Ojisan EditionPrevious Thread: >>>>8795001>LOCAL UIreForge: https://github.com/Panchovix/stable-diffusion-webui-reForge | Stable commit: 6964cebComfy: https://github.com/comfyanonymous/ComfyUI | https://comfyanonymous.github.io/ComfyUI_examples>NOOBAI-XLhttps://civitai.com/models/833294/noobai-xl-nai-xlhttps://huggingface.co/Laxhar>NOOBAI SHITMIXES102d custom: https://civitai.com/models/1201815?modelVersionId=1491533291h: https://civitai.com/models/1301670/291h>RESOURCES4chanX Catbox/NAI prompt userscript: https://rentry.org/hdgcbEasy booru tag export userscript: https://github.com/Takenoko3333/Danbooru-Tags-Sort-ExporterTags: https://danbooru.donmai.us/wiki_pages/tag_groups | https://danbooru.donmai.us/related_tag | https://tagexplorer.github.io/#/Inpaint: https://files.catbox.moe/fbzsxb.jpg | https://huggingface.co/Wenaka/NoobAI_XL_Inpainting_ControlNet_FullUpscalers: https://openmodeldb.infoBooru: https://aibooru.onlineControlNet: https://rentry.org/dummycontrolnet (OLD) | https://civitai.com/models/136070 (look at the green links in the description)>TRAININGGuide: https://rentry.org/yahbgtr (WIP)Anon's scripts: https://mega.nz/folder/VxYFhAYb#FQZn8iz_SxWV3x1BBaJGbwTrainers: https://github.com/derrian-distro/LoRA_Easy_Training_Scripts | https://github.com/bmaltais/kohya_ss | https://github.com/Nerogar/OneTrainerOP Template/Logo: https://rentry.org/hggop/edit | https://files.catbox.moe/om5a99.png
Arouselights anyone?
>>8805373Please do.
>>8805373don't make me ghost-post in the old thread
>>8805347you drew this??
any free websites similar or better than betterwaifu? my pc can't handle generating locally. pic related
Any lorabaking GODS here? I'll have one wmmzkl artist lora, thanks.
>>8805683looks surprisingly non-hag
>>8805683Make the dataset
>>8805699There's like 12 pics total on Danbooru/Gelbooru, counting out the versions of the same pic, would that be enough for a Noob lora? Or a Zit Turbo lora.
>>8805721if you don't mind it being overfitI think he meant collecting images from other sources, grabbing the danbooru pics takes like a minute
>>8805721not really and I don't think sdxl will learn this style well anyway
>>8805537did you prompt pink dicks or something?
>>8805734It's over, GOODBYE
>>8805734nope thats definitely a shortcoming of that site. thats why im asking.
>>8805764well, I went local due to limitations of pixai and their hate of anything remotely lewd.But there should be better options out there, even ignoring nai.
>>8805734Hews hack coming in clutch
how do you add strength to something with brackets, unlike (big boobs:2) you can't do (2b \(nier\):2))
>>8805855Like this: (shrug \(clothing\):0.8)If you are using forge you can just highlight the word and press ctrl + up or down key
>>8805857I don't think that works, if you do (shrug \(clothing\):0) you get a shrug, the brackets might be breaking it
Come collect your schizo he's shitting in /edg/ >>8805858I get what your saying kinda but, what is your specific problem cause it sounds like a model thing. Trying it out myself and it seems largely working
>>8805858does :0 even work with a standard one word tag?
>>8805858>I don't think>mightWhere do you people come from honestly, you could just do a single test>https://files.catbox.moe/q6ue6d.png
>>8805858Try that with any other tag. Weights are non-linear. 0.0 is more than not prompting the tag at all. The difference between 1.0 and 0.9 is greater than the difference between 0.9 and 0.5. etc.
can you negative prompt (thing:-1.0)?
>>8805869You can use negative weights in the positive prompt with NegPip extension, which is more powerful than usual negatives but also worsens any side-effects.
>>8805876interesting, thanks anon
wao...
wawaweewa...
>check the patreon>it's realI appreciate the hustle of your larp, chat.
>>8805921>>>/e/
>>8805858>>8805867>0.0 is more than not prompting the tag at allcomfy++ weighting doesn't have this problem, not my problem you are using obsolete a1111 garbage :3
comfy should be dragged and shot
>>8806186dragged where :3
>>8806185normal comfy does it too, no idea where the ++ is but I doubt it's worth re-learning everythingnever said it's a problem either, that's just how weights work
>>8806192Normal comfy is even worse than A1111.https://github.com/asagi4/comfyui-prompt-control
>Normal comfy is [delusional schizobabble]
There are different comfys now?
Heh. For a moment there I thought I was on the /hdg/ tab.
>>8806207I guess they are talking about the difference between vanilla comfy and modded comfy.
>>8806197prompt control uses the same weight system
>>8806219The point is you can specify>Style(comfy++)in the first line and do many other things which are essential, like lora and prompt scheduling, in a single prompt box. It's confusing to me that most people still use comfy without prompt control.
It's confusing to me that some people still use comfy.
>>8806220Prompt scheduling is nice. I prefer to do lora scheduling with builtin nodes, don't have to type out the filename like an A11 fag.
>>8806221what else would we use?
>>8806321eww
>>8806323Should have I made the bulges bigger?
>>8806321me on the left
Any good site to train loras online? I have datasets from old loras I made on collab (like 1.5 years ago) but now I don't have a good pc at the moment so I depend on horde sites for gens. The problem is that my loras worked best on based64 and the closest on the horde is hentai diffusion, which isn't a perfect fit. So I need a way to update my loras for illu or sdxl
Why no audio streams allowed?
>>8806458nice nips
>>8806457>>8806458>>8806459>>8806635woaww
>>8806722its /e/
>>8806458weighing scale box?
>>8806454just post a tagged and curated etc dataset and some anon like me will bake it for you
>>8806728its /d/
what's the current model meta
>>8806875nothing. waiting patiently for anime finetunes of newer models.
>>8806731Sure, why the hell not.I have the rar for a style lora for Kakuji Fujita:>https://files.catbox.moe/3woak7.rarAnd one for Nona:>https://files.catbox.moe/9965r5.rar
>>8807112we may get some problems because of the prevalence of white bg images for both of those but who knows i'll find out
>>8806875Any noob model with a side of kanikananika lora.
https://files.catbox.moe/ags10u.png
>>8807231>kanikananika loragib
>>8807231>kanikananika lorawhere
yo hgg got a questionDoes anyone know a lora for something like this?https://x.com/poke0664/status/2006927573595587018Keeps pumping them out
>>8807407>>8807439Search the archive :^)
>>8807457ah, it's another le funny hdg meme
>>8807459Yes but anon swears it's there. I've personally never been able to find it going back 6 months so guess he got me.
