#๐ฅ๏ฝanime
1 messages ยท Page 248 of 1
2 big.

flat is justice
Big good
Big bad ๐
iunno , to me even something like this isn't anime anymore
as in pure anime
make it a little flatter tho
and it's back to anime
all the ones i'm getting that are not anime at all, like semi-3d aren't postable kekw
I need negatives but I'm lazy

I love animaPencilXL ... just made a new merge with my Real Crystal Pony XL - YF model
this was still pure animaPencilXL
new images coming up
nice colors
yea animapencilXL is my 100% fave rn
i honestly believe
it surpasses nai
i did a lot of tests back and forth
i don't even need any negative prompting 99% of the time
anima has a lot of artist styles too. but yeh nai has WAAAAAAAY more knowledge
when it comes to characters (and artists)
animaPencil misses some older age lady looks
like geisha stuff or something?
I hope my mix including the fractal lora can change that, the fractals tend to increase age
i was able to do geisha stuf fearlier but didn't save any of it
since my original model in the merge mix included 40% pony, im gonna need to test this pony stuff
hm, yeah ok, its still somewhat pony, but not the same
surely it's not because of your vibe transfer 

Definitely not

uses her as vibe transfer

*hmmmmm, wait. trying to sense the vibe. . . *
picking up some bits, chibi style, big eyes, bows in her blonde hair, hands up
1st try
funfact if u like animapencil there is one with pony stuff in merge too, ponypencil and it still uses more nat language/regular prompting but for more pony visuals
do claw pose
nah, ill save some good claw poses for later, once I get to know this new merged model better
not sure if im gonna post it on civitai, need some testing first
...testing, testing...
adding a bit of the magic of animaPencilXL to my model
original model was maybe too much photorealistic, needed a bit of anime influence
and the pure animaPencilXL seems to generate quite young woman images. my new merge could potentially generate older looking women. just adding a bit to the variation.
i think the notes on anima pencil's merge could help u a bit on that?
like even tho the previous versions of anima pencil included stuff it already has
they re-add stuff like animagine
and maybe even something like neta art xl2 might help a bit
i found these notes super insightful lol
even tho anima contains all of it you might wanna double dip on specific parts of the model to emphasize it more
using the base models
seems a bit over complicated
it kinda isss...
if you wanna make like
that's more or less why i leave it to the more experienced makers now, i used to make some simple merges
but if you want it to be really good you have to go thru a lot of work
why not just use 2 or 3 good (according to taste) model, and maybe some loras?
i meaaaan
the results of animapencil speak for themselves tho, don't they?
there are cases where just a few things being merged
can result in a really good model
but even then u have to weight it all properly
i dont want to be the best, I want to be effective for what I want myself. And then share it with the world, hoping it could be of some use ๐
idk where i'd even start if i was going to make my own merge nowadays
i still like some of the stylistic stuff from like jemnite models
and kohaku epsilon looked amazing as a base model
Check my models if you like: https://civitai.com/user/youhnr5/models
the composition of the cute lunars look good
what is used to merge checkpoints nowadays anyways?
i might wanna try something like a kohaku epsilon/pencil merge
I use a comfyui workflow to merge models
know where to find one of those, or could u share?
i haven't done model merging stuff in a loooong time x_x
It's pretty simple, just looking at the example on https://comfyanonymous.github.io/ComfyUI_examples/model_merging/ or even just download those images with the workflow included
ty
you might want to run comfyui in --lowvram
i have 24gb vram
they actually bumping nums this time?
lewwwwwwds

merge attempt is innerestin
lol
since the main model would actually output miku's tattoo (animagine doesn't i think)
it really wants to draw it xD
definitely more style variation immediately
that workflow worked nice for just a flat 70-30 merge
the eye coloring is so gud
and the sorta jaggy line handdrawn look is nice
yup i like this xD
i need to test a lotta prompts tho
well, for niche cases it's decent. i might still be waiting for something like the next anima pencil tho. (or wdv since it sounds like it's sooner than i thought... maybe.)
cuz anima pencil can still pull off stuff like this on its own
can you do a sexy older lady wearing sunglasses, drinking a classic cocktail? no cute, no girly, not young. Using anima pencil
huh
i mean i literally just did this but u want sunglasses too
o
no cute either
hmm
ok, maybe a bit cute. old and cute can combine
im just challenging anima pencil do generate older looking women
love those green eyes on black
u just made me realize the wildcard did bibi and not towa herself
LOL
normally one eye should be pink i think
Big 
ugly 
How dare you

momoi is better 
hmmm.. there`s something wrong
What can cause my generation to be wildly different from someone else's despite same model and all generation details?
Loras, quality tags, etc
that reminds me, how are people using loras without it being in the prompt?
They use comfy nodes, probably
how to use it in discord?
is this comfyui thing connecting to the internet?
ahh sorry I'm just complaining to myself, I didn't mean to just write it here lol.
what
hands 


