, got distracted.
>>9002730thanks. I also found out that if I take it off draft mode (I had only ever used draft mode) I can increase the CFG scale to be more precise. It helps some but even with the weights and CFG=10, I still get around 1/3 of the pics with nudity but at least most of them are non-nude now.
>>9002834Really weird, start the prompt with SFW and rating_safe at some weight to reinforce the concept. Prompt structure is important, the tokens are taken into account in placement order, getting less preponderance until the 75th, that goes back to top relevance.Also, I´d recommend upping the number of steps, and maybe moving up from SD 1.5 checkpoints.
>>9001600I don't have the whole process anymore but it's img2img of an anime pic, that ended up looking like this https://litter.catbox.moe/tjhncpb1vsbf3l50.pngAnd then a 0.15 denoise img2img with Sambase to change the skin texture a bit, and then Adetailer for the face with some other checkpoint.
>>9002685Civitai is the platform hosting models, you have to mention the name of model itself. Many photorealistic ones are heavily overfit on nsfw.
>>9002862thanks, I'll try some of that>>9002979>Many photorealistic ones are heavily overfit on nsfwinteresting. Yeah, I thought I noticed that CyberRealistic Pony (which was my favorite one) does NSFW more than some other models. Do you have any recommendations for ones that don't? Pony Realism seems the best I've found but they still throw nudity in there too much.And does the "Sampler" matter at all for this? I haven't tried changing that yet.
1/3
2/3
3/3
>>9002978I find it way easier to generate a realism-adjacent anime pic for img2img than to try and wrangle shitty realism checkpoints. They don't listen to danbooru tags very well, and who knows what sort of captions were used for the realism training.
1/4
2/4
3/4
4/4
This is an update image from https://desuarchive.org/aco/thread/8944432/#q8944516I haven't quite figured out the regional prompter yet, but had a new idea with Adetailer for faces. I experienced typically the 1boy and 1girl sharing the same facial emotions, but individual Adetailer with inpainting now helps. So I enhanced the girl's face, but accidentally I think gave the 1boy the Joker grin (prompted evil grin) and this sleazebag is enjoying himself disturbingly....Anyway helping to bump the thread as it is page 9.
By the way thanks to the anon who suggested Adetailer and inpainting tools. Sorry it was 1.5 months late for this. https://desuarchive.org/aco/thread/8944432/#q8944526By the way, before figuring it out, here were some noob mistakes:- enabling skip img2img- enabling restoring faces after AdetailerNot sure why results were messed up when they were turned on, but my gens improved when these were disabled. What do these do?- 'skip img2img' caused either an obvious box or a highlighted area to have seams >>8994463, like a change appears too obvious, but then increased mask padding.- 'restore faces after Adetailer' never gave me the intended eye color, but after disabling it I no longer had issues.
So far I've only been slopping it up on the civtai online creator but i don't like being buzz cucked and features like inpainting not being a thing really, so I saw you can download the models? What does "pruned" mean in this context, if I download will it just work like online or do I have to train that shit or whatever?
>>9004951retard moment, forgot to attach the pic
>>9004951Hi. I'm kind of maybe Forge beginner and off-on using it for a couple of months but I've posted some gens here. I used Civitai before for a short while but got fed up with the buzz credits as your tries are limited. So Forge is a local genner and you generate from your PC itself. From there you can gen with the model/checkpoint that you have. It is the local genner that you could add on, like Regional Prompter, Adetailer, input your models, etc. I'm not too sure what is Pruned either, but it can be downloaded to be used for your local genner, i.e. Comfy or Forge. Preferably Forge as the UI is more straight forward, and you could experiment how to prevent NSFW or nudity in your prompts. Perhaps you could prompt specific clothes onto the girls.
>>9004951Pruned means superfluous images on the dataset were removed, making it smaller.
>>9003017I recommend Big Love, Big Lust etc. Although geared toward nsfw, they are highly realistic (with less plasticity) and if you craft your positive prompt carefully (negative prompt has less value with these models), you can keep the clothes on :)
>>9004951>>9004953yeah you download the models and gen on your comp offline. be warned that models you can run at reasonable speed / quality depend on your GPU mostly (rn Nvidia cards are heavily favored)
>>9004933Cheers.I can't tell you what those settings do as there's no obvious equivalent in comfyUI.
>>9005102That's ok, anon. Yup it's Forge. I chose the Adetailer by Bing-su from github, and somehow those extra settings were there. But at least I'm glad I got the hang of it regardless :D
is there any step by step guide how to generate those except those trash links on the top of the thread?
>>90051401-Get a PC with 16+Gb RAM, a GPU with the most amount of VRAM you can (no less than 8GB these days, better 12+)2-Download and install a WebUI, one-click installers are available3-Download checkpoints and LoRAs to your taste.4- ???5-Profit.
>>9005144step 4 is to grab prompts/settings from a Civitai image, or metadata from a catbox itt
>>9005140get pc with as much RAM as possible, and Nvidia vid card with most VRAM as possible (probably be your highest cost), and a large M2 disk.While VRAM is king, SysRAM does help in a myriad of ways and is cheap (unless you have DDR5 of course)Then use stability matrix as your hub as it'll take care of many things. And if you begin, start with Fooocus, it'll kickstart you
>>9005144I'm on a 5080 + 128GB RAM (work related requirement) 3. How do I know which checkpoint I use?
