>>2813593 hit the image limitPlease keep loli on >>>/b/, western styles on >>>/aco/, and vtubers on >>>/vt/Non-Stable Diffusion (Bing/Dall-E/Midjourney) is OK but keep it /e/.> Where do I start?https://rentry.org/sdg-link> Newshttps://www.sdcompendium.com> Models and Mixes>>>/t/1190826https://rentry.org/sdmodelshttps://rentry.org/hdgrecipeshttps://huggingface.co/WarriorMama777/OrangeMixs/tree/main/Modelshttps://huggingface.co/andite/anything-v4.0https://huggingface.co/AnonymousM/Based-mixeshttps://huggingface.co/gsdf/Counterfeit-V3.0> LoRAshttps://gitgud.io/gayshit/makesomefuckingporn#lora-listhttps://rentry.org/hdglorarepohttps://rentry.org/trashcollectshttps://rentry.org/zp7g6> Upscalinghttps://openmodeldb.info> Recommended Extensionsmultidiffusion-upscaler-for-automatic1111a1111-sd-webui-tagcompletesd-dynamic-promptssd-webui-controlnethttps://github.com/Bing-su/adetailerhttps://github.com/jexom/sd-webui-depth-lib> Recommended models for /e/ contentAbyssOrangeMix3, Anything-V4.5, Based67, Counterfeit-V3.0> ComfyUI resources:https://github.com/comfyanonymous/ComfyUIhttps://www.sdcompendium.com/doku.php?id=comfyui_0087https://github.com/shiimizu/ComfyUI_smZNodes> ResourcesA1111 Wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/FeaturesOld guide: https://rentry.org/sd-mashupTags: https://danbooru.donmai.us/wiki_pages/tag_groupsLoRa Training guides: https://rentry.org/59xed3 | https://rentry.org/lora_train | https://rentry.org/lora-linux-troubleshootingadvanced merging: https://rentry.org/BlockMergeExplainedModel similarity calculator: https://huggingface.co/JosephusCheung/ASimilarityCalculatiorAnimation: https://rentry.org/AnimAnon | https://rentry.org/AnimAnon-Deforum | https://rentry.org/AnimAnon-AnimDiffRegional Prompting: https://stable-diffusion-art.com/regional-prompter/> Related>>>/aco/asdg>>>/aco/csdg>>>/d/ddg>>>/g/sdg>>>/h/hdg>>>/trash/sdg>>>/vg/aids>>>/vt/vtai>>>/wsg/aicg
Request anchor
FAQ:Noob guide based on questions I see a lot (upscaling, danbooru tags, xyz plots, | and [from:to:when] syntax)https://youtu.be/Jya6qcokqZQ> What's with catbox?4chan strips metadata containing prompt info, but catbox preserves it. Prompt information can be loaded from the PNG Info tab of the WebUI.https://catbox.moe/For temporary storage and slightly higher size limit: https://litterbox.catbox.moe/4chanx extension: https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fbPro tip: Look at people's uploads to see if their filename contains a catbox link. If it does, you don't have to ask for catbox, it's already there.> How do I fix bad hands?Try the sd-webui-depth-lib (needs sd-webui-controlnet) and this inpainting guide:https://rentry.org/inpainting-guide-SDAdditional hands library: https://civitai.com/models/67174/900-hands-library-for-depth-library-or-controlnetDepth lib inpainting tutorials:https://files.catbox.moe/8r4y83.webmhttps://files.catbox.moe/j46t3b.webmThere are also embeddings in the negative prompt that attempt to address this. I personally don't like the approach because it can change the style too much, but some people swear by them.https://huggingface.co/zylux/orangechillmix/blob/main/badhandsv5-neg.pt> I can't make large images, I'm running out of VRAMTry to generate at a base resolution within a 768x768 box and upscale after that.If you're on a card with 8GB or less VRAM, use --medvram, which saves VRAM but takes more time.Keep in mind that highres fix is very VRAM hungry. Don't use it if you're running out.The ultimate upscale and multidifussion extensions upscale using tiles, allowing you to upscale images in pieces, taking only the VRAM necessary for the tile size you choose. The tradeoffs are tiling artifacts (it's stitching photos together), and it takes more time.https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Troubleshooting#low-vram-video-cards
Previous thread highlightsMar 23 - May 01: https://files.catbox.moe/6kbf5g.zipMay 01 - May 28: https://files.catbox.moe/foykah.zipMay 28 - Jun 17: https://files.catbox.moe/2ahwt8.zipJun 17 - Jul 22: https://files.catbox.moe/qkprij.zipJul 22 - Aug 17: https://files.catbox.moe/8m4chv.zipAug 17 - Sep 10: https://files.catbox.moe/5wj806.zipSep 10 - Oct 03: https://files.catbox.moe/t1y1t6.zipPrevious thread catboxes (all catbox images scraped from the archives):12-20-22: https://files.catbox.moe/9db5fc.zip01-03-23: https://files.catbox.moe/tquitr.zip01-12-23: https://files.catbox.moe/y5pema.zip01-26-23: https://files.catbox.moe/44hi5e.zip02-13-23: https://files.catbox.moe/a1zfgl.zip03-03-23: https://files.catbox.moe/tbxb20.zip03-26-23: https://files.catbox.moe/3alkfb.zip04-09-23: https://files.catbox.moe/og6ffw.zip04-20-23: https://files.catbox.moe/j86g5k.zip05-01-23: https://files.catbox.moe/okbfdl.zip05-11-23: https://files.catbox.moe/dhayxj.zip05-28-23: https://files.catbox.moe/zphj2u.zip06-07-23: https://files.catbox.moe/t10yt8.zip06-17-23: https://files.catbox.moe/8do6za.zip06-25-23: https://files.catbox.moe/w02b88.zip07-08-23: https://files.catbox.moe/u34buy.zip07-22-23-1: https://files.catbox.moe/432nug.zip07-22-23-2: https://files.catbox.moe/5g6vq0.zip07-22-23-3: https://files.catbox.moe/72r8oo.zip08-04-23: https://files.catbox.moe/mu10y5.zip08-17-23: https://files.catbox.moe/1e64oy.zip08-27-23: https://files.catbox.moe/17k3k1.zip09-10-23: https://files.catbox.moe/uivxkw.zip09-20-23: https://files.catbox.moe/znyect.zip10-04-23: https://files.catbox.moe/e6ak7w.zip10-14-23: https://files.catbox.moe/yzovui.zip
Survey time. What models are you using these days (for /e/ related stuff)?
