This is a thread for /u/ images generated using Stable Diffusion, and for technical discussion related to generating images with Stable Diffusion.This thread is for ANIME STYLE /U/ IMAGES ONLY.Realistic styles and images from proprietary alternatives like DALL-E and Bing are not allowed (>>>/aco/ has several).> Where do I start?https://rentry.org/sdg-link> Local UIsWebUI Forge, an A1111 fork optimized for SDXL (faster genning speed on low-mid vram gpus): https://github.com/lllyasviel/stable-diffusion-webui-forgeComfyUI: https://github.com/comfyanonymous/ComfyUI | https://github.com/shiimizu/ComfyUI_smZNodes> Popular ModelsAutismMix SDXL (PonyXL base), AnimeConfettiTuneXL (PonyXL base), AnimagineXL 3 (SDXL 1.0 base), Based68 (SD 1.5), Olympus v4 (SD 1.5)all available at https://civitai.com/> Prompt SharingUpload your gens to https://catbox.moe/ to share prompt metadata or consider using the stealth png info extension: https://github.com/neggles/sd-webui-stealth-pnginfo> Resources and guidesA1111 Wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Tags: https://danbooru.donmai.us/wiki_pages/tag_groupsUpscalers: https://openmodeldb.infoLoRa training: https://rentry.org/lora_train | https://rentry.org/59xed3 | https://civitai.com/articles?tags=128645Git-wiki of the technical Stable Diffusion papers: https://codeberg.org/tekakutli/neuralnomiconSD links hub: https://rentry.org/rentrysd | More useful links: https://rentry.org/sd-mashupAnimation: https://rentry.org/AnimAnon | https://rentry.org/AnimAnon-Deforum | https://rentry.org/AnimAnon-AnimDiff> LoRA downloadshttps://rentry.org/ponyxl_loras_n_stuff | https://rentry.org/ponyxl_lora_previewshttps://rentry.org/trashcollectshttps://rentry.org/zp7g6https://rentry.org/hdglorarepo> Related generals>>>/aco/asdg>>>/aco/csdg>>>/d/ddg>>>/e/edg>>>/g/sdg>>>/h/hdg>>>/trash/sdg>>>/vg/aids>>>/vt/vtai>>>/wsg/aicg
As with most boards, we recognize that AI images aren't necessarily welcome in other threads. Please be respectful and don't post AI stuff in threads where they aren't wanted.Have fun and welcome to /udg/.
>>4165582This one's very nice, level of detail like a good 1.5 model but I see pony tags in the filename (and 1.5 models are old & busted)Could you provide a catbox, or what loras were used?
>>4165585This may not reflect the base prompt due to several rounds of inpainting.https://files.catbox.moe/xwxgxr.pngI do have stealth png info on all my gens so feel free do download any of them if you have that extension, the metadata is preserved.
>>4165580Love this style
>>4165590do you have a link for the Sokunasu lora?
Ah yes, purest form of love
>>4165595https://civitai.com/models/333041?modelVersionId=373076>>4165598https://civitai.com/models/336671/ponyxl-style-sokonasu
>>4165600Thanks!
Cunnilingus please
>>4165606I had some trouble trying to get a good one on the past but I might reattempt later and post it here
>>4165578>ANIME STYLE /U/ IMAGES ONLY>crosslinking to hetshit and futashit porn boards>OP pic that isn't even yuri, just two naked girls staring at the camera>retarded /vg/h/ style general tag in the subjectIt's apparent you're not actually from this board and you really don't need to come here just to make shitty duplicate satellite threads for whatever garbage board you crawled off of.
>>4165606It's exceedingly difficult. For me at least.Symmetrical docking is easy mode for anyone looking to get started.
>>4165620Really? I didnt find it that hard, pony is trained on loads of sex positions. I've had trouble getting a good 69 but just getting Starfire to eat Raven's muff (by asura:0.7) is pretty trivial.