Everyone has kanikananika lora at this point, stop trolling.
>>8807504Im going to train one
>>8807510Will it also be in the archive?
>>8807456I'd be very surprised if SDXL could do something even remotely like this, and yet the right side is nonsensical enough to be AI. In theory it shouldn't require character knowledge if you provide the 1girl standing, so maybe it's one of the large SFW services. Or a more involved process with manual edits.
>>8807461There's one version but I don't think he ever uploaded the current one
>>8807459nta, I also thought it has to be in the archive or the lora rentry for sure. But first mention in the archive is someone memeing as if it the lora was already well-known >>8698758 Coupled with the style being shit and misspelling the name https://danbooru.donmai.us/artists/490309 leads me to believe it's a long-running meme.
>>8807625It's in the archive, but when bro uploaded it, he only linked to pixiv of the artist and didn't actually give mention the name in the post. Hope this helps you find it!
>>8807637Thanks? I was just curious what the fuss is about, but it might help the other guy.
I'm happy that my refusal to reupload something already on catbox coupled with anons being unable to use the archive is giving the thread some traffic.
>>8807660Upload the new version nigga
>>8807670I'm away from my main PC right now.
wtf is kanikananika even? can't find anything on booru or pixiv. some niche pedo artist`?
>>8807678https://danbooru.donmai.us/posts?tags=kani_ka_nanika&z=1
Can't use the archive, can't use search engines. The level of competence is through the roof here.
>>8807660tried three different spellings before doing kani* and even that didn't work at some point you gotta accept the lora is not worth the effort spent searching for itwho knew it'd be something stupid like the pixiv url
>>8807670https://files.catbox.moe/11zv1k.safetensorsHere, now please stop bringing my shitty loras up for your shitposts, thanks.
>>8807822I've never done that, but thanks
Do you anons recommend a specific model to uncensor mosaics or do I just roll with NoobAI inpaint?
>>8807523It could be that ya, the speed which they post and the obvious ai giveaways oh well I'll keep looking
what did lil bro softly inpaint
>>8807917a little bit of everything, but mostly the glans job/foreskin insertion. i just like soft inpainting because it's better at hiding inpainting seams at high denoise values.
>>8807913There is something off about her arm/wrist angle. Nice gen tho.
https://civitai.com/user/MisterTwister69/modelsholy kino
>>8807952What is this supposed to be?
>>8807970Testing my batch processing tool that I am currently developing
>>8807991Batch processing what
>>8807993well the idea is that you can quickly do your entire favorite cast as a merged picture in the pose/clothing/angle/setting/style (and any combination of that) you like in one gohttps://github.com/MyBodyIsLava/HaremBatch/tree/main
>>8807952What was the controlnet for that one? Canny?
>>8807999dw_openpose_full give me the best results, canny force the wrong lines too much to my testing
>>8807822>stop bringing my shitty loras up for your shitpostsBased. What's the difference with this one and the "yn4wxk" one?
>>8807456It's AI yeah but It's Gemini, I've seen people post prompts for this before. Not sure how they manage to get it to output lingerie though. SDXL can't do this even with a lora but you might get similar results from qwen edit or something like that.
>>8808493>output lingerie thoughprobably because it's just the clothes, not on someone
>>8807456There was another user posting the same thing under genshin characters but it screamed slop to me because it kept labeling the fabric "silk".
>wearing garter belt and stockings under long pantsslop? who wears such thing under pants?
>>8808505not even a garter belt, the straps go directly to panties
>>8805732you'd be surprised but yeah you're never gonna get that pixelation without some post processing
>>8807456also interested in the mix a similar reference sheet style genner is using:https://www.pixiv.net/users/117543073
>>8808576that just looks like the base style of any generic illust mix
>>8805340>Year of AI videos>People still generate AI .pngsI heard you can gen 5 sec video with a single 5070 nowadays
>>8808582I did a bit on my 3090 but the 4-5 minutes wasn't worth it for the results.What loras are you using now?
>>8808584isnt 4 - 5 minutes for 5 sec videos already enough ? That webm i just the one i found on Civit
>>8808586desu I probably just bailed because I knew it would lead to a bad lifestyle, I actually quit AI gen for like half a year after trying out the vid gen shitKinda felt like we were reaching a point of no return with the technology at the time. Actually thanks for reminding me fuck this shit why am I back here
>>8808589You can thank AI jews for that. They limit Local models on videogen on purpose
>>8808493>It's AI yeah but It's Geminigemini 3 can do this indeed and the inconsistencies that are omnipresent will always betray it's AIthis technology truly is a dead end, no matter how much better it gets at following prompts (by training on lots of artificial curated data to force behavior to prompt) it will always, always get the details wrong, it's inherent to the stupidity of LLMs and diffusion models
>>8808582the models suck
>>8807939cute
>>8807112here are some v1s (without a trigger tag, like you provided). the nona one just doesn't work at all unless you prompt something related to the dataset like toned female or a dougi or something. or a man. gonna bake again tonight and add a trigger for themnonahttps://files.catbox.moe/pcc5tj.safetensorskakuji fujitahttps://files.catbox.moe/ly29ja.safetensors
>>8808641Another hundred trillion dollars to OpenAI, Nvidia and Oracle will fix that!
>>8808673>nooo you can't brute force it, this is the ceilingbetter models drops>nooo you can't brute force it, this is the ceiling.better models drops>nooo you can't brute force it, this is the ceiling.better models drops>nooo you can't brute force it, this is the ceiling.better models drops
t. sam altman
>>8808741Are those better models in the room with us right now?
>>8808745no progress has been made from sd1 to nanobanana
>>8808747Ah yes the progress from SD1 was not brute forcing it, nano is still running on 1b parameters.
>>8808641>no matter how much better it gets at following prompts (by training on lots of artificial curated data to force behavior to prompt) it will always, always get the details wrongThere's also the problem of HUMANS not bothering to personalize shit per image.I don't remember where to find the prompt, but I know it specifies outer, mid, and inner layers. Notice it's missing undies in my simple quick test. Possibly mentions stockings as an example so the model picks it up; admittedly a "smarter model" might have a better idea of what an "example" is, but it's harder to work around the pink elephant.
but yeah, lots of inconsistencies if you don't spell out what you want to see, with current modelsparticularly the labeling in this case
>>8808582i boughtened a 5060 like an idiot...