I thought it was her chest
but its some long thumb lmao
it runs on your own pc and you access it through a web browser https://github.com/comfyanonymous/ComfyUI
what are you currently using?
that's a loaded question, we're all using a lot of things
I can reproduce my own ai art in txt2img
but it's completely different in comfy
so i assume you're using either automatic1111's webui or forge currently. the main reason that it's different is that the comfy creates the seed which is the noise that the image starts from on your CPU while automatic1111 and forge both create the seed on the gpu by default.
oh sorry yeah automatic
i'm sure there's probably a node somewhere to make comfy generate the seed on gpu like automatic1111 does, otherwise you can change a1111 to create the seed on cpu like comfy if your only concern is later being able to make the same image on comfy (this will mean that a1111 will no longer recreate the images that you made in a1111 though).
tbh I'm just trying to see if something is up with my stable diffusion cuz everything I generate is garbage
so I was trying to replicate other people's stuff to see if I can get it to match at least a little
another thing to look at is your prompting. a1111 normalizes weights which allows you to get away with setting some parts of the prompt to higher weight than comfy will let you get away with
@supple raptor got a problem with animagine style lora traning, I trained a lora for pony then tried those same settings and dataset for animagine, uhh it doesnt work, pony is perfectly fine.

tbh I'm kinda liking what comfy produces instead
how do people get hands and feet to draw correctly on animagine? been trying different prompts but nothing works
shikanokonokonokokoshitantan
hard to say for sure what the problem is. significantly different base models can have different optimal hyperparameters, but you should still get something that's at least reasonable on animagine with the settings that worked for pony. did you take a look at loss numbers as it was training to make sure it didn't explode somewhere (NaN).
hmm those decimal numbers?
yeah, those numbers all look fine. if it was NaN instead of one of the decimals then you'd have a problem.
I see
shikanoko noko noko noko tan tan
did you just curse me with black magic
yeah, now you have to watch the show
I'm trying to get anime realism... did I succeed or does this belong elsewhere?
the hands seem a bit too real
the proportions are lame
minus the obligatory extra fingers of course, ๐ but yeah I might have to mess around with the detailers ๐ค . Thanks for the feedback.
maybe check out 2dnPony model and increase the str on some anime artists
for example
some of them are a bit anime+irl
The size of the hands in this case is quite large.
I'm actually doing an autismmix blended with sdxl using a model merge to try to play with that node more of as an experiment with the node.
any better? I can't tell anymore...
math
tru
The messages says that mouse girl needs books or else she won't be happy
๐ - Books are always wonderful! The mouse loves books and reading! but there must be a break
until i find books, i have food to offer
Anime girl 
i took a book from an anime grill who's now sad about it. will this book suffice?
Inana`nis!
lol. wildcards hit vtubers a lot
cuz got mumei right after ina
even tho like 4970 chars in wildcard
All 
wat about this book
Funya save me from vtubers 
this is destiny!
I can't do it that fast
i'd like to question destiny about so many things xD
This is a very serious issue that should be discussed over a good drink ๐
o
lmao
funnily enough
i already had an image
with a drink.
that had "destiny" sort of written out
from earlier today i think xD
c. destiny is totally weird.
No more books for mouse girl
damn! Why are you all so fast..... ๐ญ
i had some mouse girl i really liked back in like waifu diffusion 1.4 era
i probly still have it on my pixiv
nvm thought it was mouse girl
but i guess it was cat girl xD
he looks like FELIX!
lol. it's another reminder to me of how like
bad background quality and more complex things were in older models
to be fair i think backgrounds still need a lot of work
haven't seen felix argyle in a while

and why did you offend her?
Is three legs really bad?
Well, she ate the cheese... why would offend the mouse?
cruel!
Pouting everywhere

she is one. and unique!