>>4165625I just find it difficult to get good tongue-on-pussy action without it becoming garbled in the area of interest.
i have a scarce few gens that turned out ok while poking around
>>4165627Who is the blonde girl?
>>4165630Lillie
Post was deleted apparently due to the male in the painting in the background. My bad, this has been corrected.
>>4165633Didn’t even noticed there was a guy in the picture
>>4165634I think someone just reported it because they don't like the thread. Can't really argue with the rules though it's not a big deal, just be aware.
>>4165635Why don’t they like the thread? Is it because of the AI?
I will not participate in this thread, I'm just stopping by to politely request that you AI people stay in your thread and don't post your trash in normal threads, thank you.
>>4165636Maybe, or maybe it's just a real stickler. I'm not gonna dwell on it. Just a reminder to be aware of what's in your background.
>>4165638This kind of stuff is way beyond my understanding, so I’m just gonna enjoy the things other people made.
>>4165637thanks for participating
>>4165637I know you'll likely not see this, but I agree that AI stuff doesn't belong in most threads. I'm a dude who appreciates real art as much as the next guy, I make regular posts just like any other anon, and I have never personally posted AI in non-AI threads./u/ has historically had some bad problems with the previous AI threads that have disrespected the board. I've been dismayed by it myself. I hope we can all be reasonable moving forward.
>>4165636AI somehow managed to mindbreak a lot of people, the moment they recognize/see AI they all have the same pavlovian response. It's kind of funny and sad at the same time.
>>4165659can u plz tell me who made this, or who the character is?
>>4165683Magia Baiser and Magia Azul from Gushing Over Magical Girls.https://civitai.com/models/313943/magia-baiser-xlhttps://civitai.com/models/329022?modelVersionId=368704
>>4165644can u plz tell me who made this, or who the character is?
>>4165609Very impressive,very nice>>4165637I only post on IA related threads, don't worry
>>4165690not any particular characters, generated with autism confetti mix, using img2img to get poses from random photos
I don't care much about yuri AI pics, but I do care a lot about AI novels. Is there a good model for yuri text generation? Preferably one that I can run on Google Colab because my PC is dogshit.
>>4165728I don't know about running on collab, but there are a couple general purpose roleplay models that I've used. Not specifically tuned to yuri but certainly capable of sticking to your scenario. They're tuned to roleplay, but estopian maid especially is capable of writing prose.Kunoichi 7B can run on as little as 8GB RAM, even less if you offload some to the GPU.https://huggingface.co/TheBloke/Kunoichi-7B-GGUFhttps://huggingface.co/TheBloke/EstopianMaid-13B-GGUF
AI is fun because it lets me make all the yuri twin/selfcest I want
>>4165770Hot
>>4165770>>4165776
>>4165780Are they twins or the same person?
>>4165780>>4165782They aren't specific characters, just whatever is conjured by "short hair, red hair, freckles, blue eyes, small breasts," etc.Up to interpretation I guess but I think the latter is hotter. Down for both though.
>>4165786Personally, it’s hard to choose between the two because they’re both hot
I forgor how much I did of them the other day
Really gets hard to wrangle with more than 2 characters interacting, just too many opportunities for logic and anatomy to fuck up
>>4165801Looks pretty good to me
>>4165802They do have kind of a rough charm. But these are the cream of the crop, sitting on a pile of Cronenbergian atrocities. Well, more than normal for working with gen AI at least.
>>4165805I like to think this is somehow canon. Probably happened during one of the movies.
Does anyone have a link to all the previous A.I threads? It's impossible to search for them in the archive because the archive only accepts words with at least four letters.
>>4165809There's normally not a need to search because the person who makes a new thread backlinks.>>4117053
>>4165805Ah, so it hasn't advanced so much that you get decent looking pieces at the press of a button.
>>4165825Definitely not if you're looking for something more interesting than "1girl, standing, huge_breasts." Even then you get horrors occasionally, especially if you don't have certain settings dialed in, which can change depending on the model you're using and sometimes depend on each other.Hate to waste an image but god some of these need to be seen.