>>8808757Still time to buy a 5090 before they hike the price and consumer gpus will become unaffordable
>>8808760nta but its out of my price range.Also RTX 5070ti super 24gb being cancelled make me seethe
*laughs in 4070ti super*
>>8808771at this point just wait it out anon, save up and get a better card when the 6xxx is out in 1 year
>>8808798>as low as $2000 dollars for 6070Nope
>>8808799no, doomer anon, doom harder, it's gonna be $1M for a 6050
>>8808803Realistically NVIDIA would just release 6090 at $7000 and thats it. No 6080, no 6070. These fuckers are 100% AI company who only sell their GPU to the highest bidder
>>8808652Haven't tested the nona one yet but the other one works like a charm, thank you anon!
>>8808921nice style. which lora?
>>8808928The Kakuji Fujita lora that anon made plus height ratio slider at 0,3. Prompt had "sketch:0.2,flat color" at the end of it.
>>8805464Oh wow, someone reposted this old thing.
>>8808577it's really just that easy to get 2000 a month with this shit huh. that guy's barely been at it for half a year
>>8809001kek, am I wasting my time trying to master everything before asking for patreon bucks
>>8809032If your goal is to eventually ask for patreon bucks then yes, you should start building a following right away. At least on pixiv/twitter.
>>8808998Fix the Japanese in the bubbles, you lazy wanker.
>>8808994The art is so good, I didn't recognize it as AI at first.Lurv me Hex, simple as.
normally you wouldn't just go and brag about being blind online
>>8809346You're gay.
>>8809348very nice style. box pls?
>>8809351You can't seriously consider this image anything but utter slop. being elitist is one thing, but once you're fapping to amputee porn in surreal spaces something is wrong in your head
>>8809352https://files.catbox.moe/z1for9.png
>>8809355>riz, emmmerald, no lineartneat. i'm surprised riz doesn't have as much of an influence on the lines, but i guess no lineart helps prevent that. thanks for the box
>>8809357No prob! You're right, riz's style is strong so no_lineart just lets more of emmmerald's style come through (even though it doesn't really remove the lineart much lol)
>>8809363could try messing with weights or prompt scheduling instead
>>8809363i haven't tried prompting riz on illustrious or noob but given how much data is on him doesn't surprise me. although to be fair it seems like no lineart is working there (assuming it replicates his style faithfully) cause the lines in his art seem to be way thicker than those gens. i've been meaning to train a style lora on him but i've just been too lazy to rip/sort/clean all his shit
>>8809365Working as it it reduced the lines slightly, it doesn't remove them the way it's supposed to.
>>8809369if you're trying to get rid of lines but keep the other aspects of his style definitely look into prompt scheduling as >>8809364 said. it'll give you way more control over how much of an influence riz's style has than just messing with weightshttps://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing
oh wise ojisans, i inquire on your knowledge and wisdomwhy was illustrious v2.0 doa
>>8809379eps, less training than noob 0.5why bother?
>>8809380thank you ojisan for your knowledge and wisdomplease accept my humble offering https://files.catbox.moe/s6e9q7.jpg
>>8809379training sdxl at 1536x1536 is really stupid.really, really fucking stupid.and it fucks everything up no matter what resolution you try and output to.
>>8809385this
whats the best controlnet tile model for noobvpred? cant get rid of artifacts without toning down lora strengths and upping the cn strength to 0.75 with the xinsir union model.
>>8809547https://huggingface.co/Eugeoter/noob-sdxl-controlnet-tile/blob/main/diffusion_pytorch_model.fp16.safetensors
>>8809385some retard kept saying how amazing it is for inpaintingshit's awfulpeople at g give such horrid advices that im not even sure whether its all trolling or not
>>8809547Are you running the union model as tile? It's unique in that it has a bunch of modes you can select.Anyway this one "should" be better for noob https://civitai.com/models/929685?modelVersionId=1239319 Afaik there isn't one for v-pred specifically, but it's still the closest you can get.
>>8809548its the one you linked, not the 5 gig one named "noob-sdxl-controlnet-tile.safetensors" right? >>8809552i tried that too but its not any better unfortunately, at least for my workflow
>>8809553they're both the same, the one I linked the is fp16 one while the one on civitai is fp32
As someone just getting back into this - what samplers do you guys use (vpred/shitmixes)? Just been rocking Euler A automatic bc I'm apparently a caveman.
>>8809712>>8805340Eular A is still fine
>>8807442Sexo. How do you get size difference sex like that?
>>8809766onee-shota tag
>>8809385>>8809418>>8809550Uh.. noob-ft-1536x-extract lora sisters? Should we be concerned?
>>8809547I use the eps one that anon posted >>8809552 when I want the image to be as close as possible to the original.When I don't mind it being a bit more creative with the end result I use this: https://huggingface.co/2vXpSwA7/iroiro-lora/blob/main/test_controlnet2/CN-anytest4_illustrious2_A.safetensors
>>8809814base sdxl gens regardless of the resolution used always look kind of bad, even if you gen on a higher res you still need to inpaint some parts of itgen I made when I was testing illust 2.0 back then>https://files.catbox.moe/apqeh8.png
>>8809712I'm partial to DPM++ 2S a Comfy or DPM+SDE Comfy lately. 20 - 28 steps.
>>880982056 steps on SDXL :skull:
>>8809827chat, is this playful razzin?
>>8809369nice char. i don't play gacha shit but something about her hair color with the hair ornament she has is very hot
>>8809712>>>/e/3017440>>>/e/3008796
>>8808652>>8808921the v2s with a trigger tag worked better. they are good enough so uploading them to my mega here https://mega.nz/folder/zzZSFZAa#YZDbED0Af8SYeMb5FscJPQ
>>8809712dpm++3m sde with 36-50 steps (sde benefit from high steps), sgm uniform scheduler
how to gen downturned elf ears?
>>8810006bend them in photoshop then inpaint
>>8810047>how do I gen [thing]>uhmmmm pick up a pencil and draw it yourself you chudIf someone is asking how to gen something it's because they don't want to be an artfag you wasted chance for a postnatal abortion.
>I want to be able to 1-click everything without a second thoughtgo back subhuman
>>8810068Yes, I actually want my AI to act like AI and not require me to fiddle with art shit in a different paid program.
>paid program out of nowhere
>photoshop>paidthe fuck is this retard on
>using photoshop without generative fillwhy bother
>>8810006you can't. the dataset is overwhelmingly towards the standard kind. just make a lota.
>>8810077Yes, photoshopping is is paid you subhuman.Also, why are you afraid of replying? Did your dad smack your mom too hard when she talked back?
>>8810082I think edg might be more your speed.
>>8810088Yeah maybe people there will actually answer the question instead of acting like passive aggressive little bitches from /ic/.
>>8810080Or bend them like I suggested right away. You're anyways just gonna make three gens with it before losing interest in bent ears. Gimp is free, Krita is free, but better get hung up on the last important part of the message.
