Kissu EditionPrevious Thread: >>>>8827652>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/OneTrainer>NEWShttps://huggingface.co/circlestone-labs/Animahttps://huggingface.co/ChenkinRF/ChenkinNoob-XL-v0.2-Rectified-FlowOP Template/Logo: https://rentry.org/hggop/edit | https://files.catbox.moe/om5a99.png
>no forge neo in opanti-gradiomitic thread
>Why are you doing [thing]Why not, I am just that bored so anything goes I guess>Why is so [thing]Because it's me>Are you going to do [thing]Maybe I will maybe I won't>Can you [thing] more [thing]No, usually I determine the degree of [thing] while working on the gen with no plans on adding more of [thing]
>>8834453Oh yeah, we should start adding that one for the next threads
>>8834452Style wise? a little I guess, you should work on those anima styles, chatWhile interesting and seemingly easy to do, they are not there quite yet
Anyone tried the anima full finetune script yet? Or is the schizo reasoning on why loras are inherently flawed only applicable to sdxl?
>>8834459that was always schizo and i never saw any real evidence (i.e. and x/y plot) that proved it, everything that was said pointed inexorably towards an armchair baker with zero actual experience
>prompt @koku on anima>generic 2000s slop completely unrecognizable as him>add game cg>still slop>add some of the more recent titles he worked on with a fairly consistent style that's more distinct than his earlier works>STILL slopWhat the fuck is this model's major malfunction? Fucking WAI has more of an idea what the guy's style looks like.
>>8834458>>8834452>mfw
>>8834484it's aco
>>8834483Yeah this happens to some artists for some arcane reasonBeen waiting for some autistic anon to explain it for the peasants
>>8834475Idk, from my own testing I always liked the finetunes more than the loras of the same dataset, but the effect kinda vanished when I tried to extract them.
anima Illuga lora doko
after you die from AIDS
>>8834490it happens with the vast majority of the artists i care about for mei'm (c/)hoping that it will be better with the full release
>>8834494now this i can believe
Can any of you gents do a lora on One PieceLike copying the pre-timeskip look or the wano artstyleI love anima and can't go back to noob but I hate that it sucks for this
>>8834502no
>>8834499Like what artists anon, besides kokuI'll give some of them a try to ascertain it
>>8834502kys shonenfag. how about you ask the power of friendship for a lora?
>>8834508kek'd
>>8834507why would i care about what you think? i know my use cases
>>8834511Then stick your head up your ass you worm infected faggot
>>8834512lol
>worm infected faggotnah lil bro rly went and called blud a sakura stan
Why is chat so edgy today kek
sakura disrespect in the big 26 AINTNOWAY
>>8834502it's kinda based that you only like the early part of OP but the style is fucking ugly and i refuse to make an ugly lora
>>8834514What's with the worm thing? I only know of one /vg/ gachagame shitposter using it, and he's pretty new.
anima doesn't recognize my favorite aritst what do anons
>>8834507Kobuichi and Muririn (they almost always draw together so it's one of the few cases where artist mixing works, just poorly), Shida Kazuhiro, Izumi Tsubasu, Amazuyu Tatsuki, Harada Takehito, Pochi (pochi-goya), Ishikei, Yaegashi Nan. All of these either look nothing like themselves or just terirbly bad.
>>8834483just make a lora, unironically
>>8834520>worm thing?sakura having worms or faggots having worms?
>>8834523No, it's way too bothersome.>need to collect a dataset>need to curate a dataset>need to label a dataset>need to use the right training settings (what settings are "right"? fuck knows!)>need to wait however long it takes>need to compare the different checkpoints to see which one is the bestMeanwhile if the arch wasn't stuck in 2023 and just had an edit mode, I could upload one singular image, say "do my prompt in this style" and be on my merry way.
>>8834520It's a theory that homosexual and other degenerate sexual behaviors happen because of mind altering worms that use these behaviors as a way to propagate themselvesIt's been shown in multiple studies that some worms will make people more promiscuous, much like the flu will make you more sociable before visible symptoms show up
im still using joycaption for captioning nat language, anything better come out recently?
>>8834530is there a local joycaption? i'm still using eva03 large tagger or whatever the fuck it's called from 3 years ago
>>8834526>Meanwhile if the arch wasn't stuck in 2023 and just had an edit mode, I could upload one singular image, say "do my prompt in this style" and be on my merry way.there is not a single model that can actually accomplish this. No, NBP, grok and dall4 cannot actually do this, they're all too corposlopped. Even nai struggles to do this consistently because vibetransfer is garbage and so is their style reproduction/consistency.
1 year ago i mean
>>8834533>zero-shot slop has zero shot at being goodnext you'll tell me that I should have bought bitcoin in 2016
>>8834532yeah its a comfy extension as well, it's fine I guess but pretty small model I wouldn't be surprised if something better was out now
Is there even a point of captioning images with nsfw nat language? with the te being censored and all? did the anima guy ever give us an example of how images were tagged?
>>8834537So there are three ways to censor a model. You remove all loli, bestiality, etc from the dataset. Or you make it so everything captioned with these produces a blank screen or a "safety" image, like what Auraflow did unintentionally. Or the SAAS way, you run recognition models on the prompt text and on the output image, then block the user from seeing the output if it's tagged with "bad" stuff.Third doesn't apply to local. Second I've never heard of anyone doing on purpose, it's unreliable and probably has negative side effects. And in the first case, the encoder still produces an output when encountering something it doesn't recognize, and you can align the Unet to whatever that output is.He mentioned training with many different caption styles, including pure tags, pure nl, tags first, nl first and more. I don't remember the exact wording, check the HF readme.
>>8834522I tested only a few of these in according to my own tastesHarada Takehito worked great, Ishikei worked, although it looks more like his older stuff than his newer stuff, Pochi and Yaegashi Nan gave me generic slop
>>8834552>worked, although it looks more like his older stuff than his newer stuffThat's a recurring theme too. You'd think "newest" would help but lol, lmao.
>>8834530Beta one or alpha two? Is there a consensus on which one is better?>>8834532Autotagger is still goated for booru tags, and you can cycle through various modelsBut I really need one for natural language
>>8834557link? or is that what it's called on github
taggui has a bunch of newer taggers as well
>>8834556Yeah I've noticed that as well. I managed to get the old fate artstyle by using date tags and downright ABUSING the weight but I never managed to get the newer styles>>8834558Autotagger? Here: https://github.com/picobyte/stable-diffusion-webui-wd14-taggerBe advised though, I couldn't get it to work with Forge Neo.
>>8834559>go to the page>look at screenshot>UI isn't missing obvious features>big ass preview panenice.
>>8834566manual annotation is a bit cumbersome with it and i still prefer waifuset for that.
>>8834559What's this taggui anon?
>>8834567i'll see. i'm using boorudatasettagmanager, the russkies have made it a lot better in the last couple years
>>8834570That's not pancho
>>8834571https://github.com/jhc13/taggui
jordach status???
>>8834596vanished in Bangkok after he gambled away the cascade donation money and then some
>>8834435I've been keeping my noob mixes down to 3 artists (honestly, most of my mixes with 4+/5+ could have been shaved to 3 and barely affected the gen but cope) and been having success using 1 at full strength and scheduling the other 2 at 0.55>>8834565>Be advised though, I couldn't get it to work with Forge Neo.Same. Crashed the entire thing for me so I personally switched over to >>8834574. I miss just having a tab to auto tag for me but I kinda like the ui on this program so I'll manage.
>>8834552>>8834556I've personally swapped from 'newest' to 'year 2025'. Another thing to note, experiment with samplers/schedulers. I've had styles look off with one that look almost artist perfect with another. Really depends on the artist too. Softer, more shading, not so flat works well on Euler A. The opposite does better with dpm++ 2m sde or er sde.
>>8834672Very :>=ly
Why all DP (anal and vaginal) poses that show actual penetration are so weird
>>8834779I'm finding that a lot of sex poses with 2 people are a lot lower quality than solo shots too. Any tricks I should be applying?
>>>/trash/81279915:joy:
>>8834574I might be retarded, what's the use case for this?
>>8834786Aside from inpainting, not really, I assume that you are already using all the tags you should be using
I want to test 4 tags and I want to see how they look in several different combinations, is there a way to do this without doing it manually?
>>8834794>tags you should be usingCould you enlighten me? I'll cope for a bit you don't just mean "masterpiece, best quality, highly detailed"
>>8834790I use it for gens where the image isn't from a booru, you know already tagged up, so maybe something I run into a trailer, twitter, or artist concepts of old games, etc. Run it there and have it do all the work for me and then just plug it into a ui to gen with.I'm sure most use it for compiling datasets for loras.
>>8834786Are you talking anima or illu/noob? The latter clearly is a vae limitation. You're stretching that shit thin the more you pile into the image. Hence why when you have /hgg/'s bread n butter, 1girl solo looking at viewer close-up, it'll effortlessly look crisp. As for anima, it does a lot better due to the larger vae but you'll still have an issue here and there which you can chalk up to the preview model and the fact it's only been trained on 512x512. Or at least that's the going narrative.
how do you find inspiration for poses and then figuring out how to type it out in a way that anima understands?I'm struggling with coming up with creative stuff so I thought I'd try looking at some references but I have no idea how I'd even begin to explain how these pose are set upI love natural language but I think it's harder to gauge what works and doesn't, unlike tags
>>8834843Oh that's a shame. Do you think a custom VAE would help?
>>8834845>I love natural language but I think it's harder to gauge what works and doesn't, unlike tagsA little yeah. With noob, I'd throw in a shit ton of tags that I think would accomplish the image I saw in my mind's eye and when it worked, great. When it didn't, I'd mald and keep at it until I got something close or something completely unique.Anima, on the other hand, you can do the same but the nl support gives you even more control so any failings are truly a skill issue or lazyness. Describe to it like the lobotomized model it literally is what you want in one sentence, throwing in actual tags in the process, and you should get what you want mo sweat. I think camera angles is the most difficult thing about it since that's always influenced by the resolution.>>8834847Nope. Those normally make it worse, especially with inpaint. The vae is only 4-channel so there's no getting around that limitation.
>>8834800x/y plot in scripts section of a1111 based uis (at the bottom)>>8834832https://danbooru.donmai.us/wiki_pages/tag_groups
>local music gen is now a thing>and it's actually surprisingly really lightweight>but the trainers for it are garbage>also captioning is absolute cancerjust kill me
>>8834857NAI status?
>>8834859absolutely fucking dead in the water. so dead they're not even a zombie or a skellington, they're just dust.
i hate nai so much
Based
i will never have 25 dollars...
>>8834857please elaboratei want to musegen for memes
>>8834857Is voice cloning, particularly music lyrics replacement, still where it was in 2023? I looked it up a bit and while normal voice cloning/TTS did make some progress with stuff like Qwen3-TTS, I couldn't find anything beyond RVC for lyrics replacement.
>>8834875there's a thread on /g/ >>>/g/108164777the model came out a few days ago. The UI and training portion of all of this is basically vibecoded trash. But it's in general really lightweight. A 12gb 3060 can generate two 90 second tracks in like 35 seconds(read: you can generate faster than you can listen to it on a $200 (used) GPU). It can do lyrics if you want, but captions are all LLM generated wordy garbage, lyric sheets require really specific syntax and it's all NL garbage.The trainers are bad. Really fucking bad. It recognizes that there's a u-net and a TE(and the ability to effectively train unet only) but it only gives you the option to variable a single, nondescript LR that I have no idea if it's actually for said u-net/TE or if it's something retarded like the GLOBAL LR that gets set for the optimizer.>>8834877honestly haven't played with it that much in that respect. Right now I'm just dipping in for the experience and to learn how it all functions. Someone posted something like a lyric replacement in the /g/ thread but it still sounded pretty TTS robotic tier.