>>4165825It has, but only partially. Certain models can do certain interactions and poses very well these days, and you don't need loras for that like on 1.5. SD3 is coming out soon and it's apparently very good at prompt comprehension. Don't expect any anime/hentai models for it in the next 6+ months though, and it's gonna be very vram demanding.
>>4165824Ah that was back in December so it really has been a while.
>>4165786>when the incestuous identical twins find out they have a long lost triplet
>>4165806This is how the clones pass the time floating in the tanks of LCL.
Are there any good local AI for yuri stories now? Smut especially. I last checked a couple years ago but it was pretty bleak, models that wouldn't do anything raunchy or were just incapable of coherence.
>>4165939Oh hell yeah brother, LLMs in general are just way better than they were even 6 months ago.Might start with >>4165731 and go down the rabbit hole from there.
I've seen Comfy on colab but it's very poorly optimized for SDXL. Is WebUI Forge better? Is it on colab? I'm running a toaster so no way I can download this and use locally
>>4165945Ah I hadn't noticed that between the image posts. I'll take a look.
>>4165950I can't speak to collab, but in general, Forge is A1111 with the backend swapped out for very aggressive memory management, specifically with XL support in mind (in fact it's kinda worse at 1.5 in some cases and I'd advise sticking with Comfy or A1111 on 1.5). It'll dynamically switch to low vram mode and so on and it's plain faster in a lot of cases. So yeah if you have the ability it's worth trying out.
Got a little bit of a detached head and long arm situation going on here, but I've seen worse.
>>4165977aaaand of course right after uploading I notice the detached thigh. Fuck me.
I should really get around to trying that Krita SD plugin to fix little errors in otherwise good gens. It's a shame to have really good ones with just a single weird eye or finger fucking it up.
>>4166210You don't need the krita plugin just to fix an eye or finger, a round or two of inpainting should do just fine.
>>4166216>>4166217>>4166218Very nice, I like the soft, hazy style
>>4165939>>4165728If there is more interest and discussion about text / llms as the thread goes on, I'm open to collecting information and making the next thread OP some sort of combination image and text thread, which would really make sense for the board over just images. (I know it will potentially be a long time before the next thread, just trying to gauge interest and maybe stimulate some discussion in this thread)
I don't see the other AI thread so I'll just post dalleshit here since I figured out how to do widescreenalso it's good at my main fetish: saints row
>>4166294OP said no dalle stuff, please follow the thread's rules. Make your own dalle thread.
>>4166304Ironic shitposting is still shitposting.
>>4166213True, it did work ok. Mostly I just want to draw the damned hands myself instead of doing 40 inpaints and picking the least fucked up one.
Kobeni and PowerK I S S I N Ghttps://files.catbox.moe/mmd4i0.pnghttps://files.catbox.moe/ery755.png
>>4165606and a bunch of this too with the same pairinghttps://files.catbox.moe/nr27zx.pnghttps://files.catbox.moe/7lxnsb.pnghttps://files.catbox.moe/wee6gq.pnghttps://files.catbox.moe/pmzu3j.pnghttps://files.catbox.moe/866yag.pnghttps://files.catbox.moe/4j2mzt.pnghttps://files.catbox.moe/awi5ng.pnghttps://files.catbox.moe/x9p9o4.pnghttps://files.catbox.moe/qvzsb5.pnghttps://files.catbox.moe/cumxf7.pnghttps://files.catbox.moe/3gtnzu.pnghttps://files.catbox.moe/ehsp44.pnghttps://files.catbox.moe/fn9fgy.png
Stable Diffusion 3 will save us brothers. Trust the plan.
>>4166383heh maybe. But Stable Diffusion 2 was a bust and the jury is still out on Stable Cascade.