>>8810107>Gimp is free, Krita is freeThey are free because noone would pay to use that trash
https://files.catbox.moe/4cniqg.png
>>8810112this desu
Can SD be as good as grok for image editing? I can put bikini with control net or img2img but doing it with a text is so much nicer.
>>8810137In theory yes, if someone trained an inpainting model for illustrious. You can try using inpaint controlnet and just inpaint at 1.0 denoise.
>>8810137Unfortunately, SD is several generations behind. Maybe if you had the time for some inpainting and praying to the RNGods you would get something similar, but nothing on the same level of ease as Grok.
>>8810122
>>8810141Hot.
>>8810141Think it's more about param size than model age, grok wouldn't fit on a server rack full of GPUs.Anyway for adding/removing a bikini all you need is NAI-tier inpaint. This >>8810140 isn't quite there even, but it can work. Boxrel is from >>3041913If you mean not having to draw an inpaint mask either, Qwen Image Edit claims to be able to do it. I never figured out how cause every roll takes minutes.
>>8810141Nice,very grey and shiny and inpaint-free.
>>8810150>>>/e/3041913I will never get used to the new crosslinks
>>8810141Whats the most modern local image generation? I just checked civitai and looks like nothing changed in 1 year.
>>8810155newBienetaLuminaz-image (or zit) and qwenChromathe first two being anime capable but horribly undertrainedqwen works as an img2img/controlnet base for /e/ stuff, replaced Flux in this usecasez-image is almost Flux-tier but very lightweight, so same as above if you're short on vramchroma is now the top photorealism model
>>8810155because nothing of interest happened last year
rip
>>8806971box pls
>>8810207box? also fair warning that jannies might not be too kind to tentacles
>>8810267nta but it'll be fine. Yes it's "supposedly" /d/, but did anyone ever get jannied for it? I don't remember one case in those 2.5 year.
>>8810171Which artist styles are you using in the depiction of this tradwife?
>>8810272Based living dangerously anon.
>>8810301https://files.catbox.moe/f3u8gq.pngIts there in the base gen, faces got inpainted.
>>8810306>furry and artist which I thought was loli coded due to shutenanonI'll be damned. Here I thought it was khyle. Thanks, anon.
>>8810317khyle never draws lipswamu was memed really hard in the late pony days, one of the best hashed artists in there. I recognize his faces right away even in a mix
>>8810317It's more like half-chibi-coded. I don't think shuten uses wamudraws at all.
>>8810319>wamu, healthy, possumWas probably the original iteration. Not sure thereafter.
>>8809991Neither one seems to work at all.
>>8810223https://files.catbox.moe/yf7keh.png
>>8808741"better model" is a relative thingnone of the newest models (I mean both proprietary AI and open) have any sort of variation based on seed, they are so overfitted that a descriptive prompt will get you almost the same image all the time no matter how many times you regen. This is true of nano banana pro and it is true of models like qwen image and z-image. Their better instruction following came at the cost of the model's ability to generate variations and show some form of creation.The same phenomenon is what amplified LLM slop too (You're absolutely right! It's not just right, it's absolutely right—here's my list of 1, 2 and 3 elements—a testamentt to the progress of LLMs!) the instruction tuning is balancing the training towards more and more synthetic data and what we have now is a human centipede.
>>8810432Nice
>>8810448thanks
>>8810322assuming you missed the trigger tag then
>>8810459What gave you the impression that a "nooo" poster is making a legitimate argument that warrants discussion.
>>8810432I'm not nearly as good as you but it's such a cute concept I'm going to make more until I'm happy with it.
>>8810322they are not burnt to a crisp like most artist loras, use the trigger tag and prompt for something in the dataset (fighting game related stuff)
>>8810549bitch got three arms
>>8810551i'm so sleepy...
>>8810540what caused this
bros someone gave me a amd 6600 because they are getting a 9070, it's possible to generate or should i just give up? no one making the girls i like doing the stuff i like
>>8810559Does it even count as giving up when you haven't even tried?
>>8810559put "radeon 6600 ROCm linux" in the google
>>8810582i have asked once before and all i got was lol'ed and told to get a real GPU>>8810583ok i try it, thanks
>>8810559Around 1080p upscaled outputs are fine on that VRAM at least so I wouldn't rule it out. I'm running ComfyUI on 7900XT so AMD isn't an issue either, though I have no experience on that specific card.
>>8810596i found sources but they conflic each other, omnesays it works in linux but not supported in windows then other one says it's the other way around, so far i got no idea which one is good
>>8810592i was able to install rocm on rx 570 back in the day, that doesn't support it, so it's almost certainly doabledon't know if the linux is still the only way to go on amd, but it's probably the easiest to setupas for ui, comfyui will work, but it's pain to use, no idea how it looks now with forge
>>8810598I'm running it on Linux. Used to have 3060TI on Windows with Forge, but couldn't get it to run properly when I switched hardware and OS to Linux. Forge has some ROCM version thing, but at least my card has the issue where you have to use tiled VAE or the generation takes like 8 minutes. With these resolutions these take about 28 seconds with tiled VAE.>>8810599Comfy is a hassle, but at least its frontloaded. Once you have a functional node setup for basic gen or inpainting etc. you can just prompt.
got my gpu up and running, still learning shit but i run the regular checkpoints.taking requests
>>8810624Sorry to say this, but we all have stuff up and running too. Any non-ironic request I would have already genned hundreds of times.
>>8810624requesting you lurk for a few weeks before you post
>>8810633sorry :(just wanted to be helpful
>>8810635what pussy does to a mf
>>8810634I didn't mean to be antagonistic, but as >>8810632 said, everyone here has their genning setup and in this general people tend to be rather experienced with this stuff.
>>8810637fair, i literally got it up and running today.ive been reading for a while but now that i have the actual thing functional i'm just testing shit out.i don't understand a significant bit of the technical conversations that are going on.
>>8810638you need to wait for jeet hours to do requests
>>8810648yeah nevermind forget i said anything i'm not gonna spend gpu time for jeets
>>8810650And shouldn't their hours be right now
>>8810658lmao nice. always love me some instant loss
>use WAI>looks great, has great anatomy, no artifacts in sight, isn't fried>but the default style is really difficult to shake off if I want something else>use NAI vPred v1.0>looks like shit, awful anatomy, artifacts out the ass, deep fried>but the style is really malleable and hits the bullseye every time if you ignore the fact that it's friedWhat do I use?
>>8810682>>8805340
>>8810682then use shitmix without grease like ΣΙΗ, shitposter-kun
I am sure this is someone's fetish, but I am 100% sure nipples aren't capable of generating pussy juice.
