Yeee, thers 2 versions, V3 has a less flexible face style, which can be good if its the kind of face you want 
https://civitai.com/models/123997
#🍥|anime
1 messages · Page 137 of 1
the lora enjoyer smile
those previews
Do you see multiple? I only see one 
Also, fair
And also that cat is so cute
No, of course I clicked to see more

Fair 
also would be a problem if it showed more 
I mean, that preview is pretty safe :o
I can not with normal hires
only with controlnet tiles
but its not as great 
takes like 4 minutes for 512x768*2
idk depends on the denoise
O, das fair 
Looks all the same
Usually thighs just increase with weight to the prompt, adding "curvy", "voluptuous", "wide hips", and "plump"
wtf
it has a vae?
do I just put lycoris in the lora folder in latest a1111?
I did a dumb, and forgot how exponential scaling works. I did a prompt matrix with a bunch of artists 💀
seems like it must've taken forever
Lmao

probably. I would have stopped it had I been at my PC. I had started it and then went to go do something and was away for a few hours and came home to PNG that crashes any image view I've tried to load it in. I have to open it with an image editor to look at it.
It's completely useless considering the comparison was on WD1.2 and it's so big that it's tough to even find the labels if you do manage to find an image that you want to check
I only keep it around for the novelty of having a 1gb PNG
33152x25057px
im hungry now
I've deleted more than I have saved btw 
O ignore the VAE, and yee LyCoris goes in the same folder
trying the niji style
O, cute pic :o
Gone to work, have a great day/night
any models that is very sensitive to loras aside from anylora and aurora?
I need to learn how to use that
I need too! too many things in this program 
I had zero idea you could use tiles like that
its awesome
you tried with like 0.35 or 0.45 denoising?
controlnet reference is actually solid
not as good as using DALLE but
I can use a dalle composition and get kinda close
without censorship
reference on the right
you do it in img2img tab ?
no, highresfix
use like 4x-animesharp at 0.35 or 0.45 for highres fix 2x, or 0.6 or 0.7 if you use a latent upscaler
txt2img, how I made this last bunny
do you need hires ?
I cant use hires just smol image 
yes
life would be easier if I wasn't using a 6 years old amd gpu
at the cost of speed
lower the tile size until it does
until it does
default tile size is too high
I set to 1024
you might need lower
dang
with that shite of a pc you're better off renting a gpu from google colab or something
:<
damn I guess tiled vae would've been your only hope with 4gb vram
Ello

How is everyone
in a constant state of trantiveness
@stable crane kinda of worked 
what is that?
CN tiles
yeah but what is that reference image?
ahh
I've only used CN Tile without reference image in img2img (it uses the img2img img)
nice
I'm on text2image
reference only model didn't quite do the tile effect idk 
can't test much 
The first one looks dope
I'm trying to gen 2 girls together, one with elf ears and other not
pretty hard it seems
I just drew hair over the ear for the one I didn't want lmao
@oak spoke
hello tomato man
hello catto man
That SDXL? 
i like it
it's cool
sutoy
Good morning everyone! sorry to again have to be nsfw policing here. As @wooden coral communicated a few hours ago, this cannot keep happening here and the next steps would be to timeout or permanently ban infringing users. Please keep the channel safe for work, horny anime is not allowed.
When u ban people u need to tell them to go to horny jail
whomstve was it
provide examples
(I'm a visual learner)
Read the first pin for clear rules, I dont think this needs examples
is a bikini sfw
I think probably if it's super skimpy and small, may aswell be wearing nothing
(I'm not a mod dont just take my word for it)
If in doubt, don't post it
is clevage sfw,or skintight clothes,or denim shorts
just post more nuns or burqa 👍 🙏

🙏
half of your stuff 😂
I'm fearful of Ara Ara species
I can't be the only one? They will wipe out the male sex
it turned more 3d than i wanted
OMG WHAT DID I MISS?!
Go read the message that I replied to to see what I missed
Yeah I saw that lmao, just didn't want to ping a mod or anything like that
I didn’t ping them u can reply to messages without the ping see
Also smol sus love the art
I've only been here a short time but the sheer amount of references to people posting NSFW is quite shocking lmao, it's almost an every day occurrence at this rate lmao