>>4165945>>4165952Thanks for the tip anon, I had some fun with the second one you suggested and it worked a lot better than last time I tried.Other than generating nonsense once which I solved just with redoing the prompt, I only really have one complaint - this model really doesn't want to make lengthy sex scenes it seems. Like I'll add some world info about character X liking to take their time with oral sex and the model will just spend two lines on it before someone has an orgasm. Couldn't make it improve at it. I guess it's down to trained weiting styles, do you know if there are any models trained on wordier smut?
>>4166383The problem is only with the lack of training datasets. We haven't even tapped the full potential of current models because no one can fully train them on yuri exclusively.
>>4166463Well on that 13B model I've got it to do quite lengthy sex scenes with some coaxing. But yeah it has to have its hand held. Basically if you edit what it says and string out the scene with more foreplay or whatever, it'll pick up on that. If you want it to be way more flowery and verbose, you might have to look for like a 30B+ sized model if your machine can handle it.As for specific models, I'm not well versed with what's out there, but here are some resources:https://rentry.org/LocalModelsLinkshttps://hf.co/spaces/HuggingFaceH4/open_llm_leaderboardhttps://hf.co/TheBlokeIf there is enough interest moving forward, maybe we can collectively figure this stuff out as we go along. I just dabble in the stuff and I'm learning about it as well.
>>4166476Miqu-70b at 8bpw is a great model, and only takes 74gb of vram. It's prose is a bit dry, and it does sometimes still feel like I'm railroading it a bit to get it to do what I want, but most of the time it just understands what I'm going for and executes, which is nice.Here's a small RP that is a knockoff of Asumi-chan. I might try it again with claude-3 Opus for comparison.
>>4166723>and only takes 74gb of vramo-ok anon... well, good on ya I suppose.
>>4165770also bathing robot yuri clonecest
I can already see the tags on this doujin ruining any hope of an actual Yuri.
Some old ones I generated a while ago.
>>4168639
>>4168642
>>4168643
>>4168645
>>4168639>>4168642>>4168643>>4168645>>4168646Those are pretty neat, you gave some gen ideas for later
>>4168645this one is especially cute.
I think this counts as a legit cunnilingus
>>4168781Looks legit to me. She's got a vacuum seal on that muff.
>>4168782This. Girl's got talent.
some stuff from my archive of gens
>>4170548Handholding, the final boss of AI image gen.
>>4171568
>>4171573
>>4171568>>4171573>>4171576Are you specifically prompting for one boob out?
>>4171579I am actually, just part of the look I wanted. AutismMix_pony is pretty good at prompt comprehension even the "tag" you're using doesn't actually exist. Specifically, "one breast exposed, exotic clothing, intricate gold jewelry" makes it come up with these weird asymmetrical harem outfits.
>>4171587Are you also prompting for the color of the elves? "two girls (or two female elves), one light/high elf, one dark elf"?
>>4171759I'm still experimenting with what prompting format gets the most consistent results with multiple characters. All of these were done by putting "2girls" at the front of the prompt, next to the art style and image rating tags, and then describing each character (red skin, blue eyes, long hair, dark blue hair, etc.) bracketed by BREAK commands. It's still pretty inconsistent with applying the right features to the right characters, but I think it works just slightly better than having the tags floating. Explicitly describing a contrast between the two characters might be a good idea actually, I think I'm not leaning enough on the model's natural language understanding still.
>>4171977The models basically don't understand natural language. It's not a llm, the input is just images and the collections of tags they were trained on. Natural language works to a limited capacity because base SD was trained on LAION, which is a dataset of images captioned in full sentences. However, most fine-tunes are trained on booru style tags, so you're just invoking disconnected tags for the most part even if you use the form of full sentences.BREAK just resets the priority within a 75 token context and doesn't really separate the concept of what it's rendering. Basically its greatest effect is resetting the priority of tags that are 50-60+ tokens in. But this becomes blurry when you manually tweak strength.