>>8810682go to sleep and wait for Z-hentai
>>8810691>aren't capable every vidgen ever
>Dozens of custom nodes>Most of them obsolete/abandoned>Latest ComfyUI update makes almost all stop working altogetherSigh.
I am sure this is someone's fetish, but I am 100% sure local models aren't capable of generating videos that aren't complete dogshit.
>>8810688>ΣΙΗNah, it still doesn't hit the mark on the style.https://files.catbox.moe/z5jvup.pngFor the record, this is the style I'm going for:https://vndb.org/v4880
>>8810711then beg somebody for lora, hard to expect native style with this dataset on booru to be completely faithful
>>8810713Sadly the only Misaki Kurehito LoRA that includes his old works (which is very different from his more recent one) is inconsistent and also only produces the intended style on vPred v1.0 (when it feels like it).
>>8810714Shouldn't it take you a whopping 2 hours at most to make the lora yourself? Grab the CG set and get cracking.
>>8810715I honestly haven't a clue how to make LoRAs, that's why I've been trying to find a way to make the style work with what I've got.
>komachi outfit/hair ornament barely doesn't workannoying, so close>>8810682use nai vpred 1.0 or illutrious 0.1 with loras
>>8810642kino
Huh. That worked much better than I expected.
i dont WANt to see vidgens
it's impossible to distinguish between honest vidgen and shitpost
I wish grok was uncensored again
Adds own gen? Honest. As shrimple as that.>>8810775Only if OpenAi actually goes through with uncensoring ChatGPT, and thus spurring competition, which I doubt.
>>8810658>>8810674>>8810721good stuff, especially for strong willed characters
>>8810782reposts are illegal
>>8810784What choice do I have? I am still waiting for my 4070 to finish with some other things I am prompting through.
>>8810708Thanks for fixing it, anon.
>>8810775I heard people mention the paid version is significantly less censored
>>8810775everything will be uncensored in the future. once every model is pretty much equal the only thing left to compete on will how censored you are and then it will be a race to the bottom. we are still too early.
Cute and Funny videohttps://litter.catbox.moe/m5g7pszrcnju33hg.webm
>>8810682Merge 291sekret with custom or funny greek. 70/30.
I wanted to make golshi be horse sized, but that's not /h/ allowed.
Chat if I gen a cute boy next to a cute girl both with matching outfits, it is gay or no
>>8810904Depends. Does it look gayer than this?
>>8810907No, it's rather cute actually
>>8810908Post it. Unless it's some shota thing, most mods will accept it.
>>8810911It is a shota with a cute and funny girl, yeah
and i almost forgot about (second) best girl>>8810783thanks
>>8810912>funny girlWomen aren't funny, anon. Now we know you're trolling.
Submit sample for analysis
Where is LTX2 for hentai
what happened? why so many images?
>>8811186/hdg/ surpassed the shitpost event horizon.
I am getting frustrated with Noob. It seems to default heavily to offscreen male, disembodied penis, and futa. It's better than Pony was for sure but god damn the training on danbooru/e621 tags has obvious hard limitations for some of the concepts I want to do.
>>8811279Yukikaze killed that poor orc...
>>8811281I never thought she had in her.
>>8811280offscreen male happens when you do not prompt features for the guyfuta happens when you do not prompt 1boy
>>8811291I wish it was that simple but 1boy is not a magic fix that is consistent and always prevents futa shit from being genned. What features do you mean that are specifically male and only apply to the male? For example, if I'm tagging for a goblin, sometimes it will just make the girl the goblin or just do a green offscreen guy.
>>8811304Can you give us an example box? could be a prompt issue, but in my experience it's also heavily influenced by style
>>8811312I will have to later. I am not at home. Negpip at least has been extremely useful for me and certain position tags and tags like full body help getting everyone in the shot.
>>8811304like >>8811312 give examples and you'll get infinitely more help. but if the girl is getting goblin features you're probably not descriping both the male and female adequately and also just add female goblin to the negative prompt (at low weight if it's impacting gen too much)
>>8810506>>8810547Maybe the ones you uploaded to mega are borked because neither one does anything at all even if I weight the trigger tags or I put every single one you provided.
>>8811331i redownloaded them and checked and you must be experiencing some sort of major skill issue since they are identicalyou have some setting that prevents loras that aren't fried to shit from working or maybe you are just retardedhttps://litter.catbox.moe/povfetdm6y1lqh08.jpghttps://litter.catbox.moe/5o5nq6f5e64puglo.jpg
>>8811352Or using a shitmerge that overpowers the style
>>8811354that falls under "retarded"
>>8811352>claims they're not fried to shit>it does, in fact, look fried to shitwhat did chat even mean by this
>>8811356if you check the artists they just look like that, it's not what i would pick
>>8811331OG requester here, I'm sorry to say you're retarded
Any reliable automated way to remove captions / speech / text from images yet? i.e qwen edit locally? Trying to prune a dataset but I'm a lazy fuck.
>>8811390qwen can do that, yeah
>>8811291Actually, 1boy helped a lot. My prompt didn't have it...so my bad. I could have sworn it didn't always work though but anyways, right now its helping.
https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLgadded maeka-(kumaekake)
>>8811571Thanks.Somehow I dislike every single style in there, but it's nice to see people still baking.
>>8811356TOASTY!
>>8810901cool to see a past gen of mine out in the wild.
>>8811571Any interest in sharing your bake settings? Ty for baking regardless.
>>8810659>>8810674>>8810721>>8810759>>8810947>>8811009Box / Artist? I really like that style
>>8811880i have the script i use to train on kohya sd scripts in the z-scripts folder; they're probably outdated but i don't care as long as i get the results i want. epochs and repeats change obviously and are on a per style basis>>8811896https://files.catbox.moe/j9wxhd.pnghttps://files.catbox.moe/grso7b.pnghttps://files.catbox.moe/6nh58i.pngthat's a lot of boxes to find but here's a few at least. these should cover the different styles i used in all of those gens
>>8811907Thanks for the boxes!
genning otoko no kos was clearly a mistake but the good thing about it is that it's now out of my head
>>8812001formerlyotokonokobox?
>>8812002no way fag, all of those gens are going into the sekrit vault along with the other degeneracy concepts that my mind decided for brief period of time to be oddly arousinghave this pokegen instead
pointy_breasts
>>8812004why does this boy's anus have a cervix?
>>8812007nice joke, clearly I didn't prompt for any type of breasts so it default to the size attached on most entries, don't even know why I am playing into this but whatevercya
>>8812007blud forgot to use anal_cross-section for his otoko_no_ko (currently trap) gen. :skull:
>>8812001>>8812004noodlefaggot lol
I have spent 2 days straight making ai porn. I have only slept for like 4 hours. Please help. I'm gonna delete everything. I need to know how to stay away. This feels like digital heroin.