Rose-colored glasses are not your style! ๐
yellow watermelon>red watermelon
I've never tried yellow watermelons
i think it's a bit sweeter
a lot bigger too
it's weird. it looks completely yellow
but tastes just like normal watermelon
I didn't even see them ๐ฅด
not as exotic as like pineberries, not sumthin u absolutely need to try or nething
but if u see em, they're like just a tinge sweeter, and a lot bigger
I thought yellow watermelons only existed in Minecraft
are they there?
Why?
liked it more i guess

i actually struggled getting yellow watermelon to spawn in image, that image i posted was like the only one that had it xD
Nai wasnt a fan of the idea either
your output look good btw
funny how ai can do so many advanced concepts but fail when u just try to change the color of something if not in danbooru database
wonder when we'll see anime models trained on sd3
ther's artfusion sd3 anime
but it didn't look as usable as the current stuff yet
They'll update the current sd3
After releasing the 8b version probably..
Gonna be a while for really good results
everything is so yellow!
And there's like not much information about how the model actually works, sdxl took a while for people to figure out
When they could give that information

Oh no! This is not a gingerbread house!
Watermelons can't stay fresh for long! Can you imagine what will happen in this city in a couple of weeks?
Just add a bit of sugar!

Yes Yes! catastrophe!
now it's possible
original!
the city is carved from a watermelon, decorated with strawberries. on cream?
Made by a chef!
Your house

what about freshness?
strawberry house! it's like a watermelon city, it will quickly rot
It's.. Magic