>>8834881>cumrag AGAIN*yawn*
>>8834881>only one LR settingend me
p-pancho aniki... *cough* ,, t-that's 2 updates we n-need.. o.. onegai.. *cough*
>>8834881>Generate a full 4-minute song in ~1 second on RTX 5090, or under 10 seconds on RTX 3090 with ComfyUIWhy can't vidgen be like this?
>>8834881the samples sound like early Sunowhich i guess is pretty good for local
>>8834882there's a gradio UI, you can completely ignore cumsoft. The better trainer is commandline only.>>8834883it gets worse. here's a very brief list of my complaints >>>/g/108194641oh also it has two different architectures for training. bogstandard LoRa and lyco-type! But it's. Uh. um. It's LoKr.>>8834884just be like me and completely ignore anima until the full res model is out. Maybe by then it'll have significantly improved training support, too.
>>8834887lmao okay i guess i'll just sign up for suno or something
i already like anima more than sdxl desuexcept for no upscaling
just upscale with nai
>>8834891sorry chuddite my only $25/mo sub is a t3 to my oshi
>>8834889I'm enjoying the vibrancy and detail from a 1.3-1.4x hires anima gen compared to the 2-3x cn upscaling coping of sdxl thoubeit
>>8834893what kind of retarded psychopath 2-3x upscaled instead of 1.5x or something, you need help
>it's unheard of to upscale a shitty 1024x1024 sdxl gen to 2048x2048 and higherCan't you civjeets just accumulate buzz to upscale or some shit if you can't afford a gpu?
>>8834896i have a 4090, you are just stupid
>>8834897silence gpulet
No you don't, Rajeesh
10 minutes on a gtx 970, being poor is no excuse
>>8834485box please?also, what are good negatives on anima
>>8834945i did some testing a while ago and imo it should be pos/negmasterpiece,(ye-pop deviantart:1.2), worst quality,personally i also add (@hero neisan:0.2) to negs because it's intensely sepia and fat two things i do not want
>>8834526This couldn't be any more accurate
>>8834948>(ye-pop deviantart:1.2), worst quality,should there be a comma
>>8834952i won't say it doesn't matter but i'll say it's hard to predict how commas affect style negatives
>>8834953russell noted spaces matter a lot so i would assume it is important. problem is the example caption in the model card has them on first line with no comma
>>8834956good to know. i found out the other day that the TE is sensitive to misspellings, i was tying in a character's name wrong and it didn't work at all which would be a problem at all with a CLIP TE
wouldn't*
>>8834952>>8834953>>8834956>>8834960Comma:https://files.catbox.moe/gl8vjb.pngNo comma:https://files.catbox.moe/evk53s.png
>>8834962images are broken for me
>>8834962catbox is fucked, use litterand anyways i'm gonna test it myself
>>8834965Fucking catbox.
>testing on a single fucking imagebegone retard
>>8834967I'm retarded, ignore the filename in the previous post, it IS with a comma. This is without.
>>8834936>gtx 970Had that ages ago. Still remember the entire memetroversy behind the 4gb but not really. Served me well enough until the 20XX class.>>8834948>and fatWe don't want you in the ojisan club.>>8834967>>8834970>the comma is the catalyst to a sweaterWhat did he mean by this?
>>8834910There is no way for that poor girl to pretend to be Taihoe on any way
>>8834982That's the joke.
>>8834983Stop bullying womanlets
>>8834986why he mad
>>8834989You ever finna nut after a coke and alcohol binge, anon? Sometimes you just wanna finish but the rest of your body does not.
>>8834989he's focused
>>8834991sorry i'm not european i don't do coke
any hopes of forge couple getting anima support?
>>8835007Why would it need special support?
I'm finding it pretty hard to control breast size on AnimaI would love to see an example of how body shapes are tagged in NL because I feel like with NL it should be easier to control the shape and size of breasts If they've even tagged breast sizes, because I seriously hope it's not just tagged "breasts"
>>8835029If you feel the need to use tags other than large, huge and sagging, you have lost control of your life
>>8835030medium is premium you dog
>bake anima lora (2 hours)>not working all because of scheduler in constant rather than consine.Anyone got a good formula by now? it takes so long where I can literally bake an illus/noob lora in less than 30 mins.
>>8835032>it takes so longmaybe try the finetuning script. finetunes were a lot quicker than loras on sdxl if you had the vram
>>8835030this >>8835031but it's also about shape, I want to be able to define the shape of the breasts but it's just giving me different sizes of sagging breasts despite negging sagging breasts
>>8835033it's anima I need help with not sdxl....also I only have 16gb vram, anima takes only 8 gb with gradient ticked on. no idea what a finetuning script is though.
>>8834974old and busted but it better last me another decade because i'm not buying a new gpu any time soonit can still spit out something small in like 3 minutes, though
>>8835007You don't need it. You can just say "the girl on the left / the girl on the right" or any other image composition thing you want. Camera angles are kind of annoying but relative direction works completely fine.
>>8835029It works the same as normal. flat chest > small breasts > medium breasts > large breasts > huge breasts > gigantic breasts > hyper breastsAnd you can use all the modifiers like sagging, bouncing, hanging, breasts apart, breasts together. It's also based on artist, obviously, and just use negs if you hit some weird place in the latent space.skill issue
>>8835032why would "good" mean "30 minutes?"i never baked anything good that quick on sdxl
>>8835056It's not the same. Nl still bleeds and quickly reaches its limit once you want to separate a lot of concepts
>>8835076well i have noticed some stuff to that effect but i put it off to 512x res
sage and reported
who
>Nearphotison has 15 posts on danbooru>Still works flawlesslyWhere does it come from
>>8835106as long as it's clear enough artist loras can work with that few images so why wouldn't normal model training also work
>>8835106danbooru has dumb rules951 pictures on gelbooru
also>look it uplooks like dogshit
>>8835106it looks awful though
let's also appreciate how fucking awful it is that it's cringe ass text over cringe ass shounenshit characters.i won't kink shame you but i WILL taste shame you
>>8835106so, this is the fabled western anime
>>8835110LoRAs have their entire file size budget dedicated wholly to just one style/concept; checkpoints have to somehow fit the ENTIERTY of Danbooru inside like 4 gigs, which usually means that any tag without at least a good several dozen examples will simply not be learned. And something tells me that dividing 4 gigs by the number of unique tags on Danbooru will land you at a FAR smaller size than a LoRA's typical file size of anywhere between a few dozen and several hundred megs.
>>8835118yeah i was bein retarded
>>8835106https://www.deviantart.com/nearphotison
>>8835121ohoh no
>>88351184 182 218 360 bytes (size of the Anima checkpoint) / 313 235 (number of lines in the 2025/09/19 Danbooru .csv) lands you at just 13.35 kilobytes per tag on average. That's the ballpark of embedding (remember those?) file sizes, not LoRAs. If we make the fairly safe assumption that the vast majority of the tags are below the knowledge threshold, meaning that a small fraction of the top tags takes the majority of the file size, that leaves even less space, and much less at that, than 13.35 kilobytes for any one of the far more numerous smaller tags.
your prompts need to be blessed by a japanese personto not be western promptsthe prompt priest
>>8835127i saw their prompts in the 1.5 daysthese niggas are a superstitious clownshow prayin to the latent space like it's the god of the fuckin mountain
>>8835127clap clap clap
start putting a little japanese flag on your gen to make it japanese :)
I think my finetune config might be chopped
I am trying to prompt two views of the same scene as I want a view from the front to also see the girl's face during sex from behind but it's messing up whether it's turning the girl's head 180 degrees on the back view or completely failing at giving me a front view at all>there are two views defined by an even split in the middle of the image. the view on the right is a pov from the man's perspective as he grab's the girl's ass with his hands sinking deep into the skin of her ass cheeks while fucking her. the view on the left is the same scene but from the front of the girl featuring her expression and reaction to the sex with the man standing behind her as he is fucking her pussy from behind with his hands on her ass, sex from behind, the man's head is head out of frame on the left view, the setting is the same in both instances in a dimly lit hotel room and the girl is leaning over a desk with a desk lamp on it her breasts resting on the deskThis is my shitty attempt, maybe someone else can come up with something better?
unc is typing out his life story and thinks 0.6b can handle it :skull:
let's see your prose then, big guy
>>8835121Jesus I haven't been to DA in like 15+ years and got jumpscared by that layout.
>>8835145Instead of trying to fit everything on the same prompt you could just overlap two different gens together
>>8835153I mean yeah, I could just do two gens but isn't the fun with NL that you experiment with how much you can prompt for?
A year later and here we are just writing...
>>8835145The approach I've been having is to use first an NL description of general aspects of the composition and then do a separate line for booru tags to support it and emphasize individual features (even in a multiple views scenario) and changing one or the other up depending on how the gens do/don't reflect what I'm aiming for.So in this case I would suggest rephrasing most of it into 1-2 sentences and all the small stuff relegated to tags as you normally would do.
>>8835159Yeah I'm currently rewriting the whole thing as it got a bit messy
>>8835154That's a good point yeah, perhaps try to do a list of things you want on the gen just for the lulz
>>8835149i should actually make at least one of the doujin ideas i've had floating around in my head now
>>8835110I don't think that's the case>>8835111So, do they use gelbooru as a database?>>8835112>>8835113kek it is pretty terrible, but it helps with the anatomy if you schedule it out at 0.2>>8835121Let us hope THAT didn't make it into the database lmfao
I condensed it to>sex from behind, vaginal, grabbing another's ass, deep skin,>the man is fucking the girl in the ass penetrating her pussy with his girthy penis while he grabs her ass with his hands sinking deep into her soft ass cheeks,>the image is evenly split in two views of the same scene by a thin black line going vertically through the middle with the view on the right being the pov of the man and the view on the left is from the frontIt just wont work. With this prompt it's giving me the pov sex from behind on the right but the left side is just the girl standing, facing the viewerWhy would the danbooru tags not apply universally
>le multiple views tag
>>8835185you need to reread the wiki my man
adding >multiple viewsdoes nothing>>8835187>the wikiWhat wiki??
>>8835193https://danbooru.donmai.us/wiki_pages/tag_groupstake a look at image comp and camera angle tags and see how they are associated with actual images, then apply that to your prompt
Just list the tags lil bro
>>8835176We know DA was a part of the anima dataset, we know he has 700 pics on DA and 15 on Danbooru. Safe to say it's included.
>>8835194I am obviously aware of the danbooru tags. The ones that would help me would be multiple views and maybe 2koma/comic but neither doSo unless someone else can successfully prompt for what I'm trying to do I think I'll just have to write it off as not possible
>>8835198You're missing a lot of tags for what you want to achieve thoughFor instance, doggystyle and missionary
>>8835198skill issue
>>8835201Half the gen works though, sex from behind is doing its job as the right side is giving me a pov of the girl getting her ass fucked. Meanwhile the left consistently features no sex at all happening>>8835202Post yours :^)
so ive been out of the loop on anima, you cant mix styles? why? isnt that like the whole point of image gen goonery?
>>8835198nta but i felt like trying it, something like this?
>>8835222why is it so blurry?
>>8835218it's just annoying, not impossible
Been out of the loop for a little under 2 years now. What's the best way to prompt two different characters with their own lora? Last I remember the best way was to use regional prompts, is that still the way to go or is there a better way now?
>>8835241most new models have pretty good characters out of the box. the actual best solution is to train a lora with the pair
>>8835241it depends on which 2 characters and what type of interaction you want, it can be as easy a just pure prompt or using several hacks like controlnet and inpaint to hell and back
>>8835245he did say lora characters>>8835241if you're not training them yourself it depends on how they were tagged, they might leak features onto each otherthere's also regional lora scheduling, a little harder to work with and slows down gen time
>>8835224It's NAIma
>>8834967>>8834970doesnt seem too different
>>8835185you're still over-complicating it. remember that it's a 0.6b model and try something like this instead:instant loss, 2koma, multiple views,top panel:\(rosemi lovelock is looking smugly at the viewer, 1girl, solo, rosemi lovelock, nijisanji en, grin, looking at viewer, hand to mouth, mesugaki, outdoors, portrait\)bottom panel:\(1girl, 1boy, mating press, plap, sound effects, boy on top, fat male, deep penetration, vaginal, anus, uncensored, leg lock, impregnation, cum in pussy, hetero, folded, on bed, indoors, bedroom, full body\)
>>8835308>that text and sound effectsyour 16ch vae model, local saars
>>8835308This is exactly how I tag 2koma and have no prob.>quality, artists, # of players, 2koma, general tags not easily conveyed in nl>top panel is yada yada yada>bottom panel is yada yada yadaIt's a lobotomized model so speak to it like a fellow indian anon when describing things. Don't be too verbose.