>>9005312On civitai, filter by model type: checkpoint, time period: month or year, order by highest rated or most downloaded and browse for something that catches your eye.
>Marin once again chillin' in the beach on a cool night and wishing Stable Diffusion bros a happy weekend :P
>>9004962>>9004976>>9004993Huge thanks my dudes, playing around with local generation now, still getting the hang of it.
All these braindead retarded models don't know what "looking down" means, I've tried a dozen now and she always stares directly at the camera.
>>9006046did you try "looking at viewer" in the negative as well as "looking away" in the positive?
>>9006160What does that even mean?
>>9006359"the viewer" as in the person looking at the image
>>9005849I do not know what it is exactly but her body somehow looks cartoony.
>>9006361it is the lack of glossiness, friendolook at her knees, they look more realistic right? they have decent specular while the rest of her body is sorely lacking it
>>9006362Got it. So because she is not glossy she looks more 2D. I understand it now.
>>9006364basically the glossiness is how light reacts to the human skin which has natural oil on it produced by it but there are also other considerations like subsurface scattering and translucency though those are more subtle
>>9006359Common tags in the dataset, man. Hundreds of thousands if not millions of images where someone is looking at the camera have the tag "looking at viewer". Don't talk to the model in plain English, think about it as a series of keywords.
>>9006359I'm not sure about Pony checkpoints, but I use these in Illustrious or NoobAi checkpoints. "Looking at viewer" are tags from Booru sites, i.e. Danbooru, or if you installed Auto-Complete it would show up as a prompt suggestion. Perhaps you could increase the weight of "looking down"? And using Booru tags could help with prompts, and see how it goes.
>>9006361What is the expectation here now, and what is wrong this time?
>>9006160Didn't try looking away as I thought that usually meant sideways rather than down.
>>9006364>>9006365So what is needed to be done for this image then? Anyway strict photorealism wasn't really the main aim, as long they were not blatantly csdg. I never factored into account about the lighting. My typical approach has been to prompt a character, specifying her clothes, and to keep in theme of this character the background was a wrestling ring. So I'm unsure if it's the arena lighting there that affected her appearance. Also I have been used to, the settings CFG 3 steps 14 sampler Euler A scheduler Beta. I was seeing examples first from the checkpoint's Civitai page, then adjusted those settings for initial gens before settling mostly on them. The face has already been redone with Adetailer.
>>9006362I think it's more the huge eyes
Is there any advantage to learning comfy ui over forge?
>>9006450Not sure it's worth it, but yeah. One is earlier access to new models. Flux, quantized Chroma, Wan video, etc they all get comfy support weeks before forge. The other is niche advanced workflows with automation.
>>9006450for me the graphs click way more than the webuis at understanding the genning process
Wow, hands are tough to wrangle
>>9006432Sorry, my intention was not to complain that the image does not belong here but to point out what was lacking in respect to realism that anon noticed about it but did not understand for himself. I do not mind that image being here.
>>9006432you're using a model with too much basic 3d and toon porn knowledge and it can't shade nor light anatomy in a correct way. (i don't mean "realistically", i mean "incorrect")megawatts of electricity are being wasted on ugly AI porn videos by people who don't know any better.the solution is somewhere between "use a better model" and "learn some human anatomy like an /ic/ beginner."
>>8999821Dumbest most unenforceable law, only going to be enforced in prominent edge-cases.
>>9006833>buddy i ain't reading thatIt's not meant for you then if it's TLDR. The settings and other stuff are stated so the shill for this might know, maybe some tweaking he could advice. I'm not that experienced with it but I really gave this model a shot.Anyway there were hundreds of other images that could looked through here rather than just this one to give a feedback too then. I have given this model a try already as this is the same one another anon pointed out about plastic skin and other stuff stated >>8987606Fine. This has been usually the model that I've been told needs correction, learning Adetailer only to get this.
Gonna try asking this here since you guys might know.Does anyone have a good checkpoint or lora for local genning that can make realistic dolls/sex dolls? Think pic related.I don't want PVC figures cause the hair is plastic and it's not lifesize, I don't want realistic humans cause I'm trying to generate anime/vidya character and they lose the appeal if they're too human/just look like cheap cosplayI like the doll look cause it's the best representation of 2d into 3d for me.
>>9007418feel like that's already what people are getting when they prompt anime characters on these models, face proportions partway between her canon ones and what a real human looks like >>9005355only most people are trying to avoid that, not do it on purpose
>>9007430Actually regarding that >>9005355, I'm still fairly a beginner but would love genning from time to time. The checkpoint directly had the dataset I'm guessing, so a direct character prompt without a lora did that. Anime in fact somewhat accurate, but western characters sometimes gives questionable results for me. Just maybe my approach but I also include specific clothes, as well as background and threw in a beachball. Also coming to asdg, I kind of thought like, just try prompting live action versions of fictional characters for fun. I additionally wish to understand about celebs like how those guys do it in MMAI, but for now I would stick to fictional characters as it is easier for me.
>>9007458It was not meant as criticism, I just needed an example.
Finally got tired of switching between 3+ checkpoints for every gen and I made my own merge for booba cosplay. Think it's worth sharing on Civitai?