I'll get a powerful elf to cast a spell to not let it rot
I'm counting on you!
people making lolis , need to go to jail and have their hard drive checked
You do not understand anything! this is an artistic move! It's called perspective distortion!
๐
I'm always surprised by people making such comments. This begs the question of what you imagine in your head in this case. and how do you feel about illustrations in children's books? ๐คฃ
Here you go. Now I want peaches!
The hands grew together a little unsuccessfully...
green tea?
beautiful style
That! so crisp and clean... glorious! car to give a little explanation?
guys i just found out that using the style of an horny author will taint even cute models >.<
yesterday i spent an hour before noticing it wasn't the prompt, but that damn lora XD
only righteous things, pure thoughts, and LoLi will save the AI โโworld
๐
i should check every author and not only the examples on civitai -.-'
well, cute stuff is cute and that's good
but honestly, inderned don't need more horny stuff lol
very often i dream or daydream magical places or situations and i would like to maifest them in reality before i forget about them
We are romantics, we are all like that in this regard!
The technical revolution has completely destroyed magic!
yesterday i started playing ender lilies
it's economy that destroyed magic
people stopped dreaming because people can't monetize them
and are totally raptured by brainrot culture in the vane hope of escaping the tram tram that living became
It's a good game
Great soundtrack also
totally agree
like a fable
i like the genre and the fact that balanced well stressful situations to not overbear the player
it's not for hardcore players
it gives me some sort of escapism before going to sleep
the chill musics, the not too challeging enemies, the story, the sweet characters
I would play creative games. I don't like shooters and killer games...
that's more a chill souls like mixed with metrovenia
i found similar filing with fragile
that was another game i loved very much
too bad it was a little incomplete
WARNING: WILL CONTAIN FULL SPOILERS!
"Hikari" composed by "Riei Saitล" and performed by "Aoi Teshima".
The Opening and Theme Song "Hikari" to the game I am currently enjoying quite a bit, "Fragile Dreams"
This song is quite the joy to listen to... and the art style for this opening is just gorgeous as well!
=================================...
time for super heroes!
children books are boring af , even when i was a child
ive read history books as a child
then you're just not a romantic.
i just want to be dominated by an amazonian woman๐คฃ
for you terrible but i like strong woman lol
and there is a lack of strong amazonian women images in this chat , im gonna fix that later
the Grim brothers?
I'll stay away.
he does not know the true history of children's fairy tales....
๐
but why is this chat obsessed with child images
well, simps like to get abused lol
i find that disturbing
stop sexualizing it and it wont disturb you as much
i aint no simp but ok
better people obsessed with cute things than sex
That's why there are always "trolls" online?
other ai server are full of very disturbing images
it's like trash on the internet
Apparently such people don't get any attention?
kids are annoying bastards
when i was young, social degradation, mental deviations and the fear o pervers wasn't so diffuse
Congratulations, you were one of them!
before all this paranoid and deviate staff there was this mentality
youngs reminds adults how to feel and live
we where their inspiration to remember what''s the mean of living
i frequented the nursery, the scouts, but even if there were bad apples, there where many good people too
people that dedicated their time and life to protect the childhood of young life
but society is deviated, scary and scared of each others
I never cease to be amazed by people who have stopped being children and dreaming...
that's why i like cute and sweet stuff, because they reminds of my childhood
crap! I feel sorry for them
yeah, people became dry
i mean, if the first thing they think about when they see a picture of a young child is "that's a sex object" then it's probably a good thing they distance themselves from children
and with internet, childs are growing deviated
That's what I told him right away!
and with the excuse of deviants, childs and young lives are isolated and left alone, growing without guides
I immediately have a question about what is in the minds of such people.
I find it weird that's always the case with these ppl lol
totally agree
Stop making it weird for everyone it's that simple
childs cry because they suffer and needs help
it's called projecting
and good people suffer with them and they want to help
but they are left isolated, crying, because deviants
yeah, and it's what makes us human
gorgeous! Mowgli!
cool
Oh. she is woman...
Kolors does good illustrative images, although anime style is tough at times.
but immediately it seemed that this was an illustration of Kipling - The Junge Book
the quality is very good
recently i did read about brainrot
and that really depressed me
childs are growing unsupervised absorbing crap from internet
becoming the same deviants the society try to isolate the childs from...
deviants that became people of power, activists and who only god know
this is not the world i grew up
ok, better stop rambling about my childhood XD
prospective is all wrong on this one
I think we all understand your point of view
most of the time it's just about me missing my grandparents lol
and gradually you become a grandfather yourself ๐
i was very lucky
i hope so, but times are becoming very crazy, it's hard to find people sane of mind
and stay sane of mind yourself
the social pressure is simply too much
in the forest! away from civilization!
ohh that would be good XD
Yeah well I tried to get a coffee cup monster but it won't give me that either. ๐
lol
mega grande
that's an epic size lol
That's why they say: - A large amount of coffee gives birth to a monster in you!
๐
i'm a simple guy who whant's a simple live, in a society where taxes are at 80% >.<
20% is left for consideration of the issue of taxation on breath!
i should simply leave this country
that's remind me of that game of hacking
Hurray I did it!
stop being so selfish! how will the great leaders get their 10th vacation a week if you don't spare your money!
charming
There we go, even mildly anime style
oh, watch dogs
the second was me during work hours
working to not lose my mind after 5ยฐ coffy
FUCK
Why does the logo look like a bad photoshop edit 
there's no real upscaling with these flow based models yet so it is what it is for the time being.
You could upscale with XL for example
yeah, that's with one level of 0.35 denoise XL upscaling. The problem is that XL doesn't do multi-subject almost at all, so unless you're doing really low levels of denoise, it'll just try to make everyone into one subject. The correct answer is for these flow models to be able to do their own upscaling without having to resort to 30 stages of 0.1 denoise with ultimate upscaler that takes 3 hours.
dall-e is doing it, so i'm sure it'll eventually be possibly with the local stuff.
Have you tried unsampling for upscaling
I have at times, although if you've still got complex multi-subject, sdxl is still going to mess it up. it's seriously amazing if you're only doing single subject, but I have it so low because otherwise stuff loses prompt following otherwise. upscaling with kolors works better because it has a higher level of understanding, but there's also a significant number of topics it doesn't know so that's it's own point of frustration at times ideally upscaling with the same model is really the best of all worlds, just needs to be worked on.
in the end is all about knowing the bias of the model and working around it
That's why it's important for a model to have innate artist styles, you can make as many loras as you want for a model but they will never be as flexible
i suppose that's what make some cute models cute
BASED!!