>>4172024Good to know about BREAK, all I knew was that it somehow manually decided where the 75 token chunks got divided. I figured any difference was mostly placebo effect anyway.>Natural language works to a limited capacity because base SD was trained on LAION, which is a dataset of images captioned in full sentencesRight, that's what I mean. PonyXL seems to retain some of that ability, but I've been using booru tags exclusively and having the problems described above. The real solution is honestly just taking the time to inpaint each character with a unique prompt.
Baroque yuri
>>4172037>inpaintHow do you do that in ComfyUI?
>>4172037I wouldn't say BREAK is exactly placebo, but a lot of people use it expecting it to do something it doesn't do. Like when people use (((tag))) without understanding that () just multiplies it by 1.1, so it's the equivalent of (tag:1.331), or when there are strange and exotic negative prompts that don't do anything.But yeah, if you want to be specific, regional prompter or inpaint is the way. One easy way to make the image pop is to use inpaint on the eyes to make sure their eyes are a different color. Adds variety for not a lot of time investment.
>>4172045Unfortunately I am still an a1111 pleb, so I can't tell you.
>>4172186heh, nice
>>4172024For the record, this is incorrect enough to give me an aneurysm. Most notably,>However, most fine-tunes are trained on booru style tagsthis is technically correct but not the case with Pony. It was trained on captions in addition to some bizarre mix of danbooru and e621 tags. Technically, you will probably get the best results by using caption + tags in your prompt.
>>4172541Though, to add to this,>technically, you will probably get the best results by using caption + tags in your prompt.is only really correct if you're using either base pony or a shitmix something close to it (AutismPony for example). Confetti is fucked in terms of prompt comprehension anyway and will give you whatever the fuck it wants.
>>4172541e621 is still a type of booru, it's just that the tags don't line up with danbooru. This just means that when you have to learn to invoke multiple styles of tags. And fair point about pony being trained also on captions, but I wouldn't call it a fine tune, it almost qualifies as a base model because it diverges so much from XL 1.0.
in pixiv i randomly tag stray pube to artist who look like draw yuri/ambiguous but mostly draw hetplease make this secret lesbian waning handshake
>>4173215>>4172930I really like these ones. :) you've made my night anon.
>>4173614Different anons but thanks. I like megukas too.
What GPU do you use for yuri diffusing?
I wonder if I should upgrade my GPU? Rtx 2060 12gb btw.
>>4174014>2060 12gbI'm afraid you should.
>>41738186800XT 16GBAMD is not very fun for AI stuff in general, but Stable Diffusion is one of the better supported things.
https://files.catbox.moe/pme2c3.pnghttps://files.catbox.moe/uoi1lh.png
https://files.catbox.moe/bcxwg6.pnghttps://files.catbox.moe/l5e3au.pnghttps://files.catbox.moe/v77f5t.pnghttps://files.catbox.moe/jgammq.png
>>4166671>>4166964cute
Been messing with sd-forge-couple. It's somewhat wonky but there's definite 2girl potential there. It's functionally the same idea as regional prompter but more intuitive. Don't expect it to be a magical solution that solves all your problems though.https://github.com/Haoming02/sd-forge-couple
box onegai, these are nice>>4175603was this with regional prompt/attention couple?>>4176936beautiful
>>4180790they both have stealth pnginfo, install this in webuihttps://github.com/ashen-sensored/sd_webui_stealth_pnginfo
>>4180790>was this with regional prompt/attention couple?No. For 2 non-specific girls, XL models can do that sort of thing. It was inpainted but the base gen was pretty much the same as the prompt you see there. And yeah as the other anon mentioned, both already have metadata with stealth png info.
>>4180802>>4180810thanks anons!