>>8812173Sell your gpu and buy a PS5
>>8812173Don't worry, it'll get boring eventually.
>>8812188At least he's in no danger of getting addicted to genned image (you)s, now that the thread is a barren wasteland.
>>8812173to add I've fallen in love with my OC and am now sad as fuck cause to quit I'm gonna have to forget about her. I just spend hours genning her in different scenarios.
what are some good artists for skinny girls with big titsI'm tired of everyone having super wide hips and mega thick thighs
>>8812173>>8812226Managed 3 hours and now I can't control myself again
>>8812236I think that's a bias baked into the boob size tags more than any art style. From my tests "skinny" is the only thing that makes hips more narrow. Thighs can be reduced with "slim legs". Seems to work pretty well regardless of artist. Avoid "thin waist" or "long torso" as those make the hips look wider by comparison.As a last resort you can use this lora for the base res gen https://civitai.com/models/1671543/huge-fake-tits-illustrious (unironically)Surprisingly little style bias, negligible at around 0.3 strength and makes boobs larger without widening the waist or legs.
>sankaku_ai_artbruh
>>8812250ah true, I suppose I'll stick to negatives then, thanks for your research
>>8812250great example of the "long torso" caused by using "skinny" in reality is the same length, but looks long because of how narrow is
>>8812307I meant the "long torso" tag, hence the quotes. It can make the hips wider sometimes.
>>8812320i meant that anons were complaining about long torsos while prompting skinny, not your post directlyalso long torso tag has only 174 pics on danbooru
>>88123251.1K on e621, might add up to something together
anyone got a inpaint workflow for either krita or comfy? it always seems to fuck something up... for krita the seamless option kinda works but it'll leave a seam and mismatched color/style and non seamless is complete shit.
>>8812353Literally just.... inpaint....
>>8812357Those settings are pretty dumb, but the workflow is correct.
>>8812360Settings are absolutely fine for dmd2, unless you're talking about the crop etc in which case works on my machine.
Put penis into robot.
>>8812245eventually you hit the limits of what can be done right now and get bored
>>8812226>>8812245That's when you learn new skills and make something that you want to see. Start a comic. Too much work? Maybe. But this is about your personal satisfaction.
>>8812363Makes no sense for guide size to be lower than max size, did you swap them by accident? They're effectively both at 768 now.
>>8812408*to be higher than max size
>>8812408I set max size lower for 1 image because it being so high res compared to everything else made it look off (it was a tshirt graphic iirc)
>>8812091Tell me about awa. Why does he prompt the tag?
awa de uwu
>>8812236>>8812250Good reason to familiarize yourself with some of the furfag tags. Slim, top_heavy, and narrow_waist work wonders for that phenotype.
>>8811914np anon>>8812453lmao took me back to the announcement that noob was adding furshit to the dataset
>>8812487why he's yellow
>>8812503He's got jaundice
>>8812503That's the latin x cartoon legend Homero Simpson.
>>8812487>>8812503psyduck gens belong in >>>/trash/slop
JK Gyaru's are for pleasuring ojisans in the seediest hotes. https://files.catbox.moe/04bpve.png
because of the slop, i recently realized that i like girls who look a little retarded the most
>>8812541it's asanagi
>>8812542nah, he's spiritually /aco/
grr..
based repostGOD
dios mio.. el sexo poco imaginativo
>>8812745Is this AI?
>>8812751No, it's Rin
>>8812091more vibrator in thigh straps
best tag with under 500 hits?
>>8812789pussy_grip
where is the sparxie lora you fucks
>>8812794ass design, and i say this as a fgo player
>>8812795>sparkle blanca and streamer is bad designstfu nig
>>8812991tits too small
>>8812991cute style nonie
>>8812992they're never too small
>>8812994*always
>>8812998kek i just realized how much of a midget the dude is
I am surprised this prompt worked on first try. Either this is a common concept, or this model is pretty good.
>>8813007the hard part is keeping the detail, anatomy and composition while aiming for a style that doesn't look like utter slop
https://files.catbox.moe/00e5qo.png
>>8813016Rather harsh to rape Klee over a children's card game, no?
>>8813019Who said it was rape?
>>8813022The uncomfortable look on her face, kinda.
>>8810432Phone gen in full size please? I think it can stand on it's own.
>>8812173wtf you are me from 3 years ago
why are goat eyes so creepy?
pls sir may i have some more pixels
no
>>8813063ok maybe just a few more pixels..
>>8813097more disgusted instant loss
>>8813097>>8813134What the fuck was her problem?
>>8813168>What the fuck was her problem?her sister more than anything
>>8812226
>>8813191I keep trying to quit everyday but I can't
>>8813191>>8813199Which acoslop artist is this, chat? I feel like I should know unironically it from the hashed artist pony days but I'm drawing a blank.
>>8813199>>8813204for this style I use the artist tags:dimwitdog wamudraws gerphand also the tag flat_colors
>>8813207>wamudrawsOk. That makes sense. Knew those faces were familiar.
>>8813207based ass slap enjoyer
>>8813207Addiction is so bad I go like 8 hours without water then eventually get up and down 1 litre in a few seconds.
el heckin bbcino in hggerino? argh i'm going insane!!!!
>>8813232you love nigga penis huh? you're kinda like a faggot huh?
>>8813179ah shit forgot to clean this up so it's actual pixel art. oh well
>>8812672box please
>>8813232this is really hot
>>8812393This is a stereoscopic video? I'm liking how smooth it came out.Could you please share about your workflow?
*rotoscopic video
>>8813232Still angry? Or just too out of it?
>>8810904No
gay
hella gay
based, actually
>>8813534Heterobox?
How advanced is stuff now? Can it generate scenes from one artist with another artist now?
>>8813588You can put the fries in the bag, lil' bro
lil bro got big bro’d :skull:
>>8813588Was that not always possible?
what causes that
>>8813366random illustrious mix -> Wan2.2 I2V model -> Owl3DI used to use https://github.com/HypoX64/Deep3D which is fine but I had to modify it to have a cross-eye mode and not need to dump all the frames to disk first, it also just isn't as good and so far I haven't found another good OSS 2D->3D model
>>8813754vidgens are disgusting
Why do all vidgens use some butiful shiny skin western henti realistic detailed background slop as base.
it's wai
>>8813771because only ones using this style are interested in vidgens
>>8813757>Wan2.2 I2V modelDid you use the official weights or GGUFs? If you used the latter, is Q4_K_M (the gold standard for LLM quantizations) fine? I've been meaning to get into vidgen for a while but I just assumed it was way out of my 3080 12 GB's league, but now that I've tried some LLMs after a long hiatus and found even 30B MoE models (WAN is MoE, too) to be a breeze to run, my interest in vidgen has been reignited.