Yeah but like I said it's important to have innate styles in a model, most of the artist tags that I use in my gens are from artists that make very NSFW art, yet it never produces NSFW stuff unless I want to prompt for it
i was trying to make 2 feral cats fight and instead i got 2 horny nekomimi in swimsuit -.-'
damn lora
Ah, the lora was most likely trained on a lot of nsfw images ๐ญ
initially i downloaded it because it looked good to gen gorgeous domineering miss..... i wanted a baionetta dressed as a teacher <.<
and than forgot it loaded <,<
What's the lora called?
let me see
My first lora attempt. Mogudan inspired style. I noticed that it can get a bit janky when doing latent highres fix 2x, so 1.5 to 1.7x works better.
Surely that's that's just ice cream on her hair.. right? 
I don't think melting snow looks like that 
that's why they are dry snow flakes
Yea Ik but still looks weird
very very dry snow
And very thick and liquidy snow for some reason ๐ญ
Might just be the artist tags that's messing that up
that's exactly what i was talking about <.<
God why is Civitai so slow 
yeah so I use models that have a good base look like Darksushi but look a bit fuzzy when done and use refiner once its about 80% done depending on the model im using as a base and refiner. Less detailed models tend to look cooler so I just used detailed ones to clean up the image as they finish generating.
It's 10x worse on a mobile browser
smol indie company
They need their daily maintenance
They're doing their best at least 
But honestly I haven't had any issues with it while I'm using my PC but on my phone it takes so long to load
hmm idk man i didnt prompt for nsfw stuff
Yeah it's probably just one of the artists you used, maybe they just draw snow like that who knows
like hopefully my god damn phone screen is not visible
i have another pic but now not sure if i should post it
why n ai2 why
Just crop it in your gallery and post it
I too have an obsession with fairies in jars
? explain pls?
Uhhh, where is her underwear 
"emotionless, dot mouth, by quasarcake, by ningen_mame, by yoneyama_mai, by misaka12003, by ciloranko, 1girl:human, dark brown wolf ears, clothes, white seethrough thigh highs, cold facial expression, one eye closed, white choker, quality hands, messy brown pixie haircut, red and yellow eyes, tan, background bed with a dark gray background, best quality, amazing quality, very aesthetic, absurdress"
Well if the image is NSFW then you can just crop the image in your phone gallery so you can post it here without the NSFW part
did i have anything bad in here? (stolen prompt)
its snow on her face
Looks pretty normal
Well you already posted one of those images earlier so it doesn't really matter I guess
it has no naughty bits its a magical freaking animal
mods pls just del the image its not intentional my screen is as bright as my future
What is N AIV2? ๐ญ
nai copium edition
Ah, I'm guessing that's that new arti mix or whatever it's called
Gonna assume it's neta v2
n ai china edition
To be fair they're on the right path to create something decent, it's just that both arti and neta seem very undertrained
I don't think it's that undertrained unlike animagine(?)
y undertrained? its just not good enough
It's just the outputs are weirdly low resolution for some reason
it does not seem undertrained to me
Maybe it's because it's not trained using sdxl bucketing or something
it doesnt have as much like artists data and pose data like nai
Like sleepti said, the outputs are very low quality and the anatomy is very weird. Not sure about how it deals with object permanence or with image composition but overall it still lacks a lot of things
Oh yeah anatomy still sucks unfortunately lol
I don't think no one can beat pony in terms of that
What do you mean anatomy my bro?? low quality, quality is subjective but some styles are near indistinguishable from novel ai
Not low quality, low res
haven't tested with nudes or anything, but i had some more revealing pics, ik my screen is fucked but come on, anatomy bad? bullshit
quality is subjective and i think i've never been able to make pics this good
I mean I think it's pretty self explanatory, if the model is just good for portraits of characters standing then it's clear that it can't do proper anatomy. Try prompting for some different poses like on all fours or lying on stomach feet up, something a bit more complex, and see how it does
Ok havent tested tomorrow i will
Yeah I meant to say low res, not low quality
can i get some tags or prompts for quick test tomirrow
for now this is my best example which says nothing
if you're talking about civit ai pics... those pics always display in low res, u gotta click. If you're talking my pics, didnt download the file, just downloaded the output lul
Look at your previous pic you posted for example, one hand is melting into her face, the other is melting into her hair, the eyes are falling out of her skull, the hair on the bottom right is disappearing into nothingness, her dress is shooting up in the air.. At first glance it looks nice but there is so many issues with it, that's why I said the model is undertrained
yeah actually lol but that pic is obviously screwed
still tho its the nicest stuff ive used for a while (and i gave no fuck about tags ordering that must ve done something too)
The creator of the model is aware of this problem, he knows that the quality and anatomy are the model's biggest problem, people have given him advice and feedback on what he needs to improve so hopefully it will be fixed
I'm talking about the dude who made Arti, not sure if he made the neta model
Well yeah that should get fixed along with the anatomy problems
Pretty much, but at least they're not trying to prune artist tags from their models so that's a huge plus.
They just need to let their models cook a bit more and it should be good
Did nai do that? like were they planning to replace artists
i think they need to up the weights on some of the pose description tags
Not sure, they removed artists from their furry model but that's probably because they think the western artists might cause them trouble for some reason. The japanese artists don't really seem to care
I heard they were planning to replace with placeholders like somewhere idk
If they were then they would probably let their users know through their discord server or something, I doubt they would remove artist tags considering that's mainly why people use their model
People will always spread misinformation, it's best to wait to hear something from the actual NAI staff
in any cases, fingers crossed neta 3 or 4 kicks nai 3 out the window ๐ฏ๐ฏ. 2 is a phenomenal leap all things considered
I wouldn't say it's a huge leap considering local anime models are mainly trying to catch up to NAI, having artist tags is the bare minimum but local finetuners really want to prune them from their models for some reason
Honestly I can't find something that I've liked this much before nothing quite come close
it's a huge leap for me
Yeah it's fun when a model actually has hundreds and hundreds of artist tags to mix and play around with ๐ญ
That's the most fun part about NAI in my opinion
When they MIX
So I 100% agree with you
Another model I remembered is uhhhh
Oh yeah, trust me I know ๐ญ
anime illust diffusion
it couldnt mix but it had a style i quite liked, and a lot of artists
but it was unstable, i posted a pic here.
when the style looks cool, i can forgive a lot of errors, but that one was just too much
I remember seeing that model when I first started using Pony and I was honestly amazed at how good it looked.
I liked some of pony's outputs too, but it looks fried lol
Never used it but some of the stuff people were posted made me realize just how bad Pony was at doing actual anime styles ๐ญ
It does have some artifact problems, but that's not really that big of an issue, to me it's just the fact that the images it produces look very basic, not really aesthetically pleasing if that makes sense
Anime looks basic 
some of the styles didnt hit, and stuff looks kind of fried. I will say I had enjoyment doing pony. When Neta v1 came out, it had a style which I kinda liked, and it was able to do pretty cool stuff
You look basic 
Thats why when they showed v2 i get hyped
Ugly eyes