>got a new GPU>this completely broke reforge because of course it did because lmao cuda problems>had a reforged neo install I didn't bother with the other day because it was requiring cuda 13.0 and I wasn't updating to that and breaking anything, but since I have to use it now anyway whatever>get it installed and running for all of about 10 minutes until I start trying to install extensions>restart UI, EVERYTHING COMPLETELY FUCKING BREAKS>can't figure out how to revert it, deleting extensions does nothing>try and just do a clean pull and restart>now it's throwing an error requiring python 3.13I want to die
>>8835322kekerina of time. I ran into the same shit trying to initially get on Neo. Just a straight up git clone and install with requirements was throwing up errors related to python 3.13 as well. I'm sure I had a separate install of it which I verified so I changed the env paths to reflect it and then it started throwing shit about numpy, some audio engine (I imagine due to the wan support) and some other minor shit. Raged and deleted everything. Do a bit more research and see some shit about Stability Matrix basically doing everything for you and the option to swap ui's on the fly. Cool. Try it. Some python error again. Fuck you. Uninstall everything go to the Neo page and actually read this time. Choose the uv install route. Everything turned out better than expected.Try the uv route, anon.
>>8835326thanks, that actually worked.also, neat, gens take about half as long as they did before. wasn't quite the goal/purpose of the GPU upgrade but still nice I guess.maybe now I'll actually fuck around with anima tomorrow.
>can't pay for Anlas via Paypal anymoreIt's over....
OINK OINK
ixybros.. our genuine response?
bros... any advice for anima gens?
>>8835357Just... write
>try out anima>try to upscale to see if it actually is fucked or if it was just people being retarded and/or a fucked cumrag upscale implementation>upscaling works perfectly finewow it really just needs controlnet tile and for actual trainers that matter to implement code for it in addition to it actually being trained the full way through since this feels pretty rough at the moment.also I'm guessing upscaling issues people were complaining about is just another failed cumrag implementation(who could have possibly seen that coming), since 1.5~x upscaling with a decent amount of denoise works perfectly fine, it just has the issue of not having a controlnet option to reinforce the composition.
you are you all obsessed with upscaling
because i am an ARTIST
>>8835374it's a problem with portrait resolutionhttps://github.com/Haoming02/sd-webui-forge-classic/issues/763
>>8835222Oh yeah, very nice! How did you prompt for that?>>8835308I'll give your style of prompting a try, but the above anon already solved it. Hopefully he can share with us how he did it
>>8835381I am quite literally upscaling images that are taller than they are wide.it's a bit rough around the edges and it has some really obvious resolution driven anatomy issues but it's functional and isn't spazzing out.https://files.catbox.moe/6ubki7.png
>>8835392what happened to her groin :(
>>8835374>>8835392Anon, plenty of us have been upscaling by 1.3-1.5x. I personally haven't tried beyond that myself but the issue seems to be 2x and beyond. Another anon is coping meanwhile with ultimate sd upscale.
>>8835395cope anon also specified the issue starts at a little over 2 megapixel
>>8835357elaborate
>>8835410quality tags/negs.prompt structure. sampler/scheduler/cfg
https://huggingface.co/nightknocker/cosmos-bert>BERT for Anima/Cosmos.>This is not an adapter model, but rather an early replacement for the T5/Qwen model.>This means that the T5, Qwen, and LLM adapter files are about to say goodbye.
>>8835412what is this? custom text encoder for anima?
>>8835412lil bro forgot about ernie
>>8835415never beating the allegations
>>8835411see >>8834948 and replies (i do think no comma is better)prompt structure is the same as before except you can use NL. so think about camera angle, composition, and body shape first, then do whatever.note that the advice on the anima page https://huggingface.co/circlestone-labs/Anima about what order to put stuff before the main prompt is mandatory, shit will break if you don't put things in order. also note that unlike CLIP typos can cause tags to just be ignored
>>8835412so can i just put this in the TE box or what? also>file named model.safetensorsmy retard alert is blinking
>>8835418>file named model.safetensorsthat's just default diffusers naming?
>>8835420yes exactly. it means they didn't even take the 5 seconds it takes to rename their file to what it is.
>>8835383sure thinghttps://files.catbox.moe/zd8fmr.png
you know, highresfix works not so bad on anima
>>8835423awesome
>>8835423now this is some good prompting
>highres fix upscale 2x, 0.3 denoise, no upscaler "none" option>send to img2img>denoise 0.32, multidiffusion tilesize at base resolution, mix of diffusers option, overlap of 512 for sdxl, rescale 2x, random seed, click orange button>open result in krita, downscale 4x image, erase any mistakes using lower resolution image, save new 2x resolution image that looks perfect with no double bellybuttons or weird shit happening in it with more detail that took 2 minutes more to make>posthighres fix works great in a1111 too have been using it forever, i dont get why anyone would use an upscaler alone when they all just take longer for the same or worse result as "none" in sdxl
Thanks for all the help, brosAnima truly is wondrous
>>8835431what the fuck are all these extra steps lmao
https://files.catbox.moe/sgd8nb.png>>8835395only a retard would bother trying to upscale that much though. it doesn't matter what you do it's always going to look like trash unless you either look at it on a phone or you have it zoomed out at half scale. Maybe if the models were trained at 2048px it would be functionally fine but we've seen how that sort of shit goes with illustrious v2.
>>8835431>>8835433"none" upscaler is LanczosMultidiffusion is tiled upscaling
>>8835428It's simple prompting as it should be. I keep saying some of ya niggas getting too shakespearean with this lobotomized text encoder.
is the default save precision of kohya fp32 or why are my finetune safetensors 8gb?
>>8835437i am aware
>>8835436>only a retard would bother trying to upscale that much thoughIt's easier to upscale to absurdres and then downscale it back compared to inpainting little details that get destroyed by 4ch vae. Honestly anima didn't feel like much of an improvement when it comes to intricate outfit details.
>>8835451>Honestly anima didn't feel like much of an improvement when it comes to intricate outfit details.well the positive effect of the 16ch vae while genning is destroyed by the shit upscaling and the positive effect while training was destroyed by the 512x dataset.
>>8835451i disagree, besides the obvious degradation that training at 512x made, i've done a bit of testing on highly specific concept loras, and the ability to pick up details is greatly improved, better than sdxl after months of testing and dataset curation. now some of this could just be simply that anima's dataset is superior but i don't think so.
>>8835451what details. to upscale to resolutions that large you have to nerf denoise to miniscule levels and all you get is nearly exactly what you started with at low resolution but fuzzier and with a lot of halo/glowing between linework. You're not introducing or improving any details over what you had at the start.
>>8835454>nerf denoisetile controlnet
>>8835455does not yet exist for animahave we strayed into the theoretical zone
>>8835455if you're setting controlnet tile to .9+ weight and 1 end step you're not really doing much with it.
>>8835457personally i like the weight fairly high but drop it off at like .6 or somethin
How do you guys manage 1.5x upscales on anima? It just artifacts for me no matter the settings
why do you want to upscale? what are you going to do with the image?
>>8835484>what are you going to do with the image?fap
>>8835485why do you need to ultra zoom in on your gen to fap
>>8835486why wouldn't I?
>>8835483Just read the thread
You upscalefags just wouldn't get the Dell E773c 17" Monitor anon.
Is being anti-upscaling the new contrarian brainlet take after anti-lora, anti-controlnet and anti-regional prompter?
>>8835501I think so, everything for the sake of those sweet (You)s I guess
>>8835501Upscaling isn't free - it costs time. On SDXL that wasn't much of a problem since it's so fast AND upscaling was a massive improvement over no upscaling because of the compressed VAE, but on Anima that's both slower and much better at details out of the box, its relatively smaller benefits are much harder to justify when juxtaposed against its much harsher downsides. You're basically paying more to get less benefit.
In order for the model to give good outputs right off the bat without upscaling/detailing, it has to be a lot bigger. Z also needs upscaling while Qwen Image can give nice results right away
Anima actually murders details because of the 512x training. You can see this pretty easily by prompting any character with clothes writing and comparing with what NAI 4.5 would give you.
>>8835509That's only temporary, it's ought to get much better once it's trained on 1024p. Still, even the way that it is, it simply artifacts less and produces much more coherent hands when they're far away from the camera.
>>8835512That's the bare minimum one would expect from a 16ch vae model, but sure.
>>8835512>once it's trained on 1024pETA never ever
>>8835512Were there any updates on this after the preview release? Any info on the progress?
>>8835517the only thing he's said is that it'll be trained for a long time at 1024px
>>8835518i will wait warmly
https://files.catbox.moe/2d4xkm.pnganima is kind of neat in dumb ways, I suppose.
>>8835543kino
drawfag lora when
>>8835604he just like me frfr
>>8835604Why is a Ukrainian pulling Burnice's panties?
>>8835604Based Napoleon enjoyer.
sirs?
>>8835607hehe
>>8835625Yes?
>>8835638good morning
>>8835641it's 5 fucking pm and I am dying of heat wtf is wrong with you chat
>>8835642do not redeem the upscale
help
>>8835666what with
>>8835667not upscaling
any new fast anima lora presets working? tired of baking overnight only to see it never works well.
>>8835667gooning
>>8835670actually on that note my first two bakes worked fine it's just in an "meh, ok" state, which took ~2hours per full complete bake at 3500 steps, but will changing from 1024x1024 to 512x512 help with the quality or does nothing?
>>8835672i only got worse results when i tried 512x512 but maybe that's just me. it's also much faster so you might as well try it out if you're curiousi've also only gotten results at the "meh, ok" level so far, although there are definitely things it does much better than sdxl models
>>8835670i just copy pasted my settings and it went from ~1:50 to ~2:30, results mostly goodnot gonna spend a ton of time baking since it's just a preview model, i still have a massive backlog of datasets to prep
>>8835671elaborate
>>8835677hentai
>>8835678like what
>>8835679girl
>I can taste... The essence of your suffering - it is sweeeeeet.
>>8835680
>>8835692thanks
>>8835692>nipple piercing based
>>8835692You never had the makings of a oji-club member.
>>8835699i respect women irlit's only in the world of imagination that i rape and enslave them
>>8835702Is that one of the roach men from Terra Formars?
>>8835702>i respect women irlshiggy diggy
>>8835736pretty much yeah
can you make the male younger
>>8835828chat, is lil bro the new meta?
>>8835828Cringe.
can you make the female younger
>>8835819cute
>>8835831her age?
>>883583425
>>8835835too old
>>8835836I liked your performance in titanic
>>8835836unc, current meta is that pubescence ends at 32
>>8835840starts*
>>8835841soon
13 and down only
blud is in the files :skull:
>>8835848at least he's probably not a cannibal
WHOMST?
/hgg/ - pedophile mossad billionaires general
>>8835851can you make her nakeder?
>>8835851What if she spills hot coffee on his penis?
>>8835853Why?
>>8835853It would be extremely painful
>>8835855>>8835855>>8835855>>8835855
I'm looking for artist working on noob with similar way of drawing eyes like https://www.pixiv.net/en/artworks/120255517 , more narrow and "almond shaped" without being aco. Any ideas?