>>2815836Wow my shizuku made it into the collage. I'd make more but my fucking linux install died for the second time this week. Sometimes my automatic1111 would get stuck loading a new checkpoint. It froze the computer, I reset, then linux is gone from options at the bios boot menu. Can't get it back, so I have to reinstall the entire OS again for the second time this week. Good thing I back up my creations to a different drive after every session.
Note to Nvidiafags: Driver version 546.01 makes the vram overflow optional, so if you were stuck on an old version for ai generation, you can update now.
>>2815824More Kuroki Rei please.
>>2815834Thank u for putting the Opal's together :D, love seeing her next to all those other amazing gens
>>2815837meinahentaiv4 and based67
>>2815824Animation I'd like to see someone take a shot at adding full detail to (and make it less western looking). Characters are Homura and Madoka. Too many loras for them to list; kind of up to whatever someone wants to go with.
>>2815850link lora then
Do we have a theme for november?
>>2815875squirrel girls for no nut november
>>2815875also maple leaves
>>2815837Store Bought Gyoza for the most part.
>>2815848>reinstall the entire OS again Alright, I reinstalled but prior to that I saved the entire stable-diffusion-webui folder to a different drive, then put it back on the linux install after. I can't exactly reproduce y old images, they're all very close, but slightly different even though the SD stuff is the same as from the old install. Am I overlooking something?Sorry to ask for tech support, I'll be back to posting images as soon as I know I have reproducibility.
>>2815883I've heard of differences from machine to machine, but this is the first I've heard of problems reproducing on the same machine. Still, it's a new setup so don't be too worried about not being able to exactly reproduce something.My understanding is that the way the GPU produces the seed image can vary slightly from machine to machine, but if you're on the same hardware that challenges my understanding of the problem.
>>2815875
how do you figure this shit out? I've wrote yoimiya a dozen different ways in the prompt and tried a dozen different models not a single one made a yoi anything, counterfeit, divineelegance, reproduction, bluepencil, nothing worked.
>>2815885It's weird. I did the load data from image and made sure it's all the same, but the left is the image I had before I had to reinstall and the right is on the newest install. The install just refers to the OS, the stable-diffusion folder is the same, I just sync the entire thing to a different drive after every session so when this happens, I can just put it back.It must be something weird, would one of the OS packages changing affect things?
>>2815892lookup the term Lorait's a model you can train on a specific character that you call for in a promptdownload the lora you want and place it where you have your stable diffusion folder: >stablediffusion>models>Lorayou can then call it in prompts looks like this <lora: name of lora:1.0>here's the catbox for this gen:https://files.catbox.moe/rbmbjv.png
>>2815824Requesting for more not!Priestess please.
>>2815866Don't think it exists.
Was it worth it?
>>2815706https://files.catbox.moe/lrauqt.pnghttps://files.catbox.moe/t2eq62.png
>>2815893Don't know if you do this, but there is an annoying bug when you switch models in the same session trash is not completely collected from memory and the results are not true to what is supposed to be loaded now. I am not a programmer and don't care if that is the right wording. It is impossible to remake images from memory mixed in such a manner, so I have to make sure to restart the UI everytime I switch models.
>>2815823Where can I get a 'Stable Diffusion' installer package?When I searched Google it gave me 'Easy Diffusion' and I'm guessing the two programs aren't actually the same thing?
>>2815953>a 'Stable Diffusion' installer packageDoes not exist. Read the OP.
>>2815849oh cool, they added it to the driver control panel and it can be set for any program.took them long enough.
>>2815937Hmm I definitely noticed this sometimes. If I couldn’t recreate an image restarting helped. I don’t think it is the problem this time. Think I’ll just do another system reinstall and see if that helps. If not then oh well.
>>2815823>>2815827>>2815814Last thread was right on the average at 10 days. Here's the scrape:https://files.catbox.moe/5ye11b.zip
>>2815953You can use Easy Diffusion, that's an option, but I recommend following the guide in the OP.>>2815958Thank you for your service anon.
Foxxy Thicc
hose time
If I draw some colored blobs in paint and then run it through img2img + controlnet (canny?) can SD transform it into a good picture? I won't be home until sunday but I wanted to test that kind of stuff when I get back so maybe I can already draw some drafts on my phone
>>2815970kinda. I've seen that work to an extent
>>2815970yes it works. you might have to fiddle around with the settings.
>>2815974>>2815971Thanks bros, at least I'll have something to keep me busy during these days.
>>2815910is it supposed to be the goblin slayer priestess?
>>2815913its fucking OVER then
Do you guys find it easier or harder to coom with material you've generated yourself vs real art? To me I just see it like an extension of masturbation, you know? Like instead of imagining something to fap to, I get my pc to imagine for me. Thoughts?
>>2815976You can even make paint level edits to pics on your phone and have SD go wild on it. Like get a sfw image and draw tits on it. Actually works.
>>2815932catbox?
>>2815987At first yes but it gets boring after I realize I could have my fetishes in the palm of my hands
>>2815987My fap sessions are now like 2 hours long because I keep trying to fine tune the perfect pics as well as create variants akin to Artist CG sets that you see on the panda, and for some reason I always need new content when it's AI.