Apologize to makima rn

๐
you look delicious

Trust me I know, after playing with Pony for a 2 months I got really bored of it
as a phantom thief i shall steal your organs
I didn't bother that much. This will keep me playing for a while tho
You taste the best ๐
I think there'll be a new article about the new Pony model soon, we shall see what astra managed to improve 
obviously more unicorn segs
Furry
change note 6.9
- more unicorn segs
Apparently the dataset is even bigger now? Which is huge if true, cause that would mean an even bigger Anime dataset
guys i think theyre the same
Ponyv6 anime dataset was only 1/5th of the model, around 500k images, so it's no surprise it's bad with anime
Avatar is the best anime
anime =! cartoon
dumb cat 
i think its != dude
at least you understood, so I dont care
Best anime girl
yes anime equals factory cartoon
no, you're wrong I'
ve looked it up
both != and =! exist. =! translated to IS NOT
oh ok so you did mean anime equals factory cartoon
mommy baionetta is totally my type <.<
No, I ment anime IS NOT cartoon, as in these two things are not the same. Or it can be read as "anime is the inverse of cartoon" which is not fully true.
that, which technically means you're wrong ๐
the IS NOT that =! means is not the same as the IS NOT we humans use
just read it as "anime IS NOT cartoon" and it is fully and completely correct. But if you want to read it otherwise, then yeah are fully correct and there is no room to discuss. So, next topic.
Im human
and I used it
but anyway, lets move on
you'r eright
anime would have to be false since that's the first operation being evaluated... ๐ค so you hate anime not cartoons?
what?
yeaaaaa... trying like t_ponynaiv6 which afaik is one of the better pony models and i am not super impressed with it other than its more stylized outputs.
but it can't attain "anime" style in any way lol
I love using the rapunzel lora combined with anime checkpoint, if that answers your question
Then it's a good model 
no i found better anime

SHE is best anime
where's the spot for the skibidi toilet
under your bed tonight
That's because that model was trained on a single NovelAI artist mix, it's not really compatible with other loras cause it is simply just a model made to replicate just a fraction of what NAI can actually do
Hence why it's called ponyNAI
Nai does horni 
figured as much from description, still seemed like there might've been a handful of good outputs from it
but ehhhhh
waiting for wdv which will hav eno pony influence
and likely be better than nai from what's been shown
til then, animagine merges ftw
Will it have good fingers 
It's just like using a singular lora and expecting it to do multiple other stuff, it just won't work
The first version of https://civitai.com/models/412672?modelVersionId=459938 was a mix of 40% pony based model + 60% SDXXXL + some lora
It does very good horni 
latest version, or perhaps another model, merged this version with the current animaPencilXL_v500
should be nsfw enough with this mix ๐
hands can be fixed. but pony-based models can never do like actual anime screencap quality
My 1girl prompts turn into massive cat girls on the beach 
Bad quality 
I'd have to see your prompt to actually know what you're talking about but that just seems ridiculous ๐ญ
that sounds very plausible I think
Yeah Pony does suck at anime tbh, on NAI if you wanted a simple anime screencap you'd just simply prompt "anime screencap"
yeh; i guess even then tho the base style/colorization/etc of nai is kinda falling behind now
I assume other models do the same
but it's only natural, been like 8 months since it came out
and yea in animapencil just anime screencap does it
How so?
i did a lot of comparisons just prompt to prompt, not even just anime screencap stuff
i think the merge i use now surpasses nai in base quality BUT
nai has the vibe transfer stuff
as in multiple init imgs
and what not
I'm confused, what do you mean by base quality?
Yeah vibe transfer is honestly insane
without adding a paragraph of prompting, same prompt across both since both use same prompting style
as to, for me which image came out better without using any init img
Yeah I just thought about it and tried before it even happened