I can definitely see why the NSFW police have such a tough job
Yeah
Really weird those guys posting
stuff
well, it's really the reason why mods and even old regulars will tell people to watch out for nsfw stuff
since this is a verified community server
When you ask for a cute zombie girl bit stable diffusion gives you a new nightmare
Cute or zombie choose one
Why not both?
I love cute horror
Strange question but you guys have any ideas of what I should do for a new D&D Campaign? I need to figure this out before Wednesday night
Maybe some gen art could spark some ideas
best i got for cute zombie is this
I love this so much
The face shape is a little off but meh
when your anime girl turns 3d
tell me if this is too much thigh, but seems sfw to me
hmm what would:
Hyper-Dimensional jar contains the pegasus Galaxy as a lantern carried by a hunch back man with a fishing rod baited with a demonic ugly one eyed imp with a lazy tongue holding a mouth at its end smiling at viewer
become
wait
there
Absolute Territory
As a, by a, with a (*3)
Lmao
Same lady, brand new profession?
Cuute
"JOJOOOOO REEEEEEEEEE" vibes
How is she so adorably expressive?
Or is she expressively adorable?
have anyone did training of embeddings? - that model training? - you put pictures in a folder and gathers info from that?
she's both 
i think there's still people doing embeddings training, but most would probably be training lora instead
What's her secret?
Is it the hair products she uses, or her upbringing?
She is strong to carry herself so well.
so i am getting this error - IndexError: list index out of range
what do you mean by that?
you meant pnginfo?
cant generate anything

text2img
gets to 55% and it says error
A: 4.17 GB, R: 6.63 GB, Sys: 8.9/15.9961 GB (55.4%)
previously was it working?
yeap
and if so, what did you do that might have made it not work
installed extension?
best way to debug, is to restart webui without extension
this was the result
why is your hires steps 1
yup
it's because you have hires step at 1
so your denoising strength at 0.56
i agant put hires at 1 but denoising at 0,7 but it works now
it reintroduce noise into your image for hires fix, but only 1 step for hires fix, so that's why it will be like that
there's not enough steps to clean up the "reintroduced" noise
hires steps should just not be 1
yeap saw what came out of it
a mess
maybe you can print it and call it art
and sell it for coupple of millions
doesn't make sense to introduce noise to the "somewhat" completed image, and then clean it up with a single step

want me to send you in dm the result
sure
hey, isn't this lycoris recoil? almost looks like it
except for the uniforms
what lora do?
https://www.youtube.com/shorts/e-rSc7lAvXE James Lee Style glasses & teeth, and darkness
Follow my Twitter: https://twitter.com/JamesLee03
Follow my Instagram: https://www.instagram.com/jamesleeanimation
Email: mail@jameslee.art
Stay tuned.
Music provided by ExtremeMusic.com
Make it Pain by Dave Randel and Bruce Fingers
#NOX #Shorts
Moo!
very cool ^^
Yeee, its a pretty unique thing 
Did you upload the lora
I'm gonna post Shrek with it of course
There must be a Shrek gen in all posts
Holy shit this looks so sick
If I can fix the hands I may change this to my lock screen for my phone 
that one is mighty cool
those are some glowy eyes
Technically, kinda the same
is this straight txt2img?
yesh. no lora, embed, or cn
that looks like it has some toonyou in it
it's Kakigori_v2.5 which is in fact based on toonyou-jp. good eye.
I like v1 style better, but anatomy is way better on v2. v2.5 is just something I'm still messing with
Miku is Magical, and she grabbed a gun.
mine didn't come out the same. hmm let me try that again
creates art while watching "Reign of the Seven Spellblades"
hmm think i'll make soemthing based on that
try adding best quality, and good lighting if you wish, and see how it comes out.
it's IP-Adapter Plus (SDXL), prompt is the good old "cute anime girl"
hello good people of the anime type
This is a cool pic.
ima shoot your ball
what ball

i like archery now 
huh
^.^ 
now drop and gimme 20
now miku like this
Powerful archery and faith in the horses.
just gen another amd 4gb gpu 
weakest amd user
👍
Rtx 7000

@worthy delta what do you bring today to this chat
Hopefully no more bald
Ayo
a batty joker
"hey wonder bozo wanna fight?"
bats is impressed
there now thats anime
hi frens
anyone use instagram
@lyric zealot

i tried going for a prushka look
Again, a zombie that is a bit too on the realistic side 
Zombies = exp