>>8835865give the tag sanpaku a try
>catbox keeps killing itself I hate it
>>8835604box please
https://files.catbox.moe/7xar1e.png
>>8834948>>8835417what sampler scheduler cfg are you guys liking on anima
>>8835874Euler a, Normal, 4.
>>8835874I'll have a cfg 5, euler normal, keep the A
>>88358743m sde sgm uniform 50 steps
>what cfg?
>>8835879my classifiers are free, imperial. can you say the same?
>>8835879between 4 and 5 same as it's been since forever
>>8835879max that bitch out
I miss all of the avatarfags that used to inhabit /h/ ai threads.
>>8835889avatarfags never stick aroundthey are fake
>>8835874>DPM++ 2M SDE, Simple, 33 steps, 4.5 cfgLet me guess, you need more?
lil' cfg on the avatar island
>>8835870Thanks, didn't know it's a thing. Better than nothing, but kinda doesn't fit generic styles.
>>8835889like who
>>8835910miqogod
>>8835910General's p boring without the goddess of felines tbqhngl
I see, my bad for asking on good faith>>8835907>>8835920
>hit bump limit with only 35 imagescoming from degen and slop, this confuses me
>>8835933Well, now you know what to do
>>8835933fuck off newfag
Does anyone know if kohya trains the llm adapter by default? or do i have to specifically set --llm_adapter_lr to zero
benis in bagina :D
penis in pussy
meatstick in mouthpussy
>>8835978incredibly based but you'll sadly get vacationed for it
What's the best Anima shitmix atm????
waiting for ntrmix
>>8836006we're not doing this again. Use base or perish
Asuka doesn't really care about Shinji's dick
>>8836008>Use base or perishYeah, because everybody used base illu when the first decent shitmixes started appearing, right?
>>8836012Everybody with white skin sure did. Also let's not pretend base anima is nearly as hard to wrestle as base illu was.
>>8836012Unlike with SDXL models and their CLIP TE that simply binds image patterns to trigger words and does nothing else, shitmixes of any model that uses an actual LLM as a TE actively destroy its knowledge and NL comprehension by diluting them with uncaptioned image slop. Your only real option is to finetune on a dataset that is as well-tagged and described as the training data.
>>8836014>dataset that is as well-tagged and described as the training data.>score_9, score_8>well tagged>implyingwhew bro
>>8836012Yeah. 1-2 loras was enough to deal with Illu's issues before noob became good. Also the early Illustrious shitmixes like SPO (or whatever advokat mix was called) weren't offensively slopped.
>>8836019>1-2 loras was enoughAlright then I'll change my question.What are the best loras for Anima atm??
>>8836017That's still infinitely better than blindly merging with existing checkpoints and LoRAs that were made on, at best, images captioned with just tags alone without a hint of NL if not straight up completely uncaptioned images.
>>8836021GoPro
if a lora adheres to a pose too much, does it mean overbaked? I tried making a unet-only lora for anima and that feels like the case, and does tag hierarchy affect the baking?
>>8836029>if a lora adheres to a pose too much, does it mean overbaked?yes>and does tag hierarchy affect the baking?yes, you should shuffle captions and probably use some caption dropout
>>8836031wouldn't shuffle kinda mix tags? also I set it at 0.00007 including something called betas": "(0.9, 0.995)", no idea what it does but I tried faithfully copying a lora setting.
>>8836033shuffle is good because you could unintentionally be creating a bias based on the order of tags. same reason to do caption dropout, you could be unintentionally be associating training tags with each other in an undesirable way. i like very high dropout, 0.5
hi :3
so glad this thread doesnt have vidgens
>>8836061yeah this thread is for white people
>>8836061only because my workflow broke and desu the models are dogshit
>>8836010he got to fug I don't think it matters to him
>anima can't do ranceI revise my opinion, it's a dogshit model
hi good morning :3
pussy pussy face lora when
cfg 12??
updated arenahttps://arena.sharkgone.com/
>>8836118why is there so many mahiro gifs there
>>8836120trooncord fag website
>>8836120yoinked is a tranny
>>8836118wtf is randc2?
>>8836125idk, looks like some shiny merge
>>8836118looks like it was just reset, lmao
>>8836089Illu and Noob can't do Rance either. There's a grand total of 9 "rance, solo" images on danbooru. This character is simply too horny, almost every image with him you can barely see him because he's mostly offscreen fucking some girl.
>>8836136You know what model can do Rance? NovelAI 4.5.
>>8835865Dotsuco
>>8836142thanks
>>8836157Did it work for you? thxzmgn is great alsoI swear niggas will do everything but recommend artists to youCuriously the thread also has very few gens so I'm inclined to believe everyone else has bad taste
>>8836158The eyes shape is kinda there. Will try mixing with something. Chink artists have sometime quite interesting styles, but like with the linked one, they are very inconsistent and low post count, so making lora from like 10 pics would be shit.
>>8836163I don't think they're very inconsistent but it's true that they have low post countI've come to find that chiggers completely mog the japanese when it comes to art and then they use it to make gay shitSomething about their use of colors is fantastic
>>8836166Yeah, I agree. When I see some interesting style on pixiv it's usually chink or gook with 10 images.Japs seem stuck and westerners either have no self awareness of their low skills or are "original" by drawing lips and noses.
>>8836169I like lips and noses, westerners just make it look fucking awful because they generally have a grotesque sense of "muh realism" which isn't even realistic, it's just ugly. Chinks will oftentimes do very realistic styles and still make them look great. So when it mixes with anime it also looks fantastic, yuming li comes to mind
>>8836170The only style with lips and noses I like is solipsist.
>>8836171>look it uplooks nice
how do we fix our sister thread?
>>8836180There is a good lora up there https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLg
>>8836187sorry i am animapilled now
>>8836169or it's a scat artist like Gemuo
>>8836183lil bro is cooked since the heiress of felines departed
any /g/ger to explain what this is?https://huggingface.co/nightknocker/cosmos-bert
>>8836278We already discussed it, Ernie. Read the thread.
>>8836288you just took a shit in replies, that's not really a discussionhow do i use this?
Oh very mature, Rajeesh.
>>8836288You mean >>8835412? Nobody gave a concrete answer on what this is or what it does.
>>8836295That was the concrete answer, it's an alternate text encoder. It's also bad, which is why it wasn't mentioned since.
>>8836306i forgot to test it but like 99% it's bad yeah
I think Anima is alright.https://litter.catbox.moe/qjshhqor1h3pnupw.png
>>8836321what the fuck is that metadata lmao
>>8836323its comfy
>>8836323First time encountering comfyhell?
>>8836325no, it has a bunch of weird shit in there like a schizoneg with a few memes in it, the N bomb at a random place, literally 10 fucking loras applies...
>>8836327>he's not using slurs when genningsoulless
when are we getting e621 anima tune from furfags?
>>8836347Probably after the full version is released, and when anons figure out natural language tagging. So never lol
post anima cunny
I'm surprised there's no rule against goading others into breaking rules. Pretty big deal in actual law.
goy alert
cunnybox exists for a reason retard
which one you'd like, anon?
>>8836370https://files.catbox.moe/sgvv6h.png
>>8836401>risahttps://litter.catbox.moe/qwvhnk5gwcl35a6r.jpg
>>8836407it is what it is
>>8836401that is not cute nor funny
>>8836410really? what is then
>>8836411plain boring cunny, make something cute and lewd now
>>8836401can you make the heads bigger?
>>8836414nah i think i'll post my degenerate fetish version instead
>>8835853giwtwm
>>8836014are there any docs on tools anima used to caption? like qwen 8b VL or smth?
>>8836118what are the names of the full models...
On a Anima lora at 2800 steps, how much caption dropout should I add?
>>8836427That's the million-dollar question because all the NL captioning models are censorslopped.
>>8836424What's with noob giving characters ridiculously long necks when prompting for deepthroat
>>8836435it's anima
>>8836432i like 0.5 dropout
>>>/b/946265031
>>8836439based
>>8836439cringe
>>88364370.5 like in the "full dropout rate" settings? wouldn't the tags be completely gone after a half steps in? well I don't really know how this shit works anyways.
>>8836449literally just ask chatgpt lol
>>8836432Zero, unless it's a character and you're too lazy to prompt all her clothes.The model was trained long enough with dropout to understand it can add things to the image without explicitly being prompted to do it. Like how a girl has nipples even if you just imply she's topless. With a Lora you just worry about the one thing you're training it for.>>8836449iirc 0.5 dropout means every time the image is processed it will be missing half of its tags, selected at random. Seems extremely high, but I have not tried it above 0.1.
next anima version when
>>88364732 more weeks
>>8836473maybe in 6 months if we're lucky
>>8836449it doesn't stack...it's "for every image, drop out 50% of the tags that aren't kept at random"
>>8836449caption dropout not full dropout that's cooked idk why anyone would use that perhaps someone can enlighten me
>>8836473by 2027 for sure
>>8836473who knowsthats why i just subbed to nai
>>8836480same
what's the anima trainer you use?https://github.com/Moeblack/AnimaLoraToolkithttps://github.com/FHfanshu/Anima_Trainerive had enough of sd scripts, maybe i just roll one with diffusers or smth
>>8836486that fork of ezscripts
>>8836486why would you willingly use chinkware
anyone have a dataset for anima i can test a config on?
>try to upscale using anima in img2img>get weird artifactsThe fuck? SD upscale works without artifacts but it makes way more mistakes.
>>8836507known issue, that is all we have atm
>>8836507Very sexo outfit
>>8836507issue caused by 512x
>>8836507Because it's genning in small tiles but still sees the full prompt, without tile controlnet. You can somewhat negate it by making your tiles larger, default size is 512 from SD1.5 days but the artifacts only start ways past 1024. Also helps to keep denoise lower.
>>8836530It's basic upscaling, not tiled. Tiled upscaling is the one that keeps making dumb mistakes (dicks growing out of necks, dicks between thighs, shitty hands, etc).
>>8836507Increase tile size to 1024 and tile overlap to 128, also add "out of frame" to the prompt.
>>88365331024 is my default tile size and I already tried the other tricks but even with them and at low denoise it just sucks and tries to shove every part of the prompt into every tile.
>>8836531I was referring to those mistakes, you probably have your tile size set very low. Then the model is working on a 512 square area of her neck too small to see the overall context, and it still has "penis" in the prompt.
>>8836534That's just how it is. Best thing we have, for now.In theory you could do the tiling yourself, and update the prompt manually to only what's currently visible. It's just a lot of manual work, especially for the seam fix pass.
>it aint gonna lick itself, lap it up like a good boy
>>8836534Are you trying to do a 2x upscale? That just doesn't work, the most I managed to get out of anima is 1.5
>>8836538sister, this isn't /slop/
>>8836539Yeah, 1.5x seems a lot better. Btw what's the deal with MultiDiffusion Integrated? Isn't it a similar sort of thing? Doesn't seem to get any progress when I try it, does one step and then just hangs. Doesn't crash but never makes any progress either.
>>8836537Or just inpaint chunks of the image manually. Less work and you can leave out the background and things that already have good details.>>8836539We're talking about tiled upscaling, as a workaround to that lets you go higher.
pretrained_model_name_or_path = "E:/sd-webui-forge-neo21/models/Stable-diffusion/anima-preview.safetensors"qwen3 = "E:/sd-webui-forge-neo21/models/text_encoder/qwen_3_06b_base.safetensors"vae = "E:/sd-webui-forge-neo21/models/VAE/qwen_image_vae.safetensors"output_dir = "E:/sd-scripts/output"output_name = "anima_finetune"save_model_as = "safetensors"save_precision = "bf16"optimizer_type = "adafactor"optimizer_args = ["scale_parameter=False", "relative_step=False", "warmup_init=False"]learning_rate = 3e-6lr_scheduler = "cosine"lr_warmup_ratio = 0.1loss_type = "l2"timestep_sampling = "sigmoid"sigmoid_scale = 0.35discrete_flow_shift = 1.0min_snr_gamma = 5.0max_train_epochs = 20train_batch_size = 1gradient_accumulation_steps = 2seed = 1337cache_latents = falseresolution = "1024,1024"enable_bucket = truemin_bucket_reso = 512max_bucket_reso = 1536bucket_reso_steps = 64random_crop = falsemax_data_loader_n_workers = 2persistent_data_loader_workers = truellm_adapter_lr = 0mixed_precision = "bf16"gradient_checkpointing = truesave_every_n_epochs = 1save_last_n_epochs = 5sample_every_n_epochs = 1sample_sampler = "euler"sample_prompts = "E:/sd-scripts/config/prompt.txt"attn_mode = "sdpa"my current finetune config. It's getting there imho
>>8836543>We're talking about tiled upscaling, as a workaround to that lets you go higher.But it really doesn't, that's the issue.