I am new to this stuff, but does anyone have any idea how this guy does his stuff? It's some of the nicer stuff I've seen from AI, and the multiple character images also look pretty good.https://twitter.com/Shark__LJ/status/1750459472647188773https://twitter.com/Shark__LJ/status/1752635503479083384https://twitter.com/Shark__LJ/status/1745999082873823729
i won the gacha
>>4184337Is the photorealism poster mad at being excluded? Last I checked, this isn't /b/ and threads can have rules. "No AI" for instance.
>>4184302most likely Midjourney/Niji with some noise added in, but I also see the pony skin colour, so pony with some LoRAs is also a possibility
Back with some more TextAI. Had a really fun machikado roleplay. I was mostly testing the new model, so I did some weird stuff like punching Shamiko, to see how the model would respond, but even that went perfectly! One of the most fun rp's I've done.
>>4184588What model are you running? That one that I have no hope of being able to run because I don't have that much RAM, or something else?
>>4184644This is Cohere's Command-R+. https://dashboard.cohere.com/You can run it for absolutely free by signing up to their API, or use it locally if you have >72gb of vram. Wrangling it can take some skill, but it's a great model.
>>4184647Ah ok. I'm kind of a local model guy but thanks for the reference. If it's a GGUF that can be split between RAM and VRAM, 72GB is achievable for someone who has a bit beefier machine than I do, but I don't have much hope for being able to run anything that uses more than 32.
>>4184647>72gb of vramWhat GPU even has anything close to that?
>>4184678Two A6000 maybe?If you've got ten grand to upgrade your pc, that is.
>>4184678There are various formats, some that allow you to split among GPUs and some that allow you to split between GPU and CPU. Probably very few people still have mining rigs from before the Ethereum merge, but a rig like that with 3-4 GPUs would probably be enough. If you have 64GB CPU RAM and a 16GB VRAM GPU, that's 80 total and leaves about 8GB RAM for the system. A high end gaming machine will meet the requirements. It would probably be pretty slow and impractical though.
>>4184678I run three 3090's.
>>4184688i've got 24gb VRAM and 64gb RAM.Would that work?
>>4184688Wait, so you can use system RAM for AI stuff too now?I have the same specs as >>4184702 (and could get 64 more GB of system RAM if necessary), what are the best models I could run?
>>4184702Yeah dude. You should be able to run 120B sized models, though 70B and 35B will be faster and more practical. The format you want is GGUF. Offload as many layers as you can to your GPU and let the CPU RAM pick up the rest. Use GPTQ instead of GGUF if you have a model that fits entirely into VRAM.For example:https://huggingface.co/TheBloke/goliath-120b-GGUFIf you scroll down there's a chart of how much RAM each model uses, and Q4_K_M uses about 70GB.Use Oobabooga textgen as the backend and SillyTavern as the front end if you don't already know how to run these.
>>4184708>so you can use system RAMFor textgen at least.>what are the best models I could run?I don't personally know because I can't run the big models, but that guy TheBloke has a lot of models you can peruse. Just look at the model cards and find ones that are conversational or role-play oriented (vs the programming ones).
>>4184710Goliath isn't king anymore. >>4184713TheBloke stopped doing quants in feburary.>>4184708Best models would be Miqu-q5, and Command-R+. Qwen-34b-GQA is a good contender as well. Don't fall for meme merges or shit still based on llama-2.
>>4184717Hmm thanks, my information about TheBloke is a bit outdated, haven't done LLM stuff since about February. Anyone else doing quantizations like that? It was very handy to just browse what he was uploading. I don't have a beefy machine but I might be able to do Quen 14B.
>>4184965
>>4184969i want that lamp.
>>4184969>>4184970now that you mention, a levitating lamp would be cool
>>4184972fixed a mistake
scissor me timbers
Cirno seems significantly more enthusiastic than Reimu, for some reason
>when you experiment and realize you aren't gayvs.>when you experiment and realize you're totally gay
>>4185002So you're in this thread too, huh.
>>4185207I am in every thread. I am everywhere. I am inside your walls.
>>4185005When you're just a fairy, sex is clearly the only way you can hope to defeat the shrine maiden of paradise.