Nijika my beloved
>>2815989This opened my eyes to a lot of possibilities, thanks anon
>>2816012Cute, catbox please?
>>2815824Requesting lewds inspired in dark fairy tales (Hansel and Gretel, Red hood, Cindrerella, that sort of stuff)
>>2815977Preferably but its OK as long as she looks like her.
Is it possible to convert this scene with Rias from Highschool Dxd Hero to the artstyle from the first 3 seasons? (Example: https://civitai.com/models/17257/dxd-all-girls-gasper-rias-sona-yasaka-venelana-akeno-asia-tsubaki-xenovia-grayfia-kunou-gasper-rossweisse-irina-kiryuu-koneko-kuroka-serafall-ophis-ravel)
>>2816025https://files.catbox.moe/sbrq29.png>>2816061ok i might bake some in a few
>>2816065>trained this lora on AnythingV4.5ewww gross don't touch it
>>2816074Well it was just used as an example to be fair. There are at least other Loras which do the same job.
>>2815987>material you've generated yourself vs real art?At the end of the day, they're both drawings. If people were drawing what i wanted to fap to, then it'd be the same but it's hard to come by LL girls with accidental thongslips.
>>2815850>>2815978The gens only need to resemble her. Pic and video ( >>2815913 ) related..
Tonight turned out to be a Ruby night for me. It's hard to get it to do these rally slight panty peeks.
>>2815824akane tendo from ranma½, goth, black eyes, black cropped shirt, black shorts, torn clothes, barefeet, split
>>2816132
>>2816133
>>2815824woman with long hair wearing long sleeve top and thigh high stirrup stockings or legwarmers.picture shows entire bodyboth hands visible/bare/uncovered/exposed.area from armpits down to upper thighs uncovered/exposedtoes and heels exposed.(sometimes called reverse bunny)or -------------------------------------------------------------------------------woman with long hair wearing only elbow length fingerless glove on left arm, right hand bare/uncovered/exposed. both hands visiblepicture shows entire body
>>2816134
>>2816132>>2816133more of this
>>2815996I've actually started pulling stuff from other e threads and seeing what img2img can do with them. Kept things fresh desu. Kinda fun too
>>2815850I only occasionally lurk these threads but when I do I always see your request. What anime is she from? I looked up the name and the results look nothing like the image you always post
>>2816172Kuroki Rei is from Vividred Operation. The pics posted are of Clara from Star Rail. They are posted to contribute something to the thread.
bride
Bigger resolution: https://files.catbox.moe/f6mrbo.png
>>2816190November is Sweater Weather
>>2816191
>>2815934catbox?
>>2815824Requesting military stuff but with a fantasy twist. Like I don't know... a female soldier with a witch hat or a tactical knight or things like that
>>2815824Requesting someone fix her panties/skirt. They basically fuse together. The designs and colors on her panties are kind of irregular too (3 different shades of black, uneven frills).And as one other minor fix- the strap on the left of her shirt shouldn't be white, and the fabric around it is kind of the same color as her skin. Character is Sakura Kyouko from Madoka.
Added boobie to some anon's gen, I don't remember which thread it was posted originally though.
Looks real good. Thanks for this, Kantoku Anon.
Definitely in my top 10 gens ever.
>>2815824Re-request for more gens of Julia from Fire Emblem.
Any models for Sian?
>>2816171catbox? nice style
Went back to my mad hatter and see if I could make her tits lactated. I manage to do it but I had to edit multiple gen together to make it work.Also, I think I did this anon >>2815824 request.
>>2816289I mean this anon >>2816026
>>2816145Lora is a real roll of the dice.
>>2816291
>>2816293
>>2816294>>2816291>>2816191>>2816171>>2816178i like those nice lightings & pose
>>2816302
>>2816303
Late night sleepy maid
>>2816294catbox?
>>2816340https://files.catbox.moe/ia1fn2.png
>>2816341where do I get the assup lora? couldn't find it in the usual places..
>>2816354>couldn't find it in the usual places?https://civitai.com/models/186916/ms-real-lite-ass-up-top-down-bottom-up-lora
Why are so many of them fat as fuck with large drooping tits?Why does art have to represent the world we live in.Love the skinny/petite ones...
>>2816355Thanks. There isn't really a practical way to go from the information in the catbox to the lora on civit since the names and hash don't match.Also: >Do you want to support me? Consider being a patron on Patreon to vote on train polls and access stuff early. >Lite models are less trained versions of the Full models but yet usable ones. Full models will be on Patreon and after 1-2 week span I'll release Full models on our Discord server.lmao
imagine being a cringelord who has a patreon
>>2816360Imagine not being one, free money
>>2816356probably cause the people behind them don't share your opinion.
>>2816367If you need to do requests instead of what you like then it becomes a job, the total oposite to "free money".I dont think anyone would get into your patreon just to see what YOU want to make and what you want to make alone, to get subscriber you need to come with actual plans that makes you get patrons that includes request like it or notIts perfectly fair of course, but the "I do what I want and get free money back" is just an illusion. IF you want money you have to commit to the bussiness model or you get jack shit.
>>2816374I made AI commissions for some money and I quickly began hating it, so ur barking up the wrong tree.Maybe some people like it though, like that guy genning elves for months on /h/. Also no, ur entirely wrong - fanbox and patreon are places you sub to see the author's content, not to request things.
>>2816360>>2816374Ai commisionning is not going to be a long career, look at bing images, once it has a nsfw version elsewhere, it's donebut i guess prompters still have 1-2 years
>>2816309catbox?