I'm trying to make a cute zombie girl, bit all I have is a cute zombie girl with semi realistic rotting flesh
I think im about to get one 🌟
best i got, some exp fodder for your knight
@steel arrow I'm sending my robocopter to deal with her
Sad
I would show you the slightly more aggressive zombies I got
But maybe ill keep my place in the server
a robocopter....
a zombie hoard with one absolutely JACKED zombie on the left
Your robocopter doesn't stand a chance
cool
ferocious
she is ready to fight zombies
zombie knight has arrived
This beautifully trustworthy girl is back.
Hoi!
I have been summoned
images aren't loading for me aaaaaaaaa
Wait there it goes

D'awwwwwww
What version is this?
v2
Damn my eyes are sharp
Frfr
lol
Tho like the differences between v2 and v2.5 are very big
Relatively speaking
Usually when I see models with various versions, all the versions can barely be identified without actually trying
Yeah, 2.5 is meant to be an intermediate mix to get me to where I want to be with v3, hence why it's not on Civ. The goal with that one was to reduce the bias towards young looking characters and try to resolve some of the background artifacts that vanilla introduced in v2. It did both of those things, but also backslid on hands and made the style much closer to the typical anime models.
how did you do this?
prompt it 😄 just trying to mix models, and if i get noteworthy output, i put it here 🙂
man, i love adetailer for.. reasons
i love the style and the colors. reminds so much of tales of
basically, on a generic sdxl model [watercolor [anime::0.25] | <the actual prompt>]
lots of testing and trying merging recipes
tried something similar on my 1.5 model, pretty good enough
very doable on sd1.5 models 🙂
@gusty palm you 3d modeling?
yus
i got into 3d modeling b4 i got into ai
i have a huge sign of relief
i applied for disability and the forms they sent me were due on the oct 5th, but i only got them in the mail on like the 4th
so i called and told them and they gave me til the end of the month to send them in
i thought for sure they were gonna ignore me and just be like 'damn that sucks for you'
just standard gov things?
aight, time to hurt my model even harder 😮
always when i find a "decent mix" i start putting on the hurt for real, to see if it bugs out in a bad or good way
https://www.giveawayoftheday.com/filter-forge-9-0/#
im gonna try this , been looking for something like this
wats your model?
i'm trying to make a v1.2 of brightprotonuke https://civitai.com/models/148871
reduce graphical artifcats even more, increase default level of detail, increase understanding
but it's been pretty hard
so, i generate, generate, generate. inspect up close to see if anything dumb happens (instant fail condition)
^^ much appreciated ^^
hebteb? 😮
the hentia one
they've been pretty good for me ^^ the hentai one being the fluke that got me into model mixing lol
hmmm... this model mix require some more testing, so i can fail it somewhere else 😮
adetailer gone wwild?
i hate how adetailer makes it feel like there's multiple resolutions going on (see that the pixel/line density differs when the whole image is stitched back together)
some of these are cool ig. i thought it had animated filters but it doesnt
it's future proof -> now ai can't recognize her
i downloaded a censor lora


it's the first time i have to use lora at like weight 2
or even 3

heh.
got even better at that
Blessed and enlightened, she is.
METADATA
first try =]
Literally
You officially joining this channel and becoming an anime man
I've turned the no into ramen
tf
this is what that looks like without anime in the prompt
also, if y'all where making fun of ComfyUI nodes being confusing, there is a frontend made by SAI that uses ComfyUI with a workflow loaded as a backend and it looks more simple than A1111
I just don't mind the mess, but this UI made by SAI is allowing to load any ComfyUI workflow and have all its optimizations running:
I usually use SDXL, but this is about the speed ComfyUI gets for me on SD1.5 with all the modern optimizations working
This is powerful.
this is neat, haven't seen this before
I mean, I have established workflows in Auto i've come to just know, but, that's neat.
oh are we doing kissy pictures?
ok then
maybe ill give that a whirl
stable swarm ?
I wanted to try that one
probably won't work
but still
guys, there have been not only multiple reports but also multiple warnings from mods or staff, from this point onwards you either follow the https://discord.com/channels/1002292111942635562/1040299965647437905 or get put in the horny jail for 1h, 24h or get banned. we've been cool and let slide more than we should, we have explained that discord is getting stricter. so there's that.
feel free to share your thirsty creations via DMs but not on the public channel


nice lighting! specially the backlit ones
interesting lighting source

worst timeline someone help
I give up on dalle
dall-e 3 is besttttt
so censored, I don't even know whats wrong with my prompt sometimes and it says the image is unsafe somehow
that's because of the resulting image, u can keep genning on a soft prompt block
and you'll get images
they censored it even more in the past days
yea its fine if u only create normal anime faces
it doesn't even have problems with 4chan stuff
openai is full of bs censorship
or using feet in prompt
now create an anime girl with a deformed demon face
i've done tons of demon girls
you can't appreciate beauty of girls that's considered harmful what a garbage policy for Ai

i stopped using chatgpt trash months ago
bruh
i prefer google bard more for it's openness
I removed cute from the prompt and it worked
flips table
like i said, prompt engineering
once you learn it, you can make lots of good things with it
luckily for chatgpt theres a bunch of uncensored ones on poe
also feel free to send them feedback, they do act on feedback.

