#🍥|anime
1 messages · Page 121 of 1
@lyric zealot 
Powerful stuff there.
can I dm you 
sure thing 
Not red-eyed this time, neat.
I mean the darker-haired one that's not having pink hair.
Twinning it up. 😮
both definitely have bad stuff planned
They about to make out?
looks more like "I'm tired of this girl, can't wait for this to end"
🐭
I generated a bunch to see if some would be cool, forgot to change the seed to random now they are all big booba
i see no downside 
She looks strong.
we got big boobs we got small boobs
but what about inverted boobs
just crater
robot girl opens booba up and shoots missiles at you
Very beautiful display there.
@dawn jasper shh...
it's a guy in this case but yea
maybe you can make some too I never see you make any

@oak spoke Do you like my rat girls
they're nice
Adorable, wanna talk on direct messages?
I'm tired of AI for now
and it's hot
damn i cant seem to find any loras or models to make some creepy monsters
@finite herald https://www.youtube.com/shorts/2oa5WCUpwD8
#shorts #animation #hatsunemiku
Anamanaguchi - Miku ft. Hatsune Miku
https://www.youtube.com/watch?v=NocXEwsJGOQ
Happy birthday!
Looking for Strange_Dedication_v5 LoRa
She is fabulous.

@finite herald we got a spammer
crush his skull.
my main account (subscribe): https://www.youtube.com/@steakofsaint

I offer smug miku in exchange for spam removal
Check out our Loras for making Niji Anime style AI in Stable Diffusion. https://www.etsy.com/shop/TheLoraCollective
ooeoooooooo
I think you guys have a slight misunderstanding of what Niji looks like
@barren lintel this one is just a semi-disguised ad
another miku for the ping
Like this?
@finite herald +1
Who is this woman?
The beauty of this lora is it can create different styles with a similar look. I believe they are comparable to what you're sharing.
u can edit the image on PS to add bloom,saturation/curves,dof to get a result very similar to niji
👀
How bad of an idea is it to upscale using another model
with different artstyle
extreme close ups are funny
You need a good base to create from and the lora works great for that. Photoshop can be used to enhance any image to the user's liking.
wait a minute, I ran this on another model to upscale, it only did the upper part differently, why is that ? >.>
I'm doing tiles through controlnet my pc can't do hires 
speakin of closeups
These glow-ups are expressive.
They are all hot, I will not lie. 😛
NO WAY
Just in time I was gonna send this
#shorts #animation #hatsunemiku
Anamanaguchi - Miku ft. Hatsune Miku
https://www.youtube.com/watch?v=NocXEwsJGOQ
Happy birthday!
I'm surprised you guys were making art of Miku
any must have settings for a fresh install of A1111?
Patchouli Knowledge is powerful.
no body has purchased any of my artwoork on opensea 😦
thats how it is 😔 unless u are famous it doesnt sell
cheap ass rich nft kids
like they cant afford to drop 1000$ on some shitty ai art on a starving artist
launch your own coin call it obamacoin
i wanna make somethin for the sticker event
do it like iceposeidon when he scammed some ppl 500k with his fake CXCoin
Noel Vermillion of BlazBlue fame?
Yes?
damn... abstract art can have some pretty spicy effects depending on the model
don't wanna say best girl, but she was first girl (for me) so
woa i thought u drew those
lol, nah I'm ok at fixing images, but can't draw something from scratch to save my life.
;3
going to replace some of my models with these to save space https://civitai.com/user/fp16_guy/models
no need to redownload. you can easily convert and prune your own models. https://github.com/Akegarasu/sd-webui-model-converter.git
ohh thats intrasting
might give that a try , does it take hours?
also wondering about trying one of these https://github.com/wkpark/sd-webui-model-mixer
nope, it only takes a few seconds per model. I think it uses CPU and ram too so low VRAM shouldn't be a problem. for model merging, I use this one is what that is based on. https://github.com/hako-mikan/sd-webui-supermerger.git
it's admittedly a bit complicated but works really well once you figure it out. I have a basic howto made for it that I can find if you want it.
ok cool
oops