>>2816389Nobody was making enough to quit their day jobs.It usually takes artist years to generate that kind of following and the "1girl, solo" market was immediately saturated with significantly more supply than demand. But people are lazy and stupid. Look at food delivery for example. If there are people that can't be bothered to get their own food and would rather pay way too much for someone else to bring it to them, then there are going to be people that will never care to learn what tools to use or how to prompt properly.Food delivery as a main job is fucking stupid, but if you need a little extra cash and you're not doing anything on your day off anyway, it's an easy way to make a little extra. Most traditional pixiv and patreon artists just make what they want and people subscribe regardless. Some will go a year without updates and still retain hundreds of paypigs.There are also tip jar type payments, and all it takes is one little whale to make it worth the little trouble it takes to set up an account.The clock might be ticking for the casual early earners, but I would be surprised if there aren't already certs or courses on how to train and prompt.
Bigger resolution: https://files.catbox.moe/0x7wb9.png
>>2816390What ur config?
SD appears to be adding this crazy contrast/saturation filter in the final step of every image I generate these days. I have no diea why. Is there a setting for it somewhere? Putting things like "saturation" and/or "contrast" into negatives doesn't seem to do anything.
>>2816426Sounds like the vae. You might want to try a different one or a model that doesn't have a baked-in vae.
>>2816427I use the same vae I've always used and it was never a problem before. Maybe I'll try tinkering with the settings a bit.
>>2816430Maybe give it a no positive prompt test. If it's still coming out whack, you have something going on with your negative prompts (something like greyscale or monochrome can change the output quite a bit) or you're suffering from the A1111 memory leak issue and should reboot the webui.
>>2816432Thanks for the tip, but I think I learned what it is now! It's an effect called "burn". There is an extension to help deal with it. I'll try it out.
>>2815824Please img2img pic related. Background optional tho a beach may be it. Expose her skin more if you can.
>>2816482
>>2816484amazing transformation
>>2816487NICE HAT
>>2816026 Larger file: https://files.catbox.moe/93z29j.png
>>2816487This is seriously impressive
>>2816391https://files.catbox.moe/7so7sk.png
>>2816484Nice. Thank you.
>>2816200https://files.catbox.moe/aggnek.png
Wanted to post the png file but it's 9mb.
>>2816598just use catbox
>>2815824Requesting: symmetrical dockingChallenge: asymmetrical docking
>>28165984chan strips metadata anyway and for high color images, the naked eye can't tell the difference between png and jpg
>>2816512>>2816510Sorry, but I won't gonna praise your inputs any more, Everything you post is so amazing that posting every time how amazing they are is just redundant
>>2816599>>2816605https://files.catbox.moe/phij9k.pngThe best thing I can do is use the jpeg version on 4chan and catbox for the png.Good thing that Ultimate SD upscale give me both files at the end.
>>2816603
>>2816714snapchat filter lookin ass
ANYONE please help, when trying to train a Lora I get this whenever I use fp16: runtimeerror addmm_impl_cpu_ not implemented for 'half', and when I use bf16 it gets too slow at doing steps, I use kohya ssI'm in Ubuntu 22.04 with an AMD GPU
>>2816603dock
>>2816722nice milkers
>>2816723youtoo
>>2816727may i ask what drove you to make this img?>>2816726why is the left one so pissed?
>>2816603Very cool: >>2816727
>>2816209Deploying tactical Witch.
>>2816603Don't you hate it when you accidentally summon two?
Best of the day.
>>2816730>>2816209One more Witch
>>2816728witnessing a crimson aurora and the transient nature of the worldalso should drink less so i don't fuck up my replies>>2816729thanks
>>2816734Sir,this is an ecchi board?
>>2816502Thanks. Spent some time getting the settings right, but it's fairly consistent now. let me know if you have any requests I can try.
>>2816026
>>2816719Sounds like kohya doesn't have a half implementation for AMD? Obviously full precision is gonna be slow, but really the best place to ask is the kohya github.>>2816727beautiful sky>>2816783we genderbendergen now
>>2816727AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA please promt her nude
hello all
>>2816833
>>2816834
>>2816728angry, angry about boobs
>>2815824Request for something similar to be done with this pic as it was done in >>2816482 and >>2816484
Good fucking work, fuckers. Many of these are substantially better than what many hentai artists crap out. No wonder twitter-and-instagram quasi-artists are crying so loudly about evil villlains writing prompts.
I feel like I'm having a stroke every time I read through the guides of how to set up this ai shit lol. I've used bing and NovelAI but this seems like a chore lol
>>2816848I'm not sure if I'm included in this or not but AI kind of blows my mind. For example the leaf shadows on her thigh is kind of crazy. Just imagine this shit 5 years from now.
>>2816850Also it allows artshitters like me to clean up stuff like 6 fingers which I missed with very little effort lmao
>>2816727amazing. was miyuki lora used for the clouds/plant?
>>2816850Don't get your hopes up. The low-hanging fruit has been plucked, so I wouldn't expect any major breakthroughs in the next 10-15 years. In the next couple of years, expect skepticism and doomsaying. I'd expect AI from 2028 to be pretty much on this same exact level, except that it'll take less computing power to generate due to optimization etc. Also, it's still only good for porn because of how meaningless the output is, and this will not change until we get AGI, the general intelligence, going. At that point, the tech will probably click, and all bets will be off.It just looks like, Artstation to Pixiv, the art is so brain-dead an AI can easily approach their level of creativity. Like, take those infinite asian artists painting the same few types of girls over and over and over. Turns out that stuff was so stupidly formulaic, a robot could do it.
>>2816860retard. you were proven wrong last month with dall-e already
Bigger resolution: https://files.catbox.moe/yalm1x.png
>>2816861I'm talking about the future, not the past. How stupid can you get?This is how tech works. You can't just expect regular breakthroughs because we've just had one. Look at the self-driving cars, you ignorant dipshit. A breakthrough followed by a lull of almost a decade now, with nary an end in sight. This is because the first 95% of progress solve the easy problems, while the remaining 5% are the truly hard ones that take more time to solve than everything prior combined.Use your fucking brain for once, you coomer.
>>2816867This is honestly a fairly good assessment. Couple that with the fact that the law is trending towards partly outlawing certain practices that we've come to rely on and it makes it that much harder to make significant advancements in the imagen field.