Idk, to me local models just seem very.. bland tbh, they lack the aesthetics and the image composition that NAI has
But I will say in terms of prompt following Pony is the best imo
@ivory pebble I sent you 
All I've been posting recently is nai or sd3
prompt: 1girl, shikimori, kawaii dake ja nai shikimori-san, blue eyes, pink hair, long hair, hair between eyes, school uniform, black sweater, shite shirt, collared shirt, long sleeves, spaghetti strap, red ribbon, grey skirt, pleated skirt, miniskirt, black kneehighs, brown loafers, medium breasts, fingernails, standing, staight-on, skinny, tanaka yuuichi chunpai kogame
i meant to show what i meant
lol
i'm running out of anlas for this kind of testing
xD
but i'm certain
these 4 outputs would be better than 4 images output in nai
with the same prompt
nai won't even overfit for a background
What even is that model
I mean, nai has that weird prompting thingy that needs you to pay attention to the order
it's kinda similar with animagine based models tho
Nah not really, nai puts a heavier emphasis on it
If you put the artist like on the far back then it's pretty useless lol
I found that prompt order or format in general does not have a very large influence
and anything can be adjusted using weights
so if you mostly control things with (somewhat heavy) weights, order does not really matter
Sticking to a sort of fixed template for prompt might be more usefull for use human, not always better for the checkpoint
1girl, shikimori, kawaii dake ja nai shikimori-san, blue eyes, pink hair, long hair, hair between eyes, school uniform, black sweater, shite shirt, collared shirt, long sleeves, spaghetti strap, red ribbon, grey skirt, pleated skirt, miniskirt, black kneehighs, brown loafers, medium breasts, fingernails, standing, staight-on, skinny, tanaka yuuichi chunpai kogame
with my new custom merged model.
only about 12% pony in the mix, but it still seems to have some pony features
talking about the smooth empty style
ewww a shite shirt ๐ฉ ๐ ... ๐ I know it's a typo but it's funny.

I just copy pasted the prompt from earlier post, didnt even fully read it
๐คทโโ๏ธ at least someones reading it ๐
so there's a typo within the original base image i used xD
mistakes happen
or however that saying goes... ๐
I once made my first fractal lora, created lots of pics for the civit page, and posted the model + images. Later I found out that the lora was NOT active, so all example images for the lora were a fake
i found a repository of like
a lot of images with prompts baked into them
they've been useful for those chars
since they keep them consistent
Don't see anything special there, those look like NAIV2 images let alone v3 ๐ญ. The thing about NAI is, the images won't look that good unless you artist tags, I use up to 10 different artists at times and it genuinely looks like real art, as I'm sure some other people who don't use AI would agree with if I showed them
oh btw, 99% of my posted images have the comfyui workflow included, fyi
if you have a prompt to share i could test it in anima lol
but in my example
That Jotaro looks clean though
The main problem is, like I said before, people just don't know how to mix artists, that's why you see the same typical glossy NAI images with extra highlights. You can do so much more if you actually took the time to experiment with the model
but that's a problem of the model itself
that it does that.
animapencil is like RNG style
tons of different styles
u can nail it down to a more specific one with artists too
nu
It's not a problem, people are just lazy to learn how to use the model. That is literally all there is to it
i used nai for like
half a year lol
and i got good results
but a lot of the time i mixed other things in like the vibe transfer
Lemme see some 
i can get those same results with no init img in animapencil
Hmmm but this chat told me that's bad 
they were all used for work-related purposessss xD except the ones i would've posted here when i was testing
I have an app I slapped together that runs ontop of comfyui but it doesn't save the metadata in the images otherwise I'd totally do that. I really need to figure out how to do that though so I can expand my apps capabilities in the future.
fractal lora activated
#๐ฅ๏ฝanime message like these ones when testing diff styles
or this which i still have on my steam profile lol
Those look pretty generic, but I guess you like the anime screencap style so that's fair
ye, we established that
to me most things that come out of nai/pony/etc seem generic cuz of how overused those styles r
If I wanted that I'd learn a good artist combo but I personally like something more detailed with softer colors and thin lines
That's why you experiment and you learn new ones
i do combine specific artists sometimes for results like this
but if i can get same or better results as i was getting on nai with the same minimal prompting style
That looks like something you'd get out of Pony ๐ญ
then i declare this merge betterrr
i literally
can not get pony to do flat colors
like that at all
xD
and pony prompting is even worse than anything else
with all the score_X stuff
tpony nai
it looks more like bunnyspoon
lick
than wut i posted

