#🍥|anime
1 messages · Page 124 of 1
it will if u can train it with 5 million images or w/e millions nai had
dont remember if it was 5 or 17 million

The term "lady" carries a tone of royalty in some situations. That may be playing into it.
like blender 3d stuff the requirements went down as quality went up i think xl will be replaced with something hopefully smaller but better
otherwise creation is just going to get more and more exclusive like only corporations will be able to run it
Little elves 
Prodia is good, Patience.AI gives me more options but it's slow and not always reliable
At least in Free Mode

whats your gpu temp?
what the 
everything is fine
Idk, not doing too much and it's usually around 70c
70-75 is just normal when generating id say u safe 👍
30 is when gpu is not in use
Isn't 70c very hot

hmm
Nice

What model is that
Hmm kizukiai's model
It's very versatile it can look many different ways

bad hands
Battle elf!
She's drunk

I'll prompt (drunk:1.2)
I'm curious xd
These don't have drunk in it
It's just elf sitting and some other simple stuff
Naturally looking drunk elf xd

Drunk :b
Incredible thing
Nothing came out nsfw
Without the nsfw negative

I think it has blush
I stopped generating
Too hot in my room, my fan died


I mean the room fan
Thic
stop blushing
@nova remnant the last of this elf endeavors

Time to sleep
I'm homeless inside home
Brought my bed mat to the living room because it's too hot otherwise, there is a ceiling fan here 
Who is this lady?
Meh, why do I have to be in office, now I can't generate waifu
Or anything for that matter
Hmm, im gonna prompt that when I get home, seems like it could work
Mom, get the camera, new Vil's OC dropped
I hate this character
Ah yes new OC 1girl, cowboy shot, something
the nice 1girl OC 
Zelda looking different
(blush:3.5)


Thanks For Watching!
Please Like & Subscribe!
Fun Fact:
In 1992, a shipping crate containing 28,000 rubber duckies fell overboard. They washed up around the world for the next 20 years.
Song:
Radwimps - Grand Escape feat. Toko Miura
Movie Version:
- https://www.youtube.com/watch?v=WwyDpKXG83A
Original Version: - https://www.youtube.com/watch?v...
i skipped to the best part

(steven seagal : 3.5)
https://youtu.be/YAXTn0E-Zgo?t=277 literally this part
Thanks For Watching!
Please Like & Subscribe!
Fun Fact:
In 1992, a shipping crate containing 28,000 rubber duckies fell overboard. They washed up around the world for the next 20 years.
Song:
Radwimps - Grand Escape feat. Toko Miura
Movie Version:
- https://www.youtube.com/watch?v=WwyDpKXG83A
Original Version: - https://www.youtube.com/watch?v...
Broo

I'm big brain
?
cheating
Put them together
hahaha
👍
i was using the lora
good guy main character who falls in love with girl with powers


there you go
yup, i think you need to for a little
by author
hires
weird gpu I'm using too
I just clicked a bunch of times to downgrade from 768 cause it screams for more vram :p
when in doubt 512x512
best extension ever

webui aspect ratio helper

The other day I was generating "girl, lying down on the road, happy"
I mean, if equestrian basketball is a thing.
How many road signs do you want?
Yes.
that looks wild
she slipping the crowd?
same, but i'm using some lora so maybe it helps
@lyric zealot whaat model do you use
elf friend 
currently I'm using calicomix_v75
and welcome back to SD
thank u
let me try that
i think SD is just bad at wide res
slight magical(?)
her right hand 
does a prompt like
"(cloud) in the background"
work ?
o:
letters reminded me of this lmao

@lyric zealot Think I got something that can work o:
I did some paint job
to make things make sense
trying adetailer
@lyric zealot look at these beautiful smudge of clouds

experimenting 
Thats pretty cool :O
it's working, maybe?
I'll img2img to get something maybe o:
how long does it take to generate something like this? 