>>2816659>>2816722>>2816723>>2816729All this are great>>2816763Ill now roll on your gacha game>>2816805>we genderbendergen nowIts curious its not more recurrinth theme in folklore. Myths are filled with heroes of all pantheons crossdressing but in fairy tales it aint that common
>>2816867Im honestly not sure where can we even go further since from what I've seen dall-e is near perfect, for image gen we dont even need much of a breakthrough, just further refinements.Though you're still retarded because we will get a breakthrough in the video/animation department soon™ once big companies get on it. Not to mention text and translation
>>2816876The only people who claim to be able to predict technological breakthroughs in the near term are scammers and retards. Since you're not trying to sell me anything, I'm assuming the latter.
Bigger resolution: https://files.catbox.moe/lmfx75.png
>>2816885>only retards predictback into the jungle with you
>>2816895There are important words after the word predict too. Without them, you won't understand the meaning. That's how language works, harambe.
>>2816885by this same logic you predicting the lack of any makes you retarded.That quote is also regarding specific predictions not a general "it'll happen". You need to have no fucking clue about AI to think it's going to stagnate now.
>>2816928Breakthroughs are hard to predict. Stagnation after a major breakthrough (because the easy stuff has been done and only the hardest stuff remains) is the default. So it's easy to predict stagnation, whereas it's almost impossible to predict when another breakthrough will actually happen.God, you're a retarded bunch.
>>2816867>self driving carswhy not look at tanks or actual carsor guns, machine gunsplanes maybehow about recent times and PCs?Of course we won't have major breakthroughs in self driving cars, there's not much more to improve.>>2816933Easy to predict (subjective much? it's easy for me to predict AI will not stagnate) or not, trying to predict technological breakthroughs (be it happening or not) is retarded by your own admission.I hope you're just baiting and are not genuinely this dumb.
>>2816806althttps://files.catbox.moe/iwitye.pngextrahttps://files.catbox.moe/1iglj4.png>>2816805>>2816855thank>was miyuki lora used for the clouds/plantnope just touko and 0.4 beksinskicolor comes from img2imgnow i wanna try it with miyuki
>>2816937>Of course we won't have major breakthroughs in self driving cars, there's not much more to improve.You are ignorant enough to not know how ignorant you are. Get acquainted with the concept of levels of driving automation and go from there.
>>2816937I think you need to revise how parantheses work, your second point is unreadable
>>2816986that's just your poor reading comprehension
>>2816867>narythis word doesn't make you sound smarter it's just cringe. Please just say "with no end in sight", that's what nary means, no
>>2816954Nice. Thanks!
>>2816849Its kind of like brute forcing learning another language.Once you get it going it's like Cypher in The Matrix talking about seeing a "blonde, brunette, redhead" instead of code.
>>2815824Requesting someone try making this more detailed/higher quality overall, but with the same general pose, body type and "outfit." It's Megumin from Konosuba playing on a guitar in her underwear.Preferably use this lora: https://civitai.com/images/1892396If possible, make her bra/panties black and her expression less aggressive.
>>2816988>cringeThis word doesn't make you sound smarter either, to put it mildly.
>>2816988I must politely yet firmly disagree with your assertion that my modest lexical selection betrayed vain affectation rather than erudite eloquence. The notion that an expansive vocabulary anatomizes one's intellect renders shallow insight; truly, it is the judiciously chosen word that demonstrates linguistic panache. Though your preference for brevity and directness has merit, to dismiss as "cringe" the embellishments of language risks sacrificing nuance upon the altar of simplicity. I shall endeavor to take your perspective under thoughtful advisement, yet cannot conscience stripping my modes of expression down to sterile words alone.
>>2816999that's funny because of the pic related but don't let the Cypher analogy dissuade you from starting AI art. there's no language learning involved. you don't need to know Python. once you get it going you're dealing with a user friendly web UI
>>2817022hi, I can try this for you but I can you clarify first? you want to keep the outfit but you want to use the Megumin LoRA. that LoRA is meant for doing the classic Megumin outfit with the hat, the dress, the cape, etc. The only element from that outfit I can see in this pic is the collar.so, which elements do you wanna see from the Megumin LoRA?
>>2815987Definitely better. It's hard to find arts that match my fetishes.>Like instead of imagining something to fap to, I get my pc to imagine for me.I have aphantasia so imagination doesn't work for me.
>>2817052Really like the styling on this one. Catbox?
>>2817060https://files.catbox.moe/rnb4ny.pngI just noticed I messed up the face when I did the ear removal. Was probably drunk
>>2817064what do you do for cleanup?
>>2817072Llama cleaner and Gimp
Can someone just inpaint Zero Suit Samus' render as nude? Just keep her darker blue boots. I want to use it for a brawl mod and can't use SD right now.
>>2817075thank you. with gimp you mean edit out flaws or do use some filter or something?
>>2817085Its good for easy fixes and getting the colors that I like. Everything in the top Colors tab is useful. For filters Enhance > Sharpen I use on pretty much every pic. Noise at a low opacity can make a nice grain effect for some gens. Components > Channel mixer > Red in Blue channel is good for skin tones.
>>2817086cool thank you very much
>>2817075this is pretty neat. i have no clue what im going to use this for, but im sure something will come up
>>2817056Didn't realize that one worked that way. I mainly went with that one because I liked the general style, in terms of the face and shading; is there any way at all you could apply the overall shading style (and the face) to the entire thing without giving her her default outfit?If not, or if it's just too much work, then kind of just go with whatever would work best (as long it's not a flat or really generic style.)Just one other thing- if at all possible, try to make her fishnet stockings into fishnet thighhighs (cut off/top of the thighhighs similar to this)
>>2817088> 1landscape, huge_mountains, (thick_bush:1.2)
>>2816860>I wouldn't expect any major breakthroughs in the next 10-15 yearslol lmao
>>2817101most of the example pics of that LoRA were done with the SakushiMix model --https://civitai.com/models/78056/sakushimix-finishedCheck it out. If you want I can try to redo your pic in that style but it won't have anything from the Megumin character besides hair style and the collar. Happy with that?