>>8836549t bh I've only tried it once but it worked, for a 2x. Didn't see much need to upscale that high on anima, most I just inpaint eyes. I'll give it a few more tries with a more complex scene. And maybe use Multidiffusion instead of ultimate SD, if it's been updated.
really hope they don't add any e621 to the full bake. 4.5 already covers the degenerate needs of mine, need an updated aesthetic-looking model now
>>8836546grok is telling me CAME 5e-6 to 1e-5 is the meta optimizer, leaner than adamw8bit better than adafactor
>>8836552pretty sure kohya doesn't have came. don't know if there's some external repo that adds it.
>>8836551i mean that already went out of the window with the deviantart and ye-pop datasets
>>8836553the ezscripts fork has it
>>8836554Which didn't seem to negatively affect any danbooru artists, only the quality tags.
>>8836554doesn't feel like the effect is too detrimental for now. but yeah I don't get what's even the purpose of deviantart
>>8836557Nor did the furry dataset on the noob bake, what's even your point
>>8836558/aco/ artists I assume. If you're re-tagging everything anyways, maybe it has more value than r34 which is only porn.
>>8836559noob training did so many things wrong that it's impossible to say exactly what did the most damage, but nai 4.5 is a better example of why e621 won't poison your model if you know what you're doing
>>8836561>nai 4.5 is a better example of why e621 won't poison your model if you know what you're doingit's the ugliest model since pony lol
>>8836562i also hate that 4.5 is still anime nsfw sota but there's no point in being delusional
>>8836565*sota for nsfwhad a stroke writing this apparently
>>8836565you're overdosing on copium my guy
i would like them to add the hentai-foundry dataset for extra sexiness
>>8836568you might need to renew your subs and run some prompts without a stack of quality tags and vibe transfer. the base model is ugly as sin and e621 has no small part in that
i want them to add cosplay hoes and knowyourmeme
>>8836568i use it almost every day anon. you have to actively fight it so it doesn't look like shit. it's very good with getting concepts, styles and characters from a very low amount of tags but its visual quality is just straight up bad by default. I still wish there was updated v3, it's still the best-looking nsfw-capable anime model in my opinion (noob with good loras is a close second though).
>>8836570it looks perfectly fine with artist tags and without quality tags, what are you even talking about>>8836573but i do agree that v3 is somehow the best looking anime nsfw model to this day
what's up with all the naiggers
>>8836552but is it better than adamw
>>8836580maybe yes https://arxiv.org/pdf/2307.02047but poorfags have to scrimp on memory bytes more than convergence
https://litter.catbox.moe/lal1e9o53kie8eci.pngThis image will spread like wildfire among browns and cause another model gen of shiny slop
>>8836581i read the 1st 3 sections and the part about concerns at the end. looks pretty good. but this is mainly applicable for proper finetuning it seems. as long as it's using the same basic logic as adam (which it is) then i agree. the other takeaway is just don't use adafactor because it's like a fuckin turbo lora that rapes your accuracy for speed
>>8836589just cross out the score 9 and comfy parts and replace the gay guys with cute girls its fine
>>8836590adafactor is chosen because it has minimal memory impact for 8-16GB poorfags
>>8836592i mean, technically, but it's more because it was a previous attempt to solve the same thing the paper writers are trying to solve
>>8836592So does came work for full finetuning anima with 24gb? My adafactor config was sitting at 23,8gb already
>>8836595it's supposed to be slightly better but only you could know how it would work on your hardware
>>8836598Another day another venv it is
>>8836589The average shitmix slopper doesn't gen locally, until civitai adds it to their roster of featured models we are safe.
>>8836118what are they?
>>8836424>>8836435>>8836436Copy that
>>8836632They're usually non-ironically filtered by a local installation lmao
Man not being able to mix artists properly is grinding my nutsI've seen some niggers talking about prompt scheduling, but at what value? Did anyone find a good, satisfying way to mix artists consistently?
>>8836706it's not gonna happen, stop hurting yourselftrain a lora or stick to sdxl for now
>>8836707>train a lora or stick to sdxl for nowCan't train a lora and sticking to sdxl would hurt moreThere's not even like general rules? I tried to hunt for patterns but it just seems inconsistent
>>8836709its just rng
>>8836709no model will ever be able to mix artists as effortlessly as CLIP-based modelsthe sooner you accept this the better
>>8836711What, really? Why?
>>8836712because the only reason model mixing worked nicely on 1.5 and sdxl models was because of CLIP tag bleeding. that wasn't intended but it was a happy accident for our purposesT5 and LLMs are much less retarded and don't bleed tags, but they have no consistent way of handling multiple style prompts
>>8836714So, in essence, it was good because it was shit?Man, that's a blackpill. I love anima a lot and it's way better but having to use a shit ton of loras or being stuck with single artist tags is really shitty.
>>8836719Gen first in Anima to get the composition and concepts right, I2I in SDXL to apply style?
>>8836719it was the one and only advantage CLIP had. i'm glad that we don't need to put up with its bullshit any longer but it definitely sucks to lose the easy artist mixing
>>8836722So anima use is just "easier" koikatsu? lmao
>>8836722I can use a bunch of things for that if I wanted, and if my hardware was good to do that in a short time I'd be making loras anyway>>8836723Shame, I was coping this would be a preview problem
How tf do I describe noses like Annie's from Attack on Titan?Aquiline nose doesn't work anymore for some reason
>>8836732add jew to the prompt
>>8836551oink oink piggy
Anima feels so much less consistent than noob shitmixes...
>>8836760at what? 1girl pov cowgirl?
>>8836763why would you even use anima for that
>>8836764that's all noob shitmixes are able to do
>>8836766if you are retarded, yeah
>>8836430>>8836637some aliases are super obvious alreadynv10 - noob vpred 1.0anima - animane11 - noob eps 1.1il10 - illustrious 1.0ch02 - chenkin 0.2rf03(a/b) - noob rf 0.3 (aesthetic/base)ckrf - chenkin rfrw - rouweilooks like sh, randc(1/2), karm, md, chry, cyber, nyl3 are shitmixes
>>8836793duh, what are randc
>>8836760Isn't not thinking about how to mix artist tags the point of a shitmix? The first few versions of any model are at least mildly inconsistent with that.
>>8836805The problem here is that any Anima shitmix that isn't an actual finetune on properly tagged AND NL-described data will actively degrade its capabilities.
>>8836807Oh yeah, I agree. I'm just waiting to see if the full version evens out that particular problem. Are there any NL description taggers even worth it as a starting point? I think there's going to be a huge skill check just because most of us have forgotten how to write a sentence (even a run-on one, lol)
>>8836793I still have no idea why has nobody trained a full finetune of illu2.0 besides a couple of terrible shitmixes. Was it that bad? Overtrained?
>>8836813Have you tried it? It's free.They started training at 1536 res which destabilized anatomy, then they added cosplay images to try and fix it. Plus it only has Danbooru knowledge on the level of early noob eps, and no e621.
>>8836111https://civitai.com/models/2411422/do-you-want-to-pet-my-catit's severely underbaked, but I got a decent result by using prompt scheduling then img2img-ing it
cunny
>>8836824you can kinda just write it
https://files.catbox.moe/zfpmyg.pngalso this is a really stupid fucking meme and I hate that there are tons of retards posting this shit to xitter without marking it as sensitive/explicit. That they're also typing in "sfw" on their tweet is even more retarded, though that part doesn't really matter.
>>88368132.0 is absolute dogshit
>>8836858thank you for your insight on the current situation
>>8836861np be sure to yike and crabscrub for more content.also if anyone actually wants it here's a shitty 5.6mb lowstep lora for the general composition trained on vpredhttps://files.catbox.moe/z4vjmy.safetensorsyou'll need to scribble the cat face on yourself but you can do that in 5 seconds in mspaint so who the fuck cares. it'll mostly carry through if you scribble over the lowres before upscaling.for context of use here's the metadata for >>8836858<https://files.catbox.moe/pq4fcj.png>generally you'll want to tag out >pussy peek, clitoris, tableit was tagged with jz235 in training but since it didn't want to actually catch the drawn on face to that tag it's probably not actually necessary to use in practice and you can just ignore it.
Whats the current situation of using generated images in final products? I know Illustrious has some license that says you can sell images made with it, but how is it for other baseline models?
>>8836863head expansion lora when
>>8836858can you make the head bigger?
>>8836813angel's galaxy mind decided to add a lot of ai images around 2.0 and you can definitely feel it
>pussy on table memethis is like one of those things you'll gen once and never again
What tags to try for a cavewoman outfit. Loincloth + fur shirt isn't doing it, but searching through the Danbooru wiki I can't find much better
>>8836925Try mixing fur trim along with just using a tribal tag. That will push it more in the direction you want.
>>8836793
>>8836925cavewoman, revealing clothes, fur \(clothing\), single strap, tribal
>>8836824Oh hey that's not too bad, it gets the job done
>>8836934Meant to reply to >>8836863 since I used that instead of the one on civitai
>>8836925I just use animal_print with tribal_dress, which isn't even a tag but works fine. You can also split it up into loincloth+bandeau, because the "tribal" tag adds bodypaint usually so you might not want it.
>>8836924this is an absolutely retarded meme
>>8836944its twitter
>>8836945it's the same picture
Remember when we had like 10 jackopose loras?
you can use the wan2.1 vae for anima btw since qwen is just a decoder finetuneqwens is more smoothed out which i dont like
>>8836953wonderful another thing to test
>>8836953wonderful another placebo to ignore
>>8836944why you hate anime
>>8836965i love anime
>>8836968piece of evidence to sustain your claim?
>>8836969will an absolutely mummified meme i saved in 2009 suffice https://litter.catbox.moe/a7m61hc9nrqi5opt.jpg
>>8836953i ran some x/ys and it's like 99.9% identical
>>8836984fell for it again award
>>8836753Doesn't work it gave me a knot for a nose, unironically
>>8836986what? it would be fell-for-it-again award if i blindly did it without testing
speaking of which>try out TEs>normal one is fine>that heretic one flattened out all the colors and made everything more blue and green, i.e., absolutely raping the artist replication for who i was promptinglol ok
>>8836991forgot to mention the cosmos-bert just made 4 black images lole
>>8837007cute lowres freng
>>8836732I think you'd need a lora for that, or find some niche artist who draws noses like this often (which I doubt you'd fine). I guess you can also just gen a normal pic and then use flux klein or qwen edit to alter the nose with a pic of Annie as the 2nd input
>>8837010>Worked fine in the inferior model>Doesn't work in the superior modelOh well, I suppose I'll wait to make my Jewish and Greek characters after the final model releases
>>8837011It works fine in the superior model (NovelAI 4.5)
>>8837014Pay it for me then
https://files.catbox.moe/zr52uu.png
>>8837034why her fingers so long
>>8837035distorted perspective. it's (kind of) intentional.<https://files.catbox.moe/2hbmon.png> this was the original planned composition but I didn't feel like fixing her thumb and went with a different one.
>>8837036fingers don't look like that from that angle tho
>>8837037stylization is a thing that exists and doesn't necessarily conform 100% accurately to anatomy. sometimes it breaks anatomy on purpose, even.