1:30 is enough for me to generate like 2 pic of 1024x1024
are you using sdxl model?

oh, is there animatediff node for comfyui?
ahhh, then maybe I can run that at a lower time as well
where do you even browse for new node? civitai?
I don't use comfyui a whole lot

with that you can browse custom nodes
okay got it
and search and stuff
but you can also just clone this into custom nodes : https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved
and you can drop this into custom nodes and run it to easily update all extensions

anyway, you were asking for prompt, so here
(magical, aura, dark, ritual circle, magic, corruption, energy, spiraling dark aura, portal:1.2),
full body stuff can be wonky but ill run it
i mean, change stuff so that it works
would be easier if my controlnet was working 
kinda cursed but really cool nonetheless
maybe go upperbody instead
Feel free to submit more 
I can also try a different animation model

send a fitting prompt 
what is the prompt for that ?
I mean what prompt did you use for that image/gif there
Very good question uhm
perfect image, best quality, watercolor, 1girl, (solo:1.3), kimono, flowers and petals, beautiful concept illustration, (detailed background:1.1), (white hair, long hair, hime cut, blunt bangs, azure eyes, pointy ears:1.1)
I think it was this
I liked the non upscaled version more here
about 90s runtime
Small images tend to look blurry
perfect image, best quality, watercolor, 1girl, (solo:1.3), Chinese clothes, lycoris flower, beautiful concept illustration, (detailed background:1.1), (black hair, long hair, brown eyes:1.1)
@wooden coral 
metadata

What's hilarious is that I can do 512x512 upscaled by 2 and 1024x1024 no upscale, but instantly OOM if I go any higher
What is happening to her clothes lol
yes
Lmao
gif things
I still haven't played with that yet
I've only been using Stable Diffusion for about a week now
A little over a week*
I've practically been spending all day, every day on it
o:
Smooth
Is this different prompt?
I need to install regional prompts when I get back home
.o
yes you do. it's awesome!
@wispy canopy 
why the sadcat? 🤗
Does sdxl recognize some games ?
probably
however, it's milage may vary
for example, i tried atziri, but i didn't get 4booba waifu
@wispy canopy (large breasts:4.6) 
Mutation
So how did she find the hat?

Guys, does anyone know if it is possible to generate on AMD video card?? because generation in AUTO1111 stops with an error of approximately 90%
hello
I am the 4gb vram amd user
it is possible
seems like you're using --xformers on your webui batch file ?
That is only for nvidia
yes im use it
Remove that and try again
i just delete it?
ye
30 seconds for a simple 512x512 image
wow ... i have a 8-10 min...
you might be using the nvidia version of A1111, it will be extremely slow
go in #🤝|tech-support and ask for help, I don't know how to fix anything 
oki-doki. where i can get version Auto1111 for AMD?
yes, I did everything from this article from the section for Win
ask in #🤝|tech-support >.<
thanks!
btw.. about xFormers for my CUDA card in WedUi i have a "xformers: 0.0.20"
bat file have a --xformers paramert
if iam start with this is --force-enable-xformers
a have a message module not installed?!
Hello am back just before going to sleep
Just wanted to ask if anyone here knows why the qr controlnet is so insanely popular?
Like can't people just use like existing controlnets
Like similar ones
Why you use that vae?
aand for good work xformers CUDA 11.3 must be installed?
Perhaps that is part of the reason why blurry
scratches his chin
distributed script on a1111?
hmmmmmmm
HMMMMMMMMMMMMMMMMMMMMMMMM
there's a extension -> distributed
which makes an a1111 instance send the prompt to other a1111 instances
are you thinking what i'm thinking?
still in jp, or back?
isn't it like 2 am or smth?
3am
Ok ok man
🙂
Gn
anywah, the idea i have is : distribute will now hit a cluster of A100's 😄
I mean there was an extension before that made it like so a ton of people hosted and people could use that hosting for free or something idk
omg it works 😮 total cheatcode 😮
xformers is for nvidia cards and wont work on AMD
whats your gpu? i can give you the right cmd args
i am have a one Pc with AMD and other Nvidea card. for AMD i am remove --xformers paramert.
it help. but i am ask about xformers for nVidea card
whats your nvidia gpu? and whats the question?
GeForce GTX 1660 6Gb
okay this gpu needs --medvram --xformers --no-half
if the webui shows xformers 0.0.20 thats good
now for version WebUI 1.6.0 i have this is
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:128```
I’m not sure that a 512x512 image generation time of more than a minute(~2min) is acceptable. or is this normal for my card?
without using aDetaler
does it take 1-2 minutes for a single 512x512, 20 steps ?
thats pretty normal for --lowvram --always-batch-cond-uncond --opt-split-attention
it would be a bit faster using --medvram instead
ok noe try start with this
what about this is --always-batch-cond-uncond --opt-split-attention
you can remove them if you use --medvram
oki-doki. try
wow it more faster
48 sec.
Nice 🙂
Applying attention optimization: xformers... done.
this means that acceleration is used for the CUDA card?
Yes, its optimized with xformers
strange, but when using Upscaler the generation process slowed down just terribly
yes, upscaling is very slow
doubling your resolution quadruples your generation time
and if you're close to running out of memory, it's even worse
no no with --lowVram my time HiRes.fix is 4-5 min
now it imposible slowwly!
perhaps it's the swapping in and out of layers (model) that's making it really slow
this is because you are running out of vram now. --lowvram is very slow, but it's faster than generating when you've run out of vram.
You have to much hires steps. Use 10-15
also, yes 30 steps is overkill for hires
Tiled Vae will also help to provent running out of vram.
remember it. now try use it
This will depend on the sampler. For any of the DPM++ samplers you shouldn't need more than about 20-25 low res steps.
Are 10-15 steps really enough for Upscaler?!
yep, since it's starting with the forms of the image, it doesn't need as much.
it just froze again
This model looks nice with hires