How do you know if a model is it fp16 or 32
2gb vs 4gb
size,if its big its 32
Interesting volcano
aight, imma satisfied enough and tested enough on my 1.1 model version
i kinda fixed the horns and circle appendage issue for a bit
typically it will tell you when you download, size is the other method. if it's ~2gb then it's already pruned and fp16. XL models are bigger and I don't remember the sizes off the top of my head, I think it's like 6.7GB for pruned fp16.
yeah, 6.7gb
damn. drunk nubby has good memory
my stupid ass still with them 8gb 1.5 models 
@supple raptor
Drinking is bad kids
i held out on them for a long time, telling myself they are better. they really aren't and you waste a lot of time switching models if you don't have them on a fast drive when you keep them at fp32
I'm old so it's fine
No
weren't we like almost the eldest around here? @supple raptor

yeah I think we were the oldest around when we asked last.
and now with a lot of people who don't show up anymore... i'm pretty sure we are now
can't 2minutepapers do another video which boosted popularity like mad?
if we get XL p*rn anime model they will come back
I think the novelty has just worn off for the masses at this point. A lot of people have also just moved to other servers.
The one model I like using is 4 
TBH I spend a lot more time in Touhou Project AI these days.
use the model toolkit, load it in -> safe as fp16
Just convert, it's no problem. You're lucky that A1111 can load fp16 models into fp32 now. In the past you would have been forced to use fp32 models since your card can't do half precision.

What do I select
I've had SD installed for a year now, and I still don't know a single term lol
yeah i have quite a few models well above 3gb that im gonna converrt
click that Delete known junk data checkbox and you're good
i was under the impression bigger is better for unknown reasons
How is this woman so lovely?
From a technical standpoint, yeah bigger means either more precision (fp32) or extra training data which is helpful for people who want to continue training a model. In practice, pruned FP16 has no discernable loss in quality for generation (though outputs will look different)
it worked, so fast 
Training is when fp32 or at least bf16 has actual benefits. Generation is sloppy enough that it doesn't matter.
i suppose training on fp32 is best?
ah. today is the day i die -> my cat is sleeping on my lap, i'm forever condemned to sit here
yep, though bf16 isn't too far behind and saves a significant amount of vram for those of us who don't have an a100 cluster to use 
@supple raptor fox girl saying thank you gen
pat pat

i was gonna go to bed but now i kinda wanna mess with this thing
what is... this thing?
the model coverter and mixer things
well, i started mixing on a drunk idea xD
not even an idea, just something random but it was fluke that worked xD lol
i'm not read up or motivated enough to train models, i suppose
i just developed a bit of a feeling how to get stuff out of a model into another
0 actual paper's read or research how to do it properlyXD
now i'm generating new images for new model post
fix clip?
you'd have to ask nubby if he's still awake
dude, do you have trigger on your name?
xD
Ok, so CLIP contains (i think) 76 integer values. When you merge models, sometimes those values are converted to floats. When that happens prompts will often misbehave and certain tokens may be outright ignored. CLIP Fix converts them back to integers which fixes the misbehavior, but will also change outputs.
lol, just happened to hop over here XD
oooh, i might have to check it then
which extension is that?
how about vae can that be set to delete
yeah, it's a good idea to check. there are other ways to do it, but this way works pretty well. My preferred method is to extract CLIP from a model with known good CLIP and transplant it onto my merge. I use Aurora CLIP normally, but I know that wouldn't really suit your tastes. https://github.com/Akegarasu/sd-webui-model-converter.git
since i pretty much always use the custom vaes
because the moment you said "ignore tokens", lightbulbs in my head started flashing
Yeah, if you set a custom VAE anyways, you can prune it from the model. It doesn't save much time or space anyways, but it won't hurt anything to remove it.
gonna post the model at some point, so imma need a vae in it xD
Yep, the first version of Everyone-Mix had broken CLIP which I only figured out like 3 weeks after uploading it. I'm not sure how or if CLIP fix works on XL or 2.x models. OpenCLIP is foreign to me.
goddammit nubby, now i need to start checking clips from various models too
This one can tell you if CLIP is broken. https://github.com/arenasys/stable-diffusion-webui-model-toolkit.git
i use that one -> says clip is fine
but testing various models on their responses to the CLIP, damn, that's a huge undertaking 😮
Here's what it looks like when CLIP is broken
Once again my drunk memory surprises me with knowing CLIP is 76 values.
there's also this
i think it was documented you can recall facts you learned while drunk, easier when you are drunk
but is it drunk, or just a bit tipsy
I learned most of this shit while sober. I have a gift for recalling random knowledge though.
state dependant memory got me through highschool
study high take the test high get high scores
xD
Honestly more tipsy than drunk. I don't get actually drunk very often, it takes too much to get to that point and the recovery is a bitch when you're old.
accidental furry
she had an accident?
she slip and fall into puddle of water
male test
he tested just fine, that's a good looking man ❤️
he look very fitting for yaoi

bara? 😛
I hate storms

Nope just stole prompt for better prompter
This one has metadata
nubby, are you going to skip sleep again?
i don't skip sleep. i have a very strict sleep schedule. bed by 5:30am saturday and sunday morning.