>>8837039ah yes the be retarded -> get called out -> double down pipeline
>>8837039it looks pretty bad in this case
>>8837039>it's my style!Welcome back, 2006 DeviantArt
head bigger status?
>>8837049i like hebesif they got curves its good (even if they are small)if they dont look like a woman its bad
Full anima when?https://files.catbox.moe/xso85r.jpg
https://files.catbox.moe/xc48ce.png
better post some gay shit in catbox, that'll show 'em
you are not a pedo, right chat? you just like to look at cute and funny things right?>https://files.catbox.moe/gs9a9i.png
>>88370593dpd has pd at the end for a reason.
>>8837061monkaGiga
i need the head bigger
>>8837059uoh
samefag
>>8837059i fucking hate breasts on lolis
Does anyone have an updated csv made made with Anima in mind? Like the training date and all
>>8837082dumbass
>>8837084chests should be perfectly flat, retard
>>8837085nyo
>>8837085"No"
>>8837052hot, post more
>>8837085based
epstein ahh posts
Genitals situation on anima is pretty bad
>completely fine and normal genitals>NOOOO we have to add 13 billion farting on prey images so the model learns the right bend of dick veins
the scat thread is next door
what made him so mad again?
>>8837129i say this
>>8837123Just.. inpaint? or did you already did that?
>>8837142Yeah, that was the least bad out of a dozen attempts.
>>8837123a lot of it is just upscaling being really mediocre due to it being a 512px train and no controlnet options.hands/fingers get spindly and fucked up pretty often.
https://litter.catbox.moe/fw6on9nqhbmzzmac.png>from semen demon to bobbleheadgrim
we love funkopops here
lolicon died, and bobbleheads killed it
>Your petty quarrels only make me stronger>SUFFER!!!!
>>8837088uwuhttps://files.catbox.moe/uljzas.jpg>>8837123well yes everything on anima is pretty bad since it isn't fully trained, except the aesthetics, heavily moggs any sdxl model on aesthetics
>>8837173left: sexright: sex
no
i liek this one https://civitai.com/models/2300210/2d-gold-fish-or-high-res-anime-xl?modelVersionId=2588259
>>8837209wow wow wow (yes this model is so butiful it deserves three wows!)
>>8837209this guy used to post here
>>8837034welcome back, been trying to find you somehow, have you tried anima yet? i'm trying to find a decent formula for it and i've been using your noob/eps ever since I got it.
>>8837220We only acknowledge chadmixanon, panchito, and 291h mixer here.
>>8837252for me it's Mipmup, the Pillarest
>>8837209Sorry I only care about anima shitmixes now
what do you even mix anima with
>find some decent-ish artists that anima can imitate>anatomy goes out of the window>have to spend hours upon hours mixing them up>or I'd have to mix in another artist that looks like utter sd1.5 slop but miraculously fixes all the issuesSurely the final version won't require some "stabilizer" artists/loras for good outputs, right?
pedo bros... atf is shutting down.... it's over
>>8837346the only good thing i ever got from there is a nude mod for yuffie in ff7 remake
>>8837173qrd? never watched/read takagi-san but the left one is indeed 100x more appealing than the bobblehead on the right
oekakibro pioneered a new era for cunny
>>8837346>>8837357Really? Now I'll really never play Onirism.
why is nobody rotoscoping here
>>8837370We only gen though civit here and buzz is hard to come by, sirs
>>8837370it's kinda bad, I haven't made anything worth posting
mmmmh which im@s should I gen getting harapan'd today
>>8837361>qrdjust the difference between the first and the last volumefor better or worse he's doing this deliberately, because for example these are from the same period https://danbooru.donmai.us/posts/7894022 https://danbooru.donmai.us/posts/7746095
>>8837380she literally lost some curves there. from right to left would make sense
>>8837361Maybe he doesn't want her to be appealing that way, seeing as she's underage
>>8837383lol
>underage appealing by default
her age?
out of 10!
why do you care?
We respect the age of consent here.
>>8837383>sees fully clothed young girl>talks about "appealing that way"i know what you are
ayo. chat, what lil blud implying?
when can my pixels become of age?
>>8837396set your cfg to 18
Why is she so red
she's an adult
post white females
with big craniums
kys hydrocephalusschizo
we like prepubescent pixels here
>>8837390>>8837408Which one is lying, chat?
>i love lolis>i hate huge headssimple as
>xhe doesn't use the forehead tagExplain yourself without sounding schizo.
love bangs
i use cum on forehead tag
>>8837412That's what I meant by >>8837383Maybe he got weirded out by pedos and so made her more chibi-like.
>>8837417Most likely it's this. Didn't want diddy ahh weirdo creeps jackin it to her. Japan doesn't tolerate this shit unlike degen west
>>8837417>weirded out by pedos>posts very suggestive pictures of schoolgirls on twitter
Not a good look, chat
gaijin chads own the lolicon
>>8837414for me its oliet bangs
discuss models
lolima when
why don't you go to /hdg/ if underaged pixels scare you
Listen to me for a sec, brosloliswith BIG fuckin tiddies
>>8837433kill yourself fucking subhuman faggot
uh oh pedo melty
For me? It's that lollipop energy.
nature is healing
big boobs = redditbig asses = twitterbig heads = 4chan
lolis with bubble butts
guys stop shitposting youre supposed to be complaining about upscaling on anima and asking when the full release is coming
why you still here, lolihater? shouldn't you post your hourly tirade on twitter?
Man the 60 image limit on joy caption steals all my joy
>>8837442>upscaling on animaBest I can do is two consecutive low-denoise Ultimate SD upscalesAnd sneedvr on top of that I suppose
Why is the shitposting much more lighthearted here than /hdg/ bros?
>nearly 700 posts and only 55 imageswe finally did it. I'm so proud of you boys.
post nai3 gens
lolis with bbcs
>>>/hdg/
this, we only gen hilichurls here
not a fan of poop *shrug*
>when the anima artist mixing prompt scheduling hits just rightWe can't keep getting away with not depositing 25 a month, chat.
year of the honse
>when [some shit i made up]
>>8837452proofs?
>>8837452what about ethereals?
I guess /hdg/ was too dead, so the schizo retards came here instead
gonna upscale your tears? lmao
dunning kruger sisters, our response?
whomstve lil blud bwehin aboot
Can I use Anima on WebUI yet? Comfy makes me too unreasonably mad to do anything but troubleshoot the entire time it's open.
>>8837492Yes, Forge Neo added Anima support ages ago.
>>8837493Oh thank fuck. Comfy is hell. I can understand why some people use it, but I'm not doing anything nearly complex enough to justify it.
i would like some sloppy cunny please milords
>>8837494this is all you need for a basic gen, i really dont understand whats so fucking complicated about it
>>8837501Ok, then how do I upscale it latently at a 0.57 denoise and a 1.5x resolution upscale?
>>8837502for anima? you don't, lol
>>8837505...Yeah I just tried doing that on Webui right after I posted, lol.
Anima status? I am not messing with that shit until the actual full model is out.
>>8837502Anima preview was barely trained on 1024x resolution, going above that is just asking for trouble. Tiled upscale is also not possible without controlnet, unless you go for super low denoise
>>8837514Seems like it'll be a step forward, so far it kinda "Just works", as far as early models go.
she eatin the poopie :(
>>8837518so, this is the power of /hgg/
>>8837518whats she doing to him?
>>8836118release randc2
>>8837469They just don't have the same appeal as good old 'churls
mmm poopie >:)
umas x hilichurls
>/hdg/BBC>/hgg/BHC
theres many flavors of feces
feces?
kek
https://civitai.com/models/2420824/thank god, anima is saved
im going to tinker with anima style training at 1536px, low LR, batch size 1 & no gradient accumulationthe model already knows artists pretty reasonably, even if they have only a couple dozen samples on danbooru. high res training and getting more high frequency detail might be more value added
>>8837573>1536pxThis is retarded. The model wasn't even trained at 1024x yet.
>>8837575that's not entirely correct```The preview model isn't that good at higher resolutions yet. It is a medium-resolution intermediate checkpoint, trained on a small amount of high-res images. The final version will have been trained on a dedicated high-res phase. Details and overall image composition will improve.```besides, if you dont fuck around youll never find out
>>8837576you still want to gradually up the resolutions during training, not blast a model that barely had any 1024x training with 1536x. I'm pretty sure that some models even went from 512x to 768x first
>>8837577sounds like some goddamn commie gobbledygook
>>8837575it's possible to boost a model ability to handle higher resolutions with trained lorase.g : https://huggingface.co/nblight/noob-ft/blob/main/noob-ft-1536x-extract.safetensors for noobI don't know how that guy did it, but it makes noob work at 1536x much better than Illustrious 2, the other sdxl model claiming to be able to handle this
>>8837370it's not actual rotoscoping it just looks like it
>>8837514it's awesome for a preview model
>>8837573well it is a different model so it's worth a try but please post a real x/y when you're done comparing a 1024x lora with all other settings the same
I did not wish to lower myself by engaging your kind, but you leave me little choice...SUFFER!!!!
Why does every single anima style lora fucking suck? Even the anima loras from people who used to train decent illustrious loras are garbage.
what is this...https://www.reddit.com/r/StableDiffusion/comments/1revwgq/clip_is_back_on_anima_because_clip_is_eternal/i fucking hate clip as much as the next guy, but hear me out, maybe this shit will allow us to use artist mixing properly.
>>8837665lack od e621 dataset
>>8836118>>8837528this
>>8837718But how does it handle more complex prompts written (mostly) in natural language? That is the main reason to use Anima over SDXL in the first place
>>8837732from a very brief test - more or less the same, qwen is still there.needs more testing
>>8837718>Unfortunately it won't let you use weighting as on SDXL, but from what i tested that also was a bit better at least.It doesn't seem to be even working as clip should, I don't even understand what the benefit of this would beAll of his examples feel either like cherry picking, or confirmation bias as the difference is not big enough to warrant losing natural language promptingI've no idea why you'd want to restrict yourself with tags again
>>8837788from the githubhttps://github.com/Anzhc/Anima-Mod-Guidance-ComfyUI-Node>Seems to weaken styles, but needs more testing.uh oh stinkyand what it's based onhttps://github.com/quickjkee/modulation-guidanceit just makes things look more like sloppa
>>8837788>restrict yourself with tags againNot necessarily 'restrict', the idea is to use CLIP in addition to qwen. If you've used Flux 1 with its own text encoder you might remember that there were separate prompt fields for CLIP and T5 encoder, it's the same thing here.
>>8837791ah, I must have missed thatwell if it's just clip running alongside I could see its worth
>>8837788>it won't let you use weighting>>8837789>like sloppanyo. i nyon't think i will.
>>8837788yeah, you're right, feels like genuine snake oil
disingenuous complaint
>>8837718this is what you call a snake oil
post oiled snakes
>gen with anima>no lampsHow the FUCK am I supposed to coom now
>>8837822>there is an intricately designed bright lamp in the background gracefully illuminating the scene
>>8837824It ain't the same if you prompt for it, man
>>8837822"local lamp" in pos
>>8837825wait for wai anima then
>masterpiece, best quality, highres, absurdres, safe, Dramatic digital illustration of a long-haired, handsome mafia boss seated with legs crossed, viewed from a low-angle upward perspective. He exudes authority and danger, wearing a sharp, dark tailored suit with a subtle pinstripe pattern, the jacket slightly open to reveal a dark vest and crisp collared shirt beneath. His long, flowing black hair cascades over his shoulders, some strands catching the ambient light. In one hand, he casually holds a lit cigar between his fingers, with a faint trail of smoke curling upward. His gaze looks down from above, brows slightly furrowed, lips set in a firm line, conveying displeasure or irritation. The low-angle viewpoint emphasizes his dominance, power, and intimidating presence. His other arm rests confidently on the armrest or his knee. The background is dim and atmospheric, possibly the interior of a lavish private lounge or office, with hints of gold trim, leather furniture, and shadowy lighting. The digital art style is refined and cinematic, using sharp contrasts, detailed facial expression, and dramatic lighting to heighten tension and elegance./hgg/ never had the makings of a varsity prompter.