just learn to use some general wordings for style, colors and lighting. for example:
(over/under)-saturated
bright colors
dull colors
realistic dark lighting
flat colors
thick (colorful) lines
colorful
backlight
etc
but no matter how much u do that, i dun see it coming close to this styleeee
except
I mean that's fair, some people just don't like going through extra trouble to make something, so I understand. However I'd rather go through that extra trouble to then have actual anime images 
2nd to last one
pony could probly do that
but the like purely anime adjacent ones
except literal anime images are my favorite style xD
and it's taken years for models to reach the point
it can actually do it without like misplaced pixels and looking odd
Boring style
stuff like nai style has been possible for agesss
try my cute lunar model, its a merge from some older sd15 models. can do pretty cute anime
not a fan of sd1.5 quality models >_< the only one i could get to do decentish results was like wd1.4 in that era
It hasn't tho, NAI style isn't a thing ๐ญ, what you're referring to is just NAI artist mixes
nai style as in, don't add any artist tags or franchise names, do a generic prompt
give me some prompts to try in cute lunar
naa i can't even anymore lol
NAI without styles? Might as well go back to anythingv3
lost all values at that point bro
It's like i'm talking to a brick wall ๐ญ, oh well...
facedesk lol.
LMAO
Is it really that good? How come no one is talking about it
I respect the dedication to burn money to use nai without styles
Fr, that's like trying to use Pony without the score tags
Everybody model has like a bajillion of styles
do they ever care enough and up the weights so that it actually has any effect?
NAI is mostly popular in Japan as well and everyone knows about it, that's why I'm confused why no one is talking about that model
pretty sure most of the people you find will be using a different model than someone else. whenever i view params for an image like no 2 people use the same model
This is the one and only nai style 
That's another thing people don't realize, you have to mess with the weights on the artist tags otherwise it just won't work, this is also extremely important for NAI
it's just not talked about on this disc or touhouai but in the jp discords for sd i'm in it's talked about frequently
Which jp discord
so many people today 
I'm talking training weights
Like, what's the point of artists tags
if the images can never resemble the actual work
Do you have any images to show? I really wanna see 
there's a handful of people today
I'd rather not join random discords to see images ๐ญ
No, only the ones that like anime "style" are arguing 
animagine has styles too
We vibing 
trust me bro.jpeg
they don't recognize the true best anime helicopter obviously
im just gonna show some sd15 anime images now being created with Cute Lunar
No arguing here, we're just comparing models and what they're capable of
like NAI sauce it's just properly weighted tags bro
and like good dataset
(it's hard to achieve but i'm just saying)
Comparing sizes

can't provide that link/info;
Many moneys
but 4lex like, the only real way to compare would be if you shared the same prompt between the two? all your images have no prompt attached

capitalism open source
I mean, the comparation wouldn't matter if the other model doesn't recognize the artist style
that's what i did with neta lul
What's interesting is, NAI can actually do a proper black and white background unlike SDXL. They even said they trained their model longer than base SDXL which makes me wonder if they actually changed its architecture to achieve that
Uhhh ๐คจ
it's a near pure animagine xl 3.1 merge, like every model it's merged with has animagine dataset, it knows a lot of artists at least.
unlikely, cause anything had that same world breaking impact too yk and it was the same architecture
Isn't that just a vae thing 
it's tags weighting, image strategies and blah bleh bloop
I've heard artist tags are fried on animagine, as in they don't replicate the artists quite well
the ones i've tried have worked well
Could be, I'm not a tech guy so I have no clue ๐ญ
are they close to the default style?
don't mind him lol
these are the artists i pull from
He's just a silly cat 
no, ones with like distinct outlines/line art styles work fine too
Sd3 can do the big black colors and people said that it has to do with vae 
Makes sense
just tags weighing, images strategies... boom NAI v3 easy non issue
(its a headache for us mortals but you get the point. slapping a dataset and call it a day gets nowhere)
I'm the most expert in anime here
stooopid ccat
You're my dinner
Agreed 

I think there is a lot more to it than that
Bait used to be believable
Huge things behind the model
It's their "secret sauce" they mentioned before, not sure whether that means their H100 cluster was the problem solver or if it's something that actually has to do with the model training
"..." i had that for a reason lol. But no I don't believe it screwed with architecture
Booba
They know as well as SAI does like atmost
can't wait for a month to pass and for naiv3 to be forgotten in comparison to wdv ;_;
pretty much copium until there are more previews lol
and I doubt SAI would change the architecture for a random model, like if so, they should not be dying rn
boy i wish
What's wdv
anything just anything
almost certain that's not gonna happen, but i wish 
wdv0.9/ppa looks kinda decent though
wdv, ani 4, neta 3 just anything