>>8813791>WAN is MoEis it really? interesting
>>8813791i'd like to know more of this vidgen stuff too, looks fun
>>8813793>>WAN is MoE>is it really? interesting> Effective MoE Architecture: Wan2.2 introduces a Mixture-of-Experts (MoE) architecture into video diffusion models. By separating the denoising process cross timesteps with specialized powerful expert models, this enlarges the overall model capacity while maintaining the same computational cost.For the record, MoE is literally 30 times faster on my PC - Qwen3-VL 30B runs at 15 tokens/s compared to Gemma 3 27B's 0.5 tokens/s.
>>8813791have tried a few different fp8 checkpoints from civ on a rented a40, haven't tried gguf yet mostly cause its more hassle to set up, might try it next time i'm doing a batch just to see if its faster since they usually are faster for llms
>>8813791all i got's a 8gb 2080 so i couldn't run even a gguf of it locally - used to have another but it died
why do they look so rotoscoped
training data is mostly not animation
>>8813808what loras are you using with Wan2.2?
>>8813817DaSiWa checkpoint with no loras mostly
>>8813819cool, i'll check it out. thanks bro
we dont need more people making ugly vidgens
Is it a MUST to input the last frame for i2v vidgen or can you somehow skip that step?
>>8813841no, if your workflow has a i2v mode and "FLF2V" mode the i2v mode only uses the first - i use the stock wan i2v one in the comfy template collection that only does i2v
>>8813627it was
>>8813855Yeah, I see now, thanks. Another thing - what workflow do you use? DaSiWa too? I can load one model just fine, but when it swaps from High to Low, it's very easy to OOM, as if it doesn't unload the first model at all. After several crashes, I finally managed to gen a vid by closing my browser and only loading a few tabs, that's how much RAM it takes.
>>8813878Nvm, I think --cache-none fixed that. Before, I was OOMing on Q4 GGUFs, now the full FP8 model loads just fine with RAM to spare. You'd think the sane default for a model as RAM-heavy as Wan would be to unload the first model the moment it's not needed since one model barely fits in the first place but cumfy gonna cumf and try to load both at once I guess. Another reminder why I avoid it like a plague whenever I can help it.
>>8805340okay i just got sent here from /g/LMG, trying to train an SDXL lora and wanted some advice about tagging, they said you guys actually care about lora generation, il just copy paste itabout tagging for the SDXL lora im working on, i know if your making your own character you should keep the tags unique so the model doesnt get confused, but when im tagging the dataset should i tag the parts of the outfit that are slightly cropped out of the image? for example gloves/shorts ect, or should i tag them even if you can only see part of them
>>8813956Do not tag anything that is not visible in the image. Partials are fine, like how you can prompt cowboy_shot with thighhighs and you'll only get the top part of them. In fact the generation parallel is how you should be looking at captions, imagine you're writing a prompt to generate this specific image using your checkpoint.>i know if your making your own character you should keep the tags uniqueYou should keep her name unique, and optionally an outfit tag only if you've decided to have a single name for her whole outfit. Everything else should be using the standard Danbooru tag, as long as one is available.
>>8813961I was actually planning on tagging different elements of her outfit actually, top/ bottoms, gloves, ect, which is why in a few images of the dataset ive had almost the full outfit but left one of the parts out such as the gloves, hopefully so it will understand what the gloves are ect, and yes i was planning on having the tags be completely unique for each and abstract enough so they arent used anywhereOh you mentioned danborou tags, i had something interesting before, i used to make lora's in 1.5 ages back, and i found having the dataset very expressive in terms of emotion but not tagging expressions worked better than if i tagged the expressions, should i do a similar thing for poses (from below, cowboy shot, side view, ect)
Has anyone had any Luck getting openpose to work in Forge Neo? apparently its still a work in progress because using a model pose made my generated image turn out like this, am i doomed to using comfyui to get posing right?
>>8813534Faggot shit.>>8813638/d/ shit.>>8813232Jewish shit.
his brain can only think of shit
>>8813981That looks like you're using a different controlnet model, with a reference image that's meant for openpose.
>>8813974>having the tags be completely unique for each and abstract enough so they arent used anywhereIf it's a glove just tag it as a glove. Then you're only retraining the meaning of glove, which starts off very close to what you want already. Otherwise you need enough images with the "gl0v3" on and off to learn exactly what the new tag means, and even then you'll like be unable to do things like glove_in_mouth.Most people either tag everything, with the downside that those pieces then become removable and you have to prompt for them to get the full outfit. Or they have a single tag referring to the entire outfit, and then some images without where she's in casual clothes or naked. Guessing it would take hundreds of images if you wanted to train every single piece on/off, when a decent character lora needs barely 20.>having the dataset very expressive in terms of emotion but not tagging expressionsSounds retarded. But hey, it only takes an hour to retry so might as well.
>>8814015No this happened when i had Controlnet Integrated (working with forge neo) enabled, preprocessor off and had one of these thigns in the file upload, notheing else was enabled as far as im aware and its how i always did it with sd 1.5 Speaking of there was an openpose extension for 1.5 which no longer works with Forge neo either, i guess my intent for it anyways is to have a lora of a character (which im currently working on) and then use openpose to pose the character in certain ways in a sense to keep her proportions and posture right, unless there is a better extension or way of doing that im not sure
>>8810150this is a week old but: how do you inpaint like this? is it because of the lora? i'm having trouble making it blend as much as possible trying plenty of adjusting settings, I tried adding this with the same setting and just gives "NoneType" of output, some things have film grain and I want to match everything as much as possible, even simple things such as "purple background" the background of the inpainted image is completely different in hue.
>>8814023thanks for the info, i wil try your way and tag them as the style of outfit they are based on anyways then, speaking of is there a way of finding out all the danborou tags used in my model, or any model really for that matter Oh and regarding the expressions, thats just how it worked best for me last time, because if i only had 1 image in a set of 20 tagged angry and i wanted her to appear angry i imagine it would only have 1 for frame of reference, has this methodology changed should i tag all of her expression and poses correctly when working on the dataset?
>>8814025fixed it, it's because of the dummy node added for some reason.
>>8814026>is there a way of finding out all the danborou tags used in my model, or any model really for that matterMost loras have metadata included you can read with a text editor, includes all dataset tags and their frequency. afaik there is no way to tell for a full checkpoint, but:illustrious trained a full Danbooru snapshot from 2024noobAI additionally has 4mil images from e621.netAssume if a tag has over 100 danbooru images it is at least partially understood.It knows what an angry girl looks like, unless she has a very unique expression that you want like https://danbooru.donmai.us/posts/3816226Just guessing here, if you were to train your own expressions they would perhaps be more accurate to her but wouldn't work well across art styles. And it for sure requires a much wider dataset to train.