/hdg/ tier prompt
llm slop prompt
>>8837833why do sloppers unironically prompt such verbose shit
>>8837842Nobody actually prompts like that, they give an LLM a short description of the scene and let it write the full boomerslop prompt you end up seeing.
>>8837842>why do sloppers use gpt to prompt for them true mystery
>>8837858Based shakespearean prompter.
i think im off my anima honeymoon, the seed variance for artists kinda sucks shit
i think im off my anima honeymoon, the [...] artists kinda sucks shit
Train a lora.
>>8837900they also fucking suck for anima
back to wai white man
tell me wai!
>>8837902untrue
>>8837916post one (1) anima style lora that generalizes well outside of its training set and doesn't significantly harm the model's knowledgehttps://huggingface.co/circlestone-labs/Anima/discussions/60
>>8837922oh i do have to say i haven't tested style loras yet, my concept worked fine
reading that thread i wonder if people are doing block weighting to remove style influence, also someone reported they are using 32dim which is way overkill for a style imo...though catastrophic forgetting does seem a bit worrisome
>>8837923who the fuck gives a shit about concept loras
>>8837927me.
>>8837927>who the fuck cares about adding new stuff that the models wouldn't know of otherwiseoh gee golly i dunno anon
>>8837937even with sdxl models controlnet was preferable to concept loras most of the time, and anima has way more knowledge and controlwhat are you even teaching the model, scat?
>>8837922They seem to be talking specifically about style loras without a trigger tag.
head size slider concept lora when
>>8837943that was just an example they gave, the problem is the same with trigger tagsunless you have reasons or examples that show otherwise
anima doesn't know bbc???
>>8837942nta but "preferable" is a matter of preference. You might want to just fap-gen things en masse, without having to look for or create a reference for every single image.Admittedly there are few things noob doesn't know when prompted correctly. Weird clothing, niche /d/ stuff, real-world objects like specific cars, guns. Various symbol-shaped pupils, because the VAE ruins all training data that's not a close-up. etc.Anima has worse tag knowledge and no e621, so some tags are gone entirely. But it works better with concepts you can explain in simple words.
>>8837945>let's just assume I'm right unless you have examples proving otherwisewhatever man, I don't care enough to argue
>>8837952i don't understand why you're mad but i would prefer to be wrong about this and for there to be style lora training settings that don't harm anima's knowledge
>>8837942you can't controlnet the things i want
>>8837956Yeah, CN helps for poses that don't have a tag or a name, or are just too niche to be recognized. But even with fixed linear and shading it's hard to push the model to produce something it doesn't understand at all. Objects, textures, backgrounds.
anima doesn't know bobblehead???
>>8837718anima is pretty decent at mixing artists. much better than 4.5 at least. you have to step back from your expectations on how it "should" work with SDXL and experiment afresh. also adding natural language description on subjective aspects of how it should look (these also work in the negatives) -- what was 1.5 schizo prompting can actually add value now.i think the DiT architecture vs SDXL's u-net has more to do with artist mixing than the text encoder.
>>8837967i hadn't considered NL neg prompting, hmm...
>much better than 4.5 lmao
>much better than 4.5 even 4 was better at artist mixing than anima, and that was widely considered a failed model
uh oh analtan melty
>>8837954My style finetunes work fine>proof?no
>https://files.catbox.moe/8jvhap.png
>>8837973NAI CEO and employees fud Anima because fully trained Anima, with finetunes and loras, is the first model that has any real chance of competing with NAI4.5.But it's not just them. /ldg/'s resident faildev openly admitted to fudding Anima any chance he gets, because he's a jealous spitefag who has suddenly put on his model training hat and is now trying to build a better version of Anima where he can take all the credit. A bunch of furries also hate Anima and are shitting on it because they're mad it's not being trained on e621.I've never seen a model with this much hostility directed toward it. Not even Chroma was this bad.
>>8837984(((they))) are afraid
noob with lora , controlnet and regional prompter mogs naiv4.5 and anima btw
>>8837982ok
>>8837983mods hes ban evading
>>8837984>I've never seen a model with this much hostility directed toward itGood morning newsir!
mixing isn't even that bad on 4.5. its main issues are atrocious default aesthetic and mediocre contrast/lighting, pretty much the opposite to how nai v3 was. mixing stability is pretty bad but it doesn't feel significantly worse than unshitmixed clip models
I just like my mixes on noob and I'll miss themOkay?!
.
>>8838002Same
>>8837983should have churl'd instead
no one wants to see a poop!
>>8837983>only 24 hourswhat the fuckand here I've gotten a 3d on /g/ for telling some fag to kill themselves.
>>8838018maybe you should act like a proper adult and stop yelling at those kids with autism
>>8838020if they didn't want to get yelled at maybe they shouldn't do things that deserve being yelled at for
>>8838002just train a lora off of your gens
>>8838006ouch
found a new snakeoil for anima. try removing linebreakslinebreaks:https://files.catbox.moe/ykx44b.pnghttps://files.catbox.moe/2um3gu.pnghttps://files.catbox.moe/bi7afc.pngno linebreaks:https://files.catbox.moe/uh0wr7.pnghttps://files.catbox.moe/8zpv53.pnghttps://files.catbox.moe/aj6rmt.png
>>8838095Also for the nostalgia, schizoneg. Wouldn't recommend, it kills style.https://files.catbox.moe/vi49kw.pnghttps://files.catbox.moe/a4nffk.png
>>8838095who the fuck uses linebreaks
>>8838101me, it makes it way easier to read and edit things quickly in a prompt when it's not all one paragraph
>>8838095Is one of them supposed to be better? Just looks like random changes.>>8838101Nobody sane I would guess. But the example prompts for anima use them.
>>8838095yup, it's placebo
>>8838114model creators never know how to promptit's a rule
Considering the usage of natural language in Anima, once the full model comes out (if it does) it'd be really cool to train a lora on historical arms and armor with the accurate terms and descriptions
>>8838123yeah good luck with that one
>>8838095no... not my linebreaks....How will I handle all my prompt poses now...
Who let /hdg/ animals in?
>>8838123that would be awesome is loras weren't outdated shitware.
>>883819710 anals have been deposited into your rectum
>>8838095>it's another "cumfytard who cannot into X/Y plotting makes bold claim based on single-seed comparison"The writers really need to find some new material.>>8838114Yeah only the clinically insane would want a legible prompt.
>>8838006box?
>>8838095bro it's just shuffling noiseyou could add "usgjlbhlgfvvfhlfdllnv" and your placebo would tell you "yup this is better" for that too
>>8838095you're trolling, right?
>>8838279But that's exactly what the majority of qualityslop buzzwords is
>>8837998I can literally tell 90% of the time what style people are using because it mixes like absolute shit.
>>883832090% of the time it's riz krekkov fellatrix anyway
>>8838303I genuinely can't tell if you're baiting or just being a retard, so it's probably bothmodels are trained on quality tags, you can literally try them yourself to see what they are associated with
Her age?
31 and 364 days
that's a toddler
churl'd cunny
>>8838392which one?
>>8838417For me, it's Yaoyao
>>8838421>>>/b/946405508it doesn't know the shoes idk why maybe just not the most popular character
>playing around with weights on artists and shit>get a random 100% cosplay genahanother thing for the neg
>>8838441So even on the lawless land they decide to cater to drama instead of content, VERY interesting
>>8838481idk man i dont read the threads i just post loli
what's the quality tag meta for anima
>>8838587masterpiece, etc.(ye-pop deviantart:1.2) worst quality, etc.personally i also add cosplay and @hero neisan to negs but it's up to you
he's the hero we deserve
>>8838591anything besides masterpiece worth it? also do you put it before or after artist?
>>8838594the anima model page tells you were to put tags, which is super important for anima. also be careful not to misspell things, it actually matters. https://huggingface.co/circlestone-labs/Animayou don't need ratings tags, it will infer that from context and make near-zero difference in x/ys. you don't want score tags since they are shiny ponyslop
>>8838602I saw near-zero difference from score_9 too, as long as there's an artist in the prompt. Same with masterpiece.
I always find hard to make new characters with my models in local, what checkpoint should I use if I want to make stuff with Aria (ZZZ)? every checkpoint I know doesnt know her tagwhat checkpoints use new datasets of images from boorus?
>>8838606make a lora
>>8838605>I saw near-zero difference from score_9 get your eyes checked anon
>>8838606Aria is new character for everyone, what do you do on situations like this is look for (or rather train) a lora so the model you are using can do images with her(chinatsu and valvet lora when you fucks)
>>8838607>>8838609im fucked and cant buy a good GPU, mine is a 4060ti (8GB)When I try to train lora I'm always buttfucked and showed that if I want I will have to wait 17 daysMaybe my configs are wrong cus im dumb, or maybe Im locked cus my pc configall I have is good will to make porn that no one is making so I can get anons happy
>Aria (ZZZ)kill yourself gachafag
>>8838612if you curate the dataset i will bake italmost nobody ever does
>>8838612same boat, never found any good guide to work with.
We love to gambling here
>>8838608Did you even finish reading the sentence? Another adhd zoomer.
>>8838615I have one ready, but its for Nangong Yu, can I send to you? if so, where do I send?
>>8838612That should be more than enough for training, people used to make sdxl loras on 3060
>>8838619>single seed comparisonnever reply to me again, retard
>>8838621upload it to catbox or mega or somethingand if it's not tagged then i won't botherdo your homework
>>8838619post a real x/y retard
>>8838623Why would I gen dozens of pics just to show that you're wrong? Stay wrong, I don't care.
live yourself gachaGOD
>>8838602>the anima model page tells you were to put tags, which is super important for animaWait so is there any actual difference in where I put the artist tag?It says to use after 1girl/character/series but it's such an odd placement
>>8838624https://mega.nz/file/E8RGRDKA#cJTHRrexvpwhBFc1Nqnpy_6FeyXwrrfDM40oi7jtr44its taggedtysm anon
>>8838623Based aggressive single seed shaming anon.
>>8838631it's not like clip where you can put things wheverer the fuck
>>8838612Do you have a laptop or something? Who the fuck willingly buys the smaller Ti card
>>8837417lol, the foreshadowing
>>8838740box
xenlo, xirs
>>8838791nah
>>8838709But do we have solid evidence? I suppose I'll have to try and run some tests
Is autotagger working with forge neo already?
Was the blonde really that big?
>>8838807Based denier.>>8838819Just get taggui, anon. Missed having a built in tagger but find taggui more flexible.>>88388294u
>>8838829no, they're both redditised
>>8838829https://www.youtube.com/watch?v=k1MLZD-ga5U
>>8838827>Negative prompt: big headuh oh
>>8838829did you not watch gou?
i don't want to hear no complaining about OFFICIAL aged up versions of characters
>>8838834I wanted to binge it once it finished airing, but it kinda slipped my mind
>>8838837it's definitely an experience...the solution is fucking retarded though
>>8838807please?
>>8838827>big head in negs based>loli, flat chest in negscringe
>>8838844>>8838836
I did some permutations on the recommended tag order,https://files.catbox.moe/sp3obx.jpghttps://files.catbox.moe/przq2h.jpghttps://files.catbox.moe/sxf7lr.jpgMaybe I am blind or picked bad examples, but I don't see much of a difference
It's a transformer, you dumb niggers, it has recency bias
But what if it's a Decepticon?
>>8838856statistically they have a high chance to die early in life
but when local needed lllyasviel most, he disappeared...
>>8838849>The image cannot be displayed because it contains errors.Sweet
https://thetacursed.github.io/Anima-Style-Explorer/>he added a sort by "Uniqueness"seems like an easy way to find artist tags to neg
>>8838864How did he quantify that?