>>2817101>>2817127also what kind of background do you want?
>>2817154>>2817155>>2817101>>2817022in the end I used a different model because SakushiMix didn't handle inpainting too well
Figured I'd post it here as well.My first attempt at a proper picture with stable diffusion.Took me 2 days lol
>>2817055Not as in learning a program language, but there absolutely is a new vocabulary to learn.As someone with no python experience and no idea what a windows batch file was, following the guide took regular goggle searches just to get through many of the steps.Beyond that, models require specific terms to get the desired output.For example, if you are going to try fulfilling a request, you cannot just enter the request verbatim into the prompt.It needs to be translated into something the program will understand.And each model will interpret the prompt differently.Then there are commands that need to be added to the .bat (or somewhere) for improvements or to even make some things work.A few threads back anons were going over how to get better results from the animation models and how to make the gifs/webms loop.I gave up trying to figure out what they were doing.
>>2817159You can try running the image back through SakushiMix with img2img and see if something serviceable comes out.
>>2817154>>2817155>>2817159Sorry I didn't get back sooner. Thanks for doing what you did here, but I wasn't looking for one this realistic; if you're up to trying it again in a more typical anime style, use whatever lora works best. I'd just like it for it to be her.Background doesn't particularly matter. If you're up to it, have it be like a concert stage I guess.
>>2817166
>>2815824Requesting Komori Yui (from diabolic lovers) gens with her all nude. Jusy her or a look-alike that resembles her as there is almost no proper art of her in her birthday suit.
>>2817162>eyesyou sure ur using a vae bubby
>>2817189
>>2817194
>>2817179I did use one, yes. What's the problem with the eyes?
>>2817206kinda blurry? lacking color? not sure
>>2817208Oh I made them like that on purpose. I like that they look like rubies.
>>2817214>>2817208Oh wait, I misread that.Do you need to tell adetailer to use a vae separately? Cause I didn't do that.
>>2817179the vae changes the color saturation, doesn't have much to do with the eyes. He's just using a western style model.
>>2817215no, no need for that afaik. It's just the model as the other anon said, if it doesn't look off to you then it's probably a preference thing>>2817233im aware, it's easier to spot with the eyes and the whole thing looked similar to vaeless symptoms
>>2815824Requesting Bloodborne or Lies of P kind of characters, Inspired on the setting I mean, not specific characters of those games
>>2817237
>>2816209Sword is too hard
I am a complete noob when it comes to this, although already browsing some wikis. I was wondering what would be the best way of creating AI art drawn in various styles of a character that has like 3 different color palettes (using one palette at a time ofc)? Thanks in advance!
>>2817295get a lora of the character that you want to create and a lora of the style you want to use and then fuck around with the weight of both until you like the output. the model can make a huge difference for certain styles. some are trained for a specific style. look around on civitai and pick what u like, but for a beginner id recommend to pick one in the op. i never tried to use a specific color palette, but a search in the extension tab gives a couple of results. im certain you´ll find something there.
>>2817295Train a lora on the images. If they're tagged properly you should be able to switch color palettes with prompts, worst case if you run into issues you'll have to make a separate lora for each color.
>>2817277this unfortunately
how much longer before AI can reliably do hands
>>2817214nothing wrong with the eyes, I like them. pretty good for something done so early on
>>2817170Thanks again- but this doesn't particularly resemble her either. Did you use a lora other than one for her? (it also seems excessively blurry)
Last one
>>2817368That one's spooky.
>>2817368>>2817370Is Cujo ecchi?
>>2817125>wow, this year there was a significant breakthrough!!>it means that every year there is one, right?!>I am very young and know nothing about tehcnological progress!!>wow I legit don't understand that 2022's breakthrough was based on decades of slow iterative development that coalesced into a critical mass!>I legit think 2022 and 2023 will just keep repeating!!
>>2816860>>2816867I think if your referring solely to the black-box realization of LLMs you are on point - and for a lot of anon coomers who are using LLM's as a time-killing dopamine button they'll get jaded once they hit a recognizable level of consistency (aka mediocrity) in their output.I think you underestimate what creative industry even the current technology can enable - good and bad - to people who possess some creative wherewithal. The example that is already being bemoaned is people using AI assets instead of hiring someone or using stock assets. Practically this enables the realization of games that wouldn't have been otherwise - but depending on the artist's taste the use of AI can go anywhere from taking away from the execution to enhancing it.
>>2817447I'm the person you're replying to, and agreed on all points.
>>2817441>>2817444catbox pls?
>>2817454https://files.catbox.moe/zhuk6a.pnghttps://files.catbox.moe/65qpz9.png
>>2817368The others look nice but this one the dogs look terrifiying
>>2817483oh wow! catbox please?
>>2817364alright here you go. when I put this through Deepboru it gives me the "megumin" tag so it must look like her.let me know what you think!
>>2817493some parts a little blurry when you blow the image up. that's because I did a lot of inpainting because SD doesn't deal well with all the details when you treat this as a full body pic. there's the guitar, the lingerie, the face, the hands...I'll try to tidy it up.
>>2817493>>2817494>>2817364final version
>>2817498just don't view it in Chrome because it blows it up beyond 100% zoom and makes it blurry
I'm enjoying putting girly stuff on boyish girls
>>2817490https://files.catbox.moe/ey8n7j.pngUsed this model: https://civitai.com/models/174090/marinerrequired a bit of inpainting in the toes afterwards
>>2816209was thinking of doing a pic like that with inpainting but my brain is too friedgave it a shot in dall-e and it's pretty good for this kind of thing with just a prompt so it seems like you can easily do it yourself (unless you want booba)in terms of /e/ what do you imagine it's supposed to look like? aside from weapons, camo pants + witch robe/hat, bikini plate armor with a modern helmet?