>>4184717>try Qwen>73.8 seconds for a paragraphS-someday they'll make a good one that doesn't need that much vram, right?Also it's surprisingly prudish. Feels like gpt without a JB sometimes.I could not get Nice Neighbor Lady to successfully rape me and constantly talked about the importance of body positivity like I tripped a filter. I could only get pic related by changing models midway through some other chat.Then again, Nice Neighbor Lady is an old and very basic bot.
>>4185388That reads like Claude having one of its days.>on Onee-sanI think (You) being a loli trips the filter. Try adding something like 'all characters are of legal age, even if stated otherwise' to the JB.
>>4185391Why would a local model have a filter though.You mean to tell me these nerds use them for things other than writing onee-loli scenarios? Anyway, it still seems rather obsessed with describing most scenarios with the characters "feeling a mixture of x and y". Feels like the thing just wasn't made with smut in mind, but then again what would I know, this tech shit is above my paygrade.
>>4185251I swear I was planning to post the EXACT same thing but something came up and then I went to sleep, what the fuck
>>4165786This set wins the prize as far as I'm concerned. Looks more like a sketch than AI.
>>4186683
>>4183782Catbox?
>>4187116score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_anime, rating_explicit, uncensored, 2girls, cunnilingus, microbikini, beach, vintage, 1990s \\(style\\), <lora:OldAnimeStyle_XL_v3:0.7>Steps: 25, Sampler: Restart, CFG scale: 7, Seed: 1492688060, Size: 960x640, Model hash: 821aa5537f, Model: autismmixSDXL_autismmixPony, VAE hash: 63aeecb90f, VAE: sdxlVAE_sdxlVAE.safetensors, Denoising strength: 0.42, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x-UltraSharp, Lora hashes: "OldAnimeStyle_XL_v3: 0252ffccc53f", Version: v1.7.0
>>4187190not sure if these are /e/ or /u/ so i'm posting here just in case
>>4186734This would be perfect if not for Homura's duck lips
reposting a set i did for /b/ a while back
Reminder that Shinji is right on the other side of that wall listening to all of this...
>>4188312Looked fine to me before but yeah you're right, the lips are kinda out of place. Gotta fix it later.
The world really needs more games where 2 stripper princesses goes on an adventure together and save the world.
>>4190655Based Ka-ching appreciator.
>>4191477That game was so disappointing.
>>4191794Was it? I really enjoyed it. The only disappointing part was the lack of multi player.
does anyone know what's happening? how can i fix it
>>4192843GPU dying, or not enough steps. Or rendering stopped halfway through for some reason.Try again and see if it happens again.
>>4192843keep the prompt, but change models. if that's not enough, remove a tag somewhere as well and maybe change the number of steps a bit.
>>4192843go into settings and switch emphasis mode to "no norm"
>>4192861>>4192843That is actually a vae issue. Set it to "none" if you are using Autismmix
>>4192898apparently it was a vae problem. "no norm" was already activate and the prompting was just too messy. thanks guys.
>>4193155What model is that?
>>4193159https://civitai.com/models/288584/autismmix-sdxl
I wish it was possible to inpaint one of the two girls and replace her with a photography/photorealism girl, so that an anime girl is kissing a RL girl.
>>4193244this might be possible now with attention mask for IP-Adapterhttps://www.youtube.com/watch?v=4jq6VQHyXjg
>>4193155I also have my asshole and pussy in the middle of my right buttcheek.
>>4165599requesting a few of these, so precious
>>4196270Do you mean those 2 girls on specific? that would be a little hard since I didn't prompt anything related to girls appearance but I'll see what I can do
>>4196278Based, thank you! And no I didn't mean them specifically, just the cute jap highscool girl phenotype
so my thing that I like is non symmetrical couples, petite girl with buxom girl, tends to randomly give the smaller girl huge breasts, im using nai3 though, i heard local has more tools, would they help for this problem?