i sent tons of anime related feedback
dont feel like engaging with those ppl on openai
poe ?
not oai, ms themselves
ms is the one putting the filter, just hit the feedback button, they do read all of it
MS is garbo
i disagree, they act pretty fast on feedback /shrug
Anything I tried to use from microsoft lately has been trash
i disagree they garbo
Idk I just started to hate anything microsoft
your copy of windows has been revoked
sometimes get like one of these a dayi get tons of 'we're working on your feedback'
based on stuf fi send in
i send a lotttaaa comments in
didn't toyed sofar with dalle 3 how would the promps fornit Look Like for what hq Anime picture
try generating an image in dalle3 with navel
that one is easy the ones i had trouble with are deformed horror demons eating everyday objects
a lot of my prompts r demon based
they come through fine but no examples rn
probs closest i got
for now
Sitze of Image also Limited ?
1024x1024 only


try with bikini
Lol

I think its a skill issue
So much blue
i dont see your point to defende dalle 3 ...its basically censored to filter out a lot of prompts
the filter isn't good, i'm not defending the filter
but the quality is great and you can prompt engineer around filter
SD quality are great too
Aslong U dont pay when stuff gets filtered 
took me ages to learn how to prompt elves without prompt blocking
now i can do elves nonstop
in like any setting
when before elves kept getting hit lol
why is elves blocked too
now do sexy shirtless donald trump pic
they aren't, just gotta prompt engineer
xD
i still use elven girl/elf girl
Do some humanized samirei Cat
but i add stuff later in the prompt

i defend dall-e 3 because it's such a huge leap from the next best thing
if you use tons of lora/style models then you could probably get something closer to an exact idea you have
btu dall-e 3 as a general/anime model is best i've seen
censor def sucks but it makes great things that surpass anything i can do in sdxl atm
yes its better only downside its censored and not local
just never use the word "tesseract" in a dall-e prompt
so for us who create lewds 1.5 still beats them all even if we have to gen a lot of images
well yes, but thats not to say SD cant catch up, and as far as i've seen openai has always been heavily censored from the start, that a big negative for many
constant ever-shifting competition, once SD catches up it may be better for a while but then dall-e is constantly being worked on too. the one thing i will say though is paint cocreator sucks.
paint cocreator uses dall-e 2.5 for some reason lmao
so much for being up to date ;_; but i think dall-e 3 is still preview/being worked on too
ye. and sdxl was a huge leap over dall-e 2
but then dall-e 3 is like multiple leaps over everything else so far
but its true SD doesn't have as much financial means like openai and ms
SDXL is a flop
i had a lot of good gens with sdxl once i learned to prompt engineer it & use the diff clips
sdxl vs dalle3 war of the censored models
SDXL can't compare with SD 1.5 in terms of accessiblity
once sd is in the lead again i instantly go back to it
SD for 4gb AMD cards when?
oh i have no issue w/ that, 24gb card here
You we don't count
but sd will only get more demanding i'd assume, as it gets to be a bigger model
SDXL bumped onto some tech limitations, hopefully when they work out those limitations the next version of SD could be more usable
not really
i'm pretty sure you can use it efficiently on an 11GB card
i even had multiple models & passthroughs
like passing sdxl+refined image through a finetuned anime model
the rendering takes too long, you need to upgrade your gear in most cases
render time seemed fine to me
i think it was like 15 seconds per fully stepped image for me
thats cause you have 24gb card, thats not the general user base
¯_(ツ)_/¯
i mean, not too long ago AI was only enthusiasts so it made sense to have enthusiast hardware
i started sd on a titan rtx then moved to a 3090
it's only been like a year since then, stuff like anime GANs needed high-end hardware
8gb should be fair standards altho many would disagree, it all comes to accessibility when running locally
i think it's fair to aim for 11-16GB at this point, push models to their limits and expand creativity/accuracy
i want something dall-e 3 level but local
18gb is probs limit since i'd blue screen if i had anything else open lol
That looks like Homura Akemi.
originally i was going for a ran mouri look but then dropped all the descriptors and let it make its own char
I dont Use XL at all
sdxl is good for the basic skeleton of the picture, then pass it through an anime model to finish it
sdxl is way more varied with expressions/pic design/creativity
also better with fingers & such than sd 1.5
but it doesn't do anime well unless you hyper fine-tune a prompt for it, so need a proper anime model to finish it