>>8814037Maybe some version mismatch, I don't update Comfy often. There was no dummy node on my side.It was an extreme example at 1.0 denoise. The most reliable way to inpaint something like this would be to bucket fill the bikini with her skin color, then use a much lower value. You can also increase the "feather" value, can make edges blend better. And make sure you're not using a custom VAE.
>>8814045thanks, I tried it and kinda works, but it'll also adjust the overall image color, subtle or major. but so far this is slightly better than the others I tested. >>8814025
>>8814049If you're not using super high denoise you can also throw away the inpaint controlnet, in case that's offsetting the colors.
>>8814042no way anyone can train all the booru expressions from scratch, just the six types of smug would take way too many pics
>>8814051it works better with it than without, the lines connect unlike without where it'll just smudge something on it. the denoise is set at 1. the noob controlnet is best for what I like but it'll always darken the entire image a little. xinsir promax union kinda makes everything brighter and/or saturated. but again, so far this is the best one I've tested and hopefully one will come out or some other model/workflow.
>>8814044I don't think I have ever seen SD generate uncut penis, strange
>>8814042No im not planning to train my own expressions i just found making a dataset with more expressions lead to more maleable character poses and such, but tagging said expressions would limit if you wanted to use that specific expression, and thanks for the Infor but im actually using a Ponyxl model still which i know is outdated somewhat but i like the model
>>8814072Can't say how much of what I've said applies to Pony, it's been like two years. But yeah, it always did struggle with facial expressions.
So with artists dying and works left unfinished, how would one go about training a model of their style? I have 1111 installed and have just been messing around with PonyXL but I feel like building a model will give me a deeper understanding to get the results as close as possible to Sora's design.
>>8814070You can try long_foreskin or unretracted_foreskin. Fag.
>>8814070>>8809653>>8777214>>8716610etc.>>8814109You can just prompt foreskin usually, works fine. Unless it's a very blacked style cause those affect the dick too much.Always thought it was weirdly rare in hentai, do nips all get circumcised the way jews do?
>>8814081>how would one go about training a model of their style?I wouldn't because their style looks awful.
>>8814074thats fine, honestly i do a lot of inpainting and messing with expressions myself like changing the pupis size and direction and then just inpaining over slightly to fix it, although now you mentione it wasnt there an Expression Extension for Forge UI which basically mimicked an expression you inserted via an image? or something like that
>>8814141Seems like a lot of trouble to go through just to keep using a shitty outdated model.
grr...
>>8805340Was just about to start my Lora training and realised i have the most up to date lora trainer kohya_ss v25, and now every single tutorial i looked at is outdated, does anyone have a guide on how to set this up for SDXL models in v25, or even a json file?
>>8814211Did they really break compatibility with everything that has worked for three years
>>8814214no they just fucked around and now everything is in a different place for no real reason, and some things are missing, ive no fucking clue how im suppose to keep up with the shit when a tutorial from 8 months ago is completely outdated and wrong
NovelAI Video coming april btw
>>8814268on April 1st you mean
Any news about pony v7? Those who used it in app said that model is a pure shitshow
>>8814347well duh, it was obvious it'll end up that way since the announcement
>>8814070Noob can do them fairly reliably with the appropriate tags, another advantage of training on furfag data.
>>8814380Nah it works fine on base illu. 11K pics is way more than enough for such a specific concept.
FUCKING WHAT I DIDN'T KNOW ABOUT THISfuck I should have saved more videos
>>8814560The character looks familiar but I can't remember who it is
>>8814597where am I supposed to upload my vidgens now???
>>8814599rule34gen.com
>>8814598think I already made the joke last timeoh well
>>8812393>a capability i can only dream only a few years back>bored of infinite custom hentai after a few monthslike even my shittiest gens are astronomically better than the anime/hentai I viewed years ago, but now I fuck around with a prompt for like 30-40 mins per week.
>>8814672That's just reverse nostalgia. Go watch mahou shoujo ai again, or viper gts.
Do you guys use wildcards?I like swapping from western to hentai artists from time to time.picrel is Taka Tony
I need more good doujin and hentai artists
>>8806902this looks nice
death to mass sloppers
>>8814778incase, sakimichan, apostle
>>8814788apostle looks interesting, my current wildcard for artist includes a mix of western and eastern, {realistic,3d,(namespace:0.8),(pantsu-ripper)|flat color, no lineart, , murata yuusuke|by kiriyama taichi, watercolor (medium), traditional media|by iizuki tasuku|(by joel jurion:0.8), (by cutesexyrobutts:0.6)|by owler|by queen complex, beautiful, elegant, perfect face, beautiful body|by akamatsu ken, by ken akamatsu, official art, artbook|by kim hyung tae|by sugimori ken, by ken sugimori|by tony taka, watercolor (medium)|by nishimura kinu, traditional media}I have used incase before
>>8813448what was he still angry about?
newfag here, what caused this
>>8814843elves
>>8814790>>8814779>>8814778Nice style, I like softcore and sfw posted here. Fuck /hdg/ and fuck /edg//hgg/ will be the new anime diffusion kingdom, lets redirect all the quality post and gens here!
>>8814882I just stay on /b/ and /g/, wtf is hgg?
Greetings from /jp/
>>8815216aww! which /jp/ general are you from?
Ignore the bullshit. Some schizo has a grudge against our general, don’t feed him./2huAI/ team
why are cuckchan users like this
>>8815253We were ignoring it just fine without your input
>>8815198I am hyped for the new movie
>>8815278they've been saying it's coming out this year for 3 years now
>>8815545Isn't that a show for little girls
>>8815546No? Don't tell me you're lumping in stuff for kids and manchildren like Precure (which is just sentai with a female cast) and mahou shoujo. Madoka is particularly grim, too.
>>8815547>Precureit has the most porn, tho
>>8815550that doesn't really have anything to do with target audience
>>8815216where's the /h/?
>>8814044oh yes. mmmm-hm. this man one /hgg/premium bibiciii saar.
>>8815547I was...am.
Greetings from /jp/2huai! Here you go.
>>8815661post kyouko getting railed as penance
>>8815794weird request but okay
>>8815794NTA but I wanna Kyoukopost too>>8815742I don't... really want this...
>>8815800thanks for breaking the awkward silence
Move>>8815821>>8815821>>8815821
>>8808582I have a 5080 and can't find a good workflow. What are you using?
>>8806902Catbox?
>>8810596
>>8810432