>>2817504>giant tits>gargantuan hips>pretty long hair>BOYISH GIRL OMG
>>2817532nice style can i get a catbox?
>>2817606
>>2817607
>>2817609
>>2817603it's dall-e 3
>>2817612i see. thank you
>>2817532Thats super cool
where the mummy fans
https://files.catbox.moe/zm7lgb.png
>>2815823>loras trained on SD 1.5 typically only work with realism models>loras trained on nai or nai-like models typically only work with cartoon/semi-realism models > using regular regularization images assists in the learning process by making sure it's not too rigid with the train directory images (for example, this allows you to put a subject in any environment or any kind of clothing even if every single train image has that person inside and wearing only one type of clothing)So here's my theory. People smarter than me probably have already figured this out but I think this is how it could work. I regularly train realism loras based on real people. Train on SD 1.5. Have a regularization directory of 500 or 600 images depending on who I'm training. That directory is to ensure the model does not overfit. My biggest gripe with training in general is that I can use these models on wheelers and models but they do not work well, if at all, on anime models like aom3, anything v4, Meina, etc. So here are my questions:1) what having a regularization directory of let's say $250 realism images and 250 anime / cartoonish images allow it to have more versatility and be able to work with not just SD 1.5 models but anime models too?2) my theory is that perhaps merging SD 1.5 with another enemy model like Nai or anything V3 could also help with this too.To the people ITT who are way smarter than me, does what I said and why I wanna accomplish make sense? I want a person-lora to work both with a model like realistic vision AND an orangemix model. Is this possible? Are the methods I described feasible? Would they even work?
>>2817652I'm not a training expert, so I can only speculate based on what I have used, but there's a gap that needs to be bridged between realism and anime style. Most modern anime models already have SD 1.5 and nai as ancestors, so that is not really the issue. I've used concept loras based on realistic style and it's been hit and miss on whether they work in an anime context. It usually ends up looking muddy and incomprehensible.After you bridge the gap, what do you expect the output to look like? It will be neither anime nor realistic. Basically the same problem applies in real life if you were to ask an artist to do something that is both realistic and anime.
>>2817656>and it's been hit and miss on whether they work in an anime contextAt which point we've faced with catastrophic forgetting?
>>2817702What do you want? A list of realistic loras that work vs ones that don't? Like I said, I'm not an expert on training so I don't know how things are trained and so I don't know what works and what doesn't. I don't know what goes into training to make certain loras compatible with the nai/anime strain of models. Maybe there are other anons that can chime in, but all I know is that it's a tall order to make realistic things work in an anime context and vice versa.
>>2815824I want to see girls getting [spoiler]headpats[/spoiler]
>>2817264catbox?
>>2817705I'm not an expert on this either, but AFAIK overbaking and forgetting are the first things the model developers should care about and avoid of. Generalizing is the purpose of machine learning after all. I want to have a model which is able to create not only photorealistic images, but also cartoon, anime etc, and it should not just be good at it all, but also be able to effortlessly float between the styles and even combine them (which is quite possible) if that's your wish. That's what you really can call a good model. In this regard, NAI is garbage. And many newer models don't even know some things the base NAI knows. And that's exactly the reason why you can't really "bridge the gap". Simply take a look at DE3, it's not perfect, but it has shown potential. In the end it all comes down to the training methods, quality and quantity of the datasets used, the size of the models and computational costs.You'll probably want to bring style or concept to your LORA anyway though, so you should try to avoid fucking up the model as well. It the base is good, it shouldn't be very hard though, I imagine, since the good model already recognizes most of the things present on your dataset and won't crap out your lora.
Bigger resolution: https://files.catbox.moe/798wz0.png
>>2817716Very nice... Now show me her butt hole.
>>2817716Is she supposed to only have three fingers? You really should work on your QC if you're using it to promote yourself.
boob curtain test
>>2817493>>2817498OR here, sorry about being late to get back (again.) These are indeed quite a lot better and much more in line with what I was looking for, thank you. Though what is leading to the blurriness in these?
>>2817781It's an artifact of the diffusion process. These things aren't magic, if you want the best results you're going to have to do some work yourself.
>>2817783I'm mainly just curious; I just haven't noticed this type blurriness in most other AI anime art.
>>>/h/7651924
>>2817825
>>2817826
>>2817827
>>2817652tl;drwhats the issue?
>>2817828catbox pls
>>2817834probably the tagging
>>2817789it depends...if you're viewing the image in Chrome it blows it up beyond 100% which makes it blurry.if you're looking at it normal size, there are some blurry patches. it's basically because we're doing img2img of a full body portrait. in trying to respect the original image, SD has a hard time filling in little details of a full body portrait, like hands, face, and the guitar. So I have to go in afterwards and inpaint details, which is why you get blurry patches like around her right elbow area."most other AI anime art" that you see on here is not img2img it's text2img, so the AI has free reign to generate original details with a huge database of reference material. It's not tied to copying one image. It's hard for SD to translate elements of a source pic in another style while keeping detail AND fidelity. I could have turned up denoising strength and gotten more detail but you were quite specific about wanting to keep certain elements. Also, in text2img the AI artist will gen 20 images from one prompt and cherry pick the best ones without defects.I gave you a huge image. if you can bare to reduce the image size, it looks fine. There is a limited amount of time I'm prepared to spend perfecting the inpainting on an image that I have no personal connection to. If you're still unhappy with the results I would echo what the other anon said and I would recommend that you consider investing in the time to learn about using SD independently.