Me waking up because of the sun
fact 
More like, me sitting on my bed to go sleep and then I realize its 8 AM
actually me this morning 
Das me everyday xD

such a messy eater
Also I may betray Aurora 
did you find a new model or something?
How old is aurora?
I'm trying one, and its not new new, but I think it has something I want
It was uploaded to civit on 14 April 2023, but I had it before that :o
what is it?
oh damn, it's been 5 months already

AnyLora, it seems to be really flexible with style LoRAs, more than Aurora, and all I use is LoRAs for style, I thrive on LoRAs
I'm still addicted to Ink wash painting 
who wouldn't be?!
ahhh, that model that's used to train most lora now
huh, lora's, what are those? /s

Oh yeah
It seems to be used for a lot :o

iirc it's pretty good for training lora too
It's your fault 
who me?! i'm innocent!
I wish y'all could generate a father for yourselves
Seems to be good, but seems to tend to overtrain :o
i wash my hands in the ketchup of innocents!
True and real
are you just a generic troll?
why you got to put him down like this? Asking if he's a troll is enough, you don't have to add "generic" 
assuming it's a "he"
that's why i'm so nice and degrading, it's not the first, not the last, and i certainly don't want to waste a single more breath before getting someone's ass banned if that's all they're trrying to do 🙂
understandable
Fair :o
I mean, I don't get too mad at em, I just go with the troll because its funnier
and well, if that's someone's first message, it might as wel be his last too
nah, i'm done with that
Probably... But I only discovered it because of this
.
lets throw that prompt back into my current generations
still generating pictures for a new post 😛
I KNOW
these kinds of images, i'll probably never be able to make, on demand or by accident
but that's ok, that's why everybody makes different images ❤️
because you are on SDXL?

because i just don't prompt for stuff like that 😄

there's sdxl models which are able to
ahhhh, well that's the beauty of SD. Letting anyone able to "create" art based on what they like

ugh... i probably can even think of a recipe that would get anime style into my model
if i could , should i?
well, in my case, im using regional prompter, which SDXL probably don't have right

why not, anime sells
regional prompter works in sdxl, even VERY well
yeh
yeah, SDXL ... IS heavy to run
are you eyeing for a card?
4090
damn bro
ooh, that's very exciting!
yup yup
just need to make sure it's the right time to get it
don't really wanna buy now, and then 6 months later, 5000 series come out

yeah, i'm sure... right now my psu is pretty much on the right spot, 750w for a ~600w system
a 4090 would kill that by itself
I think my mouse just fucking died
i think it really depends on your whole setup
computer, or real?

lmao, i didn't think of it that way
yes lol
I'd never buy a hamster or something like that
they die so easily
so many friends that got one only for it to die the next day or a week later
my hamster when i was a kid lasted a bit longer than that
download your newest AI hamster now, TRUE AND REAL
🤖 🐹
when i was a kid, i got two hamster, one of them died, and the other ate the body 💀
AHSDOHADUOSHADUOASHADHOASNAO
oof, that sucks
i've done that for over a year because i had to
due to injury
couldn't use my right arm at all
I'd say mask
I'm seriously addicted to the animation here https://www.youtube.com/shorts/2oa5WCUpwD8
#shorts #animation #hatsunemiku
Anamanaguchi - Miku ft. Hatsune Miku
https://www.youtube.com/watch?v=NocXEwsJGOQ
Happy birthday!
currently having one non anime being rendered, not as good
moment
it's still rendering, the next one is going to be in the scarouse department 😮
she's just a bunch of tentacles too
are you on a sdxl model?
no
SDXL?
regional prompt