>>8838865>Uniqueness Rank: My alternative to "global favorites." Since this is a serverless tool, I’ve used CLIP embeddings and KNN to rank artists by their stylistic impact. It’s the fastest way to find "hidden gems" that truly stand out.
>>8838863>>8838849https://litter.catbox.moe/qb6n5kvdxdcm36fo.jpghttps://litter.catbox.moe/himfkzxvzip8m5me.jpghttps://litter.catbox.moe/w9bo6g0v5tey1pgu.jpgWhat has happened to catbox that it stops working every five minutes?
>>8838870artist 1st totally rapes the style replication
>>8838872>>8838852
>>8838870i might be blind, but this is a nothingburger
>>8838873yes and? what does that have to do with how it works in practice?
>>8838875I think he's trying to invalidate XYs as a whole, by saying each gen depends on the one you genned before it.
>>8838878i was gonna say that can't be it because it's too retarded but then i remembered what thread i'm posting in
>>8838875it means that whatever comes last gets special attention. in fact I bet you'd get the best results if you put the artists tag at the start and then repeat in at the end of the prompt.
>>8838880why don't you post an x/y of it? go on
>>8838880Well how about you do an X/Y plot then
your placebo, sir
the only thing that matters is whether quality slop tags are before or after the artist tag
>>8838884see >>8838881>>8838882
>>8838884are you just gonna talk about it or what
>taps on the x/y sign
>>8838890This. not wasting compute on arguments with retards. gen slop or don't, no skin off my back
>>8838791>https://files.catbox.moe/ebzhiu.png
Share a captioned ~50-100 image style dataset and you might receive a lora for anima.
>>8838894what are your credentials as a baker? I won't cast my datasets before swine
>>8838895Just passing time while waiting for the full model to come out so I can once again spend ages trying to find a config that works okayish.
>add artists tags on the bottom of the prompt>they get considerably less inconsistent and I can even get better mixes (granted, only with mixes that already worked before)Holy shit that anon was right
Anima bros, artists are coming for ushttps://files.catbox.moe/vhss15.pnghttps://files.catbox.moe/cfklvh.png
>>8838933>2 followersprobably a random troll account from a bored zoomer (most likely than not a pro ai one if they're even aware of anima's existence)in reality anima is still basically unknown by the general public
>>8838933paid for by analtan I'm sure
>>8838894orz heres a dataset for picassotrigger8 i had sitting aroundhttps://mega.nz/file/LaoXxKhB#DIo8vcF3F_-OxBnEIS7hlh_l2txAph-9M6oTvOEPHZ0
>>8838933nobody cares about this shit
>>8838933>>8838936Paid for and/or pushed by Ani, the /ldg/ faildev who by his own admission is going on a crusade against Anima because he thinks he should have been the one to get the ComfyOrg grant to train a model.The Anti-Anima twitter account and website was posted on /ldg/ less than 24 hours after that tweet, when it had gotten zero engagement. Absolutely no chance a random ldg anon somehow saw it that quickly.
Surely chat will just ignore the drama right, chat is mostly ojisan that know better right
>>8838945The noodle posts the images or it gets the ICE
>>8838907That looks overfit, bro
>>8838949Already did you fucker motherfucker double fucker>https://files.catbox.moe/q9wo4e.png
release the cunny. NOW
>>8838945what are you vagueposting about, souichirou shit?
>>8838951It is, 32 hastily chosen images wasn't great with the config. Still handy for experimenting with resizing or removing blocks though.
>>8838955drama as a whole
>>8838956How bout I remove you from the block, homie? Stay strapped or get clapped
>>8838830Does taggui have access to the same models as autotagger? Does it do natural language too?
>>8838943>The Anti-Anima twitter account and website was posted on /ldg/ less than 24 hours after that tweetThat's crazy
I legit cannot tell oekaki anon from noodle anoneven went back though the archives, it's the same style maybe slightly jaggier lines
>>8838975do you use wai by any chance?
>>8838977on occasionit's convenient for 2.5D styles
>>8838975based legally blind anon
>>8838975>bobbleheadschizowhite background>noodlekekyellow background
>>8838982I get that, but this one is white+green >>8417028
>>8838632eta tomorrow morning btw
>>8838975giga blind
I do kinda dislike that anima is associated with c*mfy. would rather have a chinese funded euge bake
btw why is noodlekek avatarfagging with the loli vocaloid
Maybe chat is right, I should put ciloranko fdy into rotation once again>>8838989she's cute and silly and creepy and stupid and a brat and
>>8838989Thread's part caring, you could avatarfag with Peter Griffin and nothing would change.
>>8838975>oekakischizosplotch/shadow nose>noodleschizodot nosethere are a lot of other very distinct differences but oekakischizo nose type is unique(doesn't always apply when he's posting with a different style thoever).
>>8838990can you go back to not posting please? or at least use a tripcode so i can filter you
>>8838988what is cute euge up to these days
>>8838986hype
>>8838997I don't really look at child noses
>>8839001you forgot to check for dupes (there were 9) and also first time using NL captions so i'm just gonna stick her name at the start as a kept tag idk lol
>>8838998why are you so mean to me chat? :( you want some yummy yummy BBC gens? I can throw you a bone
>>8839004post churled
>>8839005I don't remember if I posted this one before
cilorango :DD
Any other breeds worth testing?https://litter.catbox.moe/5yqmur.jpg
>>8839024chihuahua
>>8839024shiba inuhonestly since anima can even do dolphins there might be no limit
>>8839024pit bull
>>8839024yjk
>>8839024african american
>>8839032anon, that's just way too low
>>8839032geeeeeeeeeeg
>>8839032winrar
>>8839032wooooow
any other lora like this one?it always make the gens better, but some character loras doesnt work with ithttps://civitai.com/models/438059/dynamic-poses-slider-ponyxl
>>8838893thanks, but is cat box just consistently down or just me
>>8839043ayo this nigga still on sdxl
>>8839043make a dataset for it
>>8839045what should I move for? lol>>8839046for what?
>>8839045>>8839046m using illustrious btw
>>8839044catbox more than often shit itself on the pants, let's try again>https://files.catbox.moe/ebzhiu.png
>>8839056probably gonna have to use litterbox. new files on regular catbox are just fucked
>>8839057alright let's see >https://litter.catbox.moe/nzxks7vxcfp54ky1.png
>>8839047dynamic angle
>>8839049datasets are forevermodels are not
>>8839064im leagues behind desu... explain me datasets, how the fuck i find one in civit? or any other site
What model are you guys currently using? Anima rocks until you need to upscale, since it was unfortunately trained on 512px... I tried chenkin rf but going back to SDXL just hurts so much
>>8839069>going back to SDXL just hurts so muchhow so?
>>8839069>Anima rocks until you need to upscaleI just did some additional finetuning. 1.3k steps @ 3e-6 and it does 1.5x -1.75x upscales just fine. tradeoff is that I'm locked into 1 style, but since i never gen much in other styles anyway, it's not much of a loss and I'll manage until the full release.
>I need to upscale :|why?> >:|
>>8839074nai user spotted
>>8839072Because raw gens are just awful, it's probably only this model but all styles I try are incredibly unstable too
>>8839078so it's not style or knowledge just convenience?
how do i find smegma lora if smegma (word) is banned on civit?
>>8839073Does that mean he broke upscaling somehow? I was worried it would be innate to the whole arch.
anima is a blunder. a great project idea locked into a shitty architecture. smaller than sdxl yet somehow slower?? for what purpose???
>>8839091It's a transformer model with an llm te and 16ch vae. you can still train loras for it and even finetune in a reasonable timeframe
nai shill shaking in his boots
>>8839091>somehow slowerHalved VAE compression will do that to you. A 1024 square latent is twice as big as it would be on SDXL.
>>8839066a dataset? for training a model? a bunch of images that have a set of patterns the ai can understand? hello????
>>8839089it's an artifact of low res training i thought
>>8839100Would have to be somehow specific to this arch or his training settings, seeing as SD1.5 was trained at the same res and didn't have this issue.
>>8839072knowing there are vae artifacts ready to rape the image quality in every gen sucks
>>8839107it's the arch i believe
Was there any explanation as to why specifically the 0.6B text encoder was used?
>>8839107except it would give you 5 heads and 10 limbs if you tried to generate directly at 1mp on most of its models
>>8839113you can somewhat reduce the effect by cranking up the cfg
>>8839115the good old 16 cfg days
>>8839116those days never ended on /hdg/ :skull:
>>8839113Yeah but high res genning breaks anatomy in every model afaik, 1.5, xl, flux, auraflow; we have ways to address that. They do not break down into artifacts the way anima does.
once again proving that every arch since sd1.5 has been downhill>slow as shit>no weighting>cant upscale>loras fry>no linebreaks>comma spacing mattersall this just to generate text on shirts.
>>8839129>all this just to generate text on shirtsalso can't do it because of the 512p training
>>8839129selling it shortthough admittedly I've not used the NL capabilities for anything other than test gensthe VAE's gonna be big, if we ever get that full bake
>>8839129>loras fryskissue
full release anima isn't coming, is it?
2 more weeks
*sigh* *spends 25 dollars*
>>8839133More likely because the Unet is too small. SDXL can do clothes writing but only short phrases two or three words tops, and it's hard to make it understand what you want.>>>/e/3048862I remember a guy genning merry christmas or something last year
it can write "blacked" on underwear, what more do you need?
>>8839146proofs?
>>8839146I wouldn't consider that writing, it's probably baked-in visually same as mega milk or sugoi dekai because of how often it shows up.
now this, this is real comfy
>>8839150now post a sexy henti gen
>>8839145doesn't have much meme potential
>>8839129forgot the age-old problem where you STILL can't have two character loras interact without regional coper. we have people throwing millions of dollars at 20b parameter models yet we still cant resolve it.
>>8839156loras are brute force solutionsthe actual solution is to train a lora with two characters btw
>>8839129based 1girl stander chad, you tell 'em!
>>883915899% of anima gens are 1girls and most styles break for anything else doe?
1girl is enough why do you want more?
>>8839159>most styles break for anything elsesource?
>>8839157>take every character lora>merge them together>you can now prompt any amount of characters flawlesslyexplain why this wouldn't work
>>8839151>>8839159haven't had much trouble so far, but I'm mostly using non-hentai artists
>been going through psd posts on kemono for the last 2 hours>actually got lucky a couple of times>now cant stopjust fuck my shit up
>>8839164it breaks if you do too many at once, but two is fine. unironically if you were really obsessed with a show or whatever you'd be better off with a full finetune or some giga dim lora just so it can contain everything it needs to
>>8839165Not bad
>>8839159Bullshit, however lots of artists have awful hands
>>8839167I remember using this one on Pony https://civitai.com/models/462775/all-characters-in-arknightsit's basically 1MB per character, but he trained it very slowly and on a huge dataset
>>8839167>>8839172there was also the hololive lora for that sd1.5 778x furry finetune which contained hundreds of styles in a single lora
any new anima baking formula? all i've tried kinda works but terrible when it comes to anatomical things.>>8839165artist? kinda nice
>>8839174i tested that one and a bunch of others and it was dogshit idk theoretically as long as it's high enough dim it should work
Why anima is so good?
>It's slow andUpgrade your GPU
>>8839184nice try njudea agent
how toratora achieves this perfect quality?WHATS HIS SECRET?
lol?
>>8839186>WHATS HIS SECRET?25$
>>8839190IMPOSSIBLE ITS NOT NAI ITS NOT
it's nai
novelai 4.5 SOTA
explain this then.
>>8839195trvthnvke
why are you brazilian?
>>8839197It's NAI
NovelAI won. It's clear that there are millions still paying, or else the service would've shut down. Meanwhile civitai is scrambling to censor loras because they're desperate for cash. Local is dead
>>8839058thanks!
quick! kokonuts in on, make porn of her and send to her!!!
who and what
>>8839208go back to your containment zone
Now that /hdg/ is dead, /hgg2/ when?
>>8839032those are less human than animals