>>2817866blurriness is also a factor of the model and upscaler. This can usually be fixed with an additional img2img pass at low denoising.Here I've downscaled >>2817498 and re-upscaled it at 0.4 denoising. Notice the outer lines especially are more crisp and defined, and this is likely because I used based67 which is known for producing very clean lines.Note that I'm not saying it's "wrong" to have softer lineart or a chromatic abberation effect from a model, it's just a different aesthetic and some people view this as being blurry.
Should I cover up the pussy too?
Their marching band outfits are fun.
>>2817866>>2817877You've really done more than I ever expected and I hugely appreciate that. I'm still trying to gauge the kind of setup I'd need to get what I need out of SD, and this has also been more than informative.
Thoughts?
>>2817836>1200+ images in todays folderhttps://files.catbox.moe/uusael.png
>>2815824Can the AI improve the coloring of a picture? Like taking this picture and make the coloring more elaborated with shades, brights, etc
>>2817907
>>2817907>>2817931i wanna see her spreading her pussy
>>2817910I'm unaware of any loras that allow for stippled coloring like that, so that detail and a lot of other fine details like the gradient coloring and linework for shading are getting lost as I crank up the denoising. This is with lineart controlnet, denoising 0.65. Below this, the colors just didn't pop. fwiw I think this came out objectively worse than the original.If you want to preserve the original details and just saturate it more, curves in photoshop/krita is probably what you're looking for.
>>2817799Very nice style. Give details if you feel like it pls
Is posting real photos for references and such permissible in this thread?
>>2817953yeah, nobody mods on /e/
>>2817907For me, it's the high ponytail with blunt bangs
Spent the day making thumbnails for my LorasIt's a ton of fun and I highly recommend it.
Gusset Pantyhose
Good old Shirow.I like a subtle lingerie nightgown.
Ishikei
>>2817994
Dako
Drow
>>2817992
Bluethebone
Viera
>>2817992>>2817993>>2817994cute
>>2817907>>2817828>>2817827>>2817931Just to see if I could do it.Needs more work, but I have four basic [without the swimsuits] loraless looks that I like.
>>2818012Looks like we're both in a similar mood.
>>2818004Thanks :D
>>2818020
>>2818021
>>2817237It's having trouble making it as gloomy and depressing as BB, but it's a start.
>>2818011
>>2817991A few more before bed.Melon22
Nyantcha
Squishy underboob
>>2817833Catbox please?>>2817907high ponytail + blunt bangs activates my neurons
im retarded, anyone got catbox for this or any similar ones?
drew
>>2818070cool
genning this produced some eldritch horrorhttps://files.catbox.moe/wmpclr.jpg>>2818073thanks
>>2818070moth Toko. Cute spider tattoos here too >>2818075
>>2818078i was attempting some sort of mosquito vampire at first but then i decided to just go along with the whims of sd, now that i look at it again it does resemble a moth kek, thanks for pointing that outa mothpire is fine too
>>2818080nice spooky expression and cool cock monster with balls on the sides of its headlol>mothpire+2 pen when biting through clothes
>>2818050https://files.catbox.moe/v0wl08.png
>>2817992this is some god tier hayrstyle
>>2818083Thanks!>>2818075NIce gen. That catbox is cursed, though
>>2818080Neat. I'm also a fan of trippy and eldritch type horror stuff. Though I'm not gonna push it too far and keep it /e/.
Some really good gens here. I'm waiting for someone to gen a Loli Tainaka Ritsu from K-on!
>>2818099>LoliMight have to go to /b/ for that.
>>2818115well it doesnt have to nude or rule breaking, Just never seen one made
>>2818080That is honestly one of the coolest AI generated images I've ever seen. Good job, man. True art. Maybe you could catbox? Also, what model were you using?
>>2815824This is an ancient photo I was curious to see if someone could try converting into anime and expanding on. Ideally, I'd like to see it as Ryuuguu Rena. Here's loras for her: https://civitai.com/models/7894/ryuuguu-rena-higurashihttps://civitai.com/models/104536/rena-ryuuguu-when-the-cicadas-cry
OP here. Page 8 and 35 images to go. Won't have time to do it tomorrow so new thread tonight.The last highlights have a couple gaps that can fit small but very wide images. If you want to fill it, feel free to post, otherwise I'll fill it myself with either downsized pics from the thread of filler gens.Good day, anons.
>>2818115>>2818123Just link it offsite.
>>2818105
>>2818177I see you've found my /vt/ posts.Anyway. Gonna try to take us to a new thread in an hour or two. Anyone browsing, feel free to help the thread get to the image limit first. 27 images left, done with highlights, so maybe post your old favorites or experiments that didn't turn out perfect.
>>2818082>+2 pen when biting through clotheshehterrifying>>2818087sd does best cursed body horror when it's an accident>>2818097those are some cool horntacles>>2818131thanks, the model is just based64there's a bit of a process to how it was genned because it was img2imghere's a catbox of the original genhttps://files.catbox.moe/zizi4u.pngpng info and send it into img2imgnext you put this whole image into img2img as the sourcehttps://files.catbox.moe/4l69i4.jpgi'm on an older version of sd so you might not get the exact replica but i just tested this process and it replicates the original gen just fineboth loras are on civit, touko lora is excellent, the artist lora might need some wrangling with neg prompt because it loves turning girls into monsters and making everything brownprompt in there is half nonsense cause i like typing in random shit and seeing if i get anything out of it
>>2818215your gens with galaxy girls are really cool
>>2818196catbox?
>>2818222Thanks for sharing your workflow. Interesting use of img2img.New thread incoming.
old highlights and old gens in no particular order
New thread>>2818252>>2818252>>2818252
>>2818227>>2818228>>2818229>>2818230>>2818231>>2818232>>2818233>>2818234>>2818235>>2818236>>2818237>>2818238>>2818239>>2818240>>2818241>>2818242>>2818243>>2818244>>2818245>>2818246>>2818247>>2818248>>2818249>>2818250>>2818251I wish you would stop spamming these collages and let the thread fill up naturally.