Merged with ghostmix, that's probably why
A lot of weird prompts that wouldn't work in the past are now working
here with metadata
but will probably have a lot of weird stuff inside, since it's using regional prompter
ugh I gotta revert to 1.5 models... sdxl and my patience don't mix 😄
More realistic take...
3090, a single highres image takes 2 minutes
ah
that's really one of the reason why i stick with SD1.5 models, and also there's not really any good SDXL model that can do good anime yet
on 1024x1024?
1024x1280 50 steps restart -> x2 15 steps sde 3
jesus
but, image quality is :chefs kiss:

why u get all these emotes? 😮
😮 dang
how do you keep that straight?
no idea why I'm in most tho
I swap between multiple accounts and wish I could merge them. 😛
nitro user
I forgot that they fixed the prices
let see who's faster
it's so fast, it's probably a bot
I'm heavily debuffed rn
there's probably a bot that looks for link that looks like nitro gift
and auto claim them
probably
😦 damn
in theory, should be pretty easy to code
sorry smoll sus

i wasn't quick enough
I wasn't ready
🤣
Can't you just see who claimed it?
no
oh
same here
I was using pokecord back when it was around
everyone staring at pokemon spawn
and everyone jump at it
first to finished typing the name got it
and I won almost everytime

masochrist mouse
spank it more
i think it's almost impossible to win against those bot, unless you use bot 
probably
This model is nice, more versatile for styles than Aurora I think, or at least more versatile for flat coloring, but its also not Aurora, so its not the same 
what is your (thicc:) weight on
#2 face is a little cursed
'yes'?
XD
jesus
the addaes shirt 🥵
ah yes I can totally read those

I gave up on cowboy shot a long time ago
Always drippy
Same
try cowgirl
I accidentally realized that the AI was sneakily giving people in the background cowboy hats
But I mostly just do "standing" and then "thick_thighs" or "thighs, legs" if I don't want thick thighs
it was funny for a few gens
Mine doesn't do that XD
sci fi cowboy when?
It probably shouldn't
that wasn't meant to be reply
but it was probably because I was using it at like... :1.3 at the time
I used it at 1.3
My prompt
Most of it
Is weighted
xD
try prompting with a crowd in the background
like this?
This is my average prompt
I always do white backgrounds XD
very nice
I love generating like... reference sheet style pics, I almost never generate like a character in a enviroment
interesting eyelash
interesting everything tbh
Cool eyelash :o
prob hurt her eyes to look out
And true
when your eyelash glow

this one just look ready to pounce you
if you add little breathing fog
do it
she got the 3rd eye tho
wait is this a lora or something?
The dress looks similar in both images
nope
like this?
no lora
just my special model + my prompt
I like to avoid using lora
me too @oak spoke
I use nothing but LoRAs
More like just 1 shot, I do multiple angles sometimes, but it can be inconsistent and harder, even with controlnet :o
This image is powerful.
This one is also amazing.
Miku Hatsune is adorable here.
got a better one
I'm vibing right now, playing miku song on loop
generating miku
life is good

peak
big hat
smol
2 sides (blood warning)
solo-leveling art = god-tier
solo-leveling story = ...
gimme some beginning after the end or suicide hunter
well, fact
suicide hunter actually had me in tears with the martial arts arc
tbate?
oh yeah, for sure
added tho
it's... surprisingly VERY good
i love it
It's soooooooooo good
infinite leveling murim? read that one too?
yup yup
im currently reading world's greatest sernior disciple
it's only like 49 chapters
but it's pretty similar to infinite leveling murim
if you like one, you will like the other
nano machine = generic
i got bored after like 144 chapters
it didn't get better
nano list tho 
i think i dropped it... dunno why
it's not for everyone, but it's kinda interesting
and the artstyle, i love it
the scribe who retired to the country side -> was interesting, becamse trash
i love her "tattoo"
never heard of that
Latna saga @lyric zealot up there with the beginning after the end, very good actually
some of those manwha's are really good, and some are just... generic meh trash
ooh, new chapter of mieruko 😮
i think it's on hold for me
for some reason
no idea why
it should've started a while back
is that a manhua?
i'm generally not liking the chinese take on those tho, they're usually extremely generic
yup true true
but the return of the mount hua sect
is probably one of the best manhwa out there
or at least imo

i had one non-anime, i just couldn't post it
y not? got open wounds?
I don't know
it was SCARY real, i'll dm
just a feeling
I'll delete it to be safe then
if you see it you see it

it was a good pic tho 
ty
post it at the other server
there's a channel for that
uhh. the unstable type server?
AnimeAI
never mind
the channel got removed
lmao
@wispy canopy @lyric zealot Have you guys read Academy’s Undercover Professor?
nope, i'll add it to the list
i'll also recommend 'the greatest estate developer' . the mc faces are just the best
It's my fav at the moment
i think it's on hold for me
but will add it back
It's been awhile since there's update, right?
still weekly for me, but i'm using an aggregate site
what chapter is the latest?
104 august 2nd
yup, it stopped there for me
ah, i'm even on 97, no need to rush then 😄
@lyric zealot meow