You'll get bonked
my hand hurts
my hand also hurts from playing too much disgaea
so went back to ffxiv today instead to take a rest from controller
These gifs don't work on my phone 
Phone issue
Maybe my discord is just extremely out of date lmao
smallest rifle
Yes
Isn't that the name of the girl
With blonde hair
(I should know where she is from but idk)

samsus
Xl isn't censored though
Try just Samus, blonde hair, blue suit
See if your model knows

it is for porn
You'll want to use Samus Aran
Nubby knows everything
that prompt kinda sus
But with guns
👍
ahh ok
let me try
it isn't though. It could certainly use more finetuning training if NSFW is what you want, but that does not mean that it's censored.
i didnt use blonde as a prompt but the image got it right
samus aran in blue suit with a big gun in her hand
Maybe I've sent this before, anyway more box fox
yeah, samus is pretty popular so she shouldn't be difficult. add zero suit to get her blue bodysuit more accurate
(giga giant sniper rifle of destruction:2)
u are probably talking about nudity while prn refers to the hardcore stuff
inb4 ban

It's still not censored, just not trained on that. None of the base models have been. Censored would mean that the ability to generate it has actively been interfered with. Nothing is stoping anyone from training whatever they want into SDXL besides the fact that finetuning it has proven to be rather difficult.
Yes

Anyways, is it because of something other than the model being heavier
That is some doom/halo weapon x3
no money 💸 😔
training the text encoders is difficult which can make certain types of training harder, it seems to have a much harder time with especially large datasets which makes the scale of training we need for a good NovelAI equivalent anime model to be trained extra hard, and on top of that it's heavier so a smaller group of people are able to do large scale finetunes.
(ultra mega wet and tight latex suit : 7.5)🤠
Absolutely love Samus and the metroid series
here is your image
too political 💀
could this possibly work as a prompt? 1girl, {blonde|brunette|red head} or does that only work with a script?
i think u need the dynamic prompts extension
yea that one
ok thanks
Serious Sam Waifu
sweet, this is working out nicely thanks man
10/10
So are you back?
AI hands 💢
from what
I still have the same bad gpu that takes a million minutes for anything
@nova remnant a woman with {blonde|brown|red|silver} hair, wearing t-shirt, jacket, mini skirt, standing in a garden full of flowers

Can you make these be chosen randomly ?
not random anymore, now that im using dynamic prompt extension
also this extension is not heavy on vram at all
fun thing to do ...
a woman with {blonde|brown|red} hair in her living room playing {guitar|banjo|ukele}
This is IMO a better alternative to the Dynamic-Prompts extension. It comes with a nice collection of wildcard files, dynamically changing several of the generation parameters, and doesn't have all of the bloated crap that you'll probably never use that's in the Dynamic-Prompts extension. I haven't used Dynamic-Prompts in months, but when I did, I frequently had compatibility issues with other extensions, I've never seen a compatibility issue with Umi-Ai-Wildcards. https://github.com/Tsukreya/Umi-AI-Wildcards
interesting, thanks for sharing, going to take a look at it 🙂
help me i dont understand words what does the (:1.7) mean in all thiss
1.7 strength for my petition
:p
ah lol
Did you inpaint to put the clothes or to remove
i added clothes
looks very clean
Would hang around in the creepy forest with
And do normal things people in creepy forests do
howto
how to get ai photos
#1100170312106127410 bots in this discord can do
Or #1072220168534642768 probably leads you on how to do it on your PC
She's still quite expressive as usual.
Can you do old weapons
Like a musket of sorts
And make it in that style
Very gun gun
idk what gun is this 
dont come to the town tomorrow 
regional prompting 😄

how do you get such clean miku's? the model knows it well?
Yep. Kakigori can pretty reliably make most characters with 10k+ images on danbooru. It can do some with as little as ~5k.
Have you tried Region Prompt Control inside Tiled Diffusion? That's the regional method that I like best
i never use tiled diffusion
i just like region prompt, prompting mode because it's auto-masking
tinfoil hat girl?