@lyric zealot I can't find it anymore
but have in mind that I liked one of your gens up in the chat


you couldn't find one of my gen? 
its been many hours since I looked at images, was going through the new ones, many cool stuff
also saw you saying about the 4090 I went to look and just the card consooms my whole PSU worth of power 
Its fiine 
i know the feel (it's images, not files) XD

i didn't even see that options

oh, that's reacton gifs too
OHHH
the little rat model and "lora"
the style is made with another model through refinement
Can someone pick a PFP for my FB account for me
is refinement even a thing for the 4gb me
Whats this league of legends looking logo
Its time to stop
Idk
But Atleast pick
I had limited prompting in #1103708504142925824
either the darker blue one or the one with the little golden wing at the end
Welp the darker one will be my fb pfp thx btw
np
A woman, 55 years old, of Portuguese decent, empathic, from Northeast in us.
wot?
there's no bot in this channel, you can head to #1100170312106127410
nooo I was gonna gen and give the image
he must think you're the bot

🤣
lmao
@lyric zealot A woman, 55 years old, of Portuguese decent, empathic, from Northeast in us.
Error
there's no way she's 55
sad
here have weird hair instead ig
hahah

power of botox
yup, but i have some quality prompts, and also lora
I’m sorry.
was too lazy to remove those
wee, new model is up 😄
A senior citizen


is it perfect ?
make another new model
or go look at this freeu thingy
https://github.com/ChenyangSi/FreeU basically this
and implement it into auto1111
but that isn't eface's model
It must be the perfect model for him
made by the one
the model maniac
you're calling my model inferior?

aren't there a lot of already really good anime sdxl models already?
the fact that she look confortable sitting like that?

maybe, just kinda weird
she's on stage for anime got talent show
ugh. the lethargy is real 😦
no more highres upscales for a bit
yeah, show them how computers should really be used!
pro gaming chair
then i'll just uhh... do songmaster prompting again ^^
same prompt, seed diff model which do you think is better?
i feel 2 is more creative -> that'd be the one i'd go with
although the lines are a bit more defined on 1
.>
Did you prompt for flowers and stuff
no
well I might've
it's an old prompt with like 300+ token
🤣
yeah just check there's flower in there
so model 1 literally just overlooked that part
We love the wings.
We appreciate this lady.

What adds the hearts?
i wanted heart-shaped pupils
but then that happened

you can try spoken heart as well
omg, regional prompter is so good 😮
This is with 3 regions
still trying to get the hang of it
I just use base prompt + mask with 3 regions


forest -> red leaves, dress -> purple summer dress
it's really good
however, the original prompt is a winter landscape
so it can be used to isolate conflicting elements
yes?
it's ... tricky and finicky
however, it allows for good compositional value, i suppose
added yellow demonic smile one the moon -> basically, it prompts teh region of the moon with that part, however, as it's attention/region prompting, it easily bleeds over into her face 😦
AI: Feet = head
Accidental 64 width gen, still nice
even makes you able to isolate WHERE the fucking wings should be 😮
instead of having winged trees lol
Gib your research progress
chibi anime ADDCOMM girl with a dress in winter, walking through a forest at night, sky , moon BREAK yellow demonic smile, moon BREAK angelic, girl BREAK (red leaves:1.5) winter, forest BREAK purple sexy summer, little chibi sticker, dress
so, basically, what i'm now trying to figure out, is threshhold
i think it's the value WHEN it starts calculating areas in relation to the prompt
What's ADDCOMM?
is it a XL thing?
add common prompt
Regional prompter extension term
so basically, ALL the individual elements get chibi anime added
yeah, regional prompter
it's pretty complicated, but i think once you're used to it, you won't be able to go without it
Never tried regional prompting
About same?
the target word should be shirt right?
or am i doing something wrong
