My gens with it are blobs of pixels
Also look dm
pat pat little amd fwen
Shope
oh sadge, it didn't embed
someone posted this to the Kakigori gallery on civitai

I was close to posting Shrek gens
👍
Every page has Shrek one way or the other 
that guy makes anything sound imposing
"MUST I DO EVERYTHING MYSELF?!"
love it when he shouts that
@nova remnant How's it going?

I'm fine like a hamburger
@nova remnant Good to talk via direct messages?
I think I will try a1111 to use adetailer
sometimes i get random fire like that and it goes away when i do campfire in negative prompt 
I'll give that a try.
aint no way merging models is actually this simple
If you want to get more specific it’ll let you fine tune but that’s pretty much the base
Like presets, block weight but idk what even goes on in there I usually just slide around until something looks good
ive always wanted to merge more than two together and this looks like you can (probably could some way with A1 but i never looked into it)
probably lose my mind over the ratios though 
I can try and explain what I’ve noticed over tons of messing around but it’s definitely not precise lol
I was trying to merge 3.0 and dreamshaper
To make the composition or style of dream shaper but flat colors and line art
Yeah, there are extensions but it’s pretty much just sliders as well for each “block”
It’s fun if you aren’t picky or precise about your model…but it’s not fun when you’ve got like 5 and you can’t pick because one interprets prompts better than the other
It's also a much better idea to only merge 2 models at a time otherwise it's really tough to know what the reason for changes are. So, merge 2 models and tweak settings until you get what you want from the merge, then merge the result of those 2 with another, repeat until you've merged all of the models you want.
yes i agree

also try and make note of when something goes right so you dont have to go back and mess with it again
comfy is nice bc if i drag in an image itll put the ratio it was generated with
wait, if you used the {brackets|to generate}, will it pull up the brackets or what was used in the bracket? so like if i had white, blue, red hair and two pictures showed up, one with blue and another white, if i toss the blue gen'd image into comfy, will it say blue hair? 
I like SuperMerger extension for A1111 because it allows merging in runtime so you can experiment without saving a bunch of models and it also saves the recipe for the merge as the model name in the PNG info for the images you gen with it, even if you aren't saving the merges.
im pretty sure comfy uses wildcards or something similar but thats all i know so id probably check with their server if you really wanna know
also yeah that sounds nice. would just depend on what youre using. a1111 has nice ui but if you cant spare the vram id use comfy
or just want it faster and youve got low-mid vram
agreed. comfy is nice for merging, I just don't really like using it to gen. for merging loras into models, comfy is especially nice.
also ex: a model i made. it probably needs more tweak before i upload but its just for personal use bc there are so many models out there already and i figure its not much different than a break/abyss mix
the weighting seems different like all the other numbers compared to A1 and comfy 
now i need to fiddle with this more 😭
yes weighting for comfyui is different if thats what youre saying
its probably impossible to get the exact image out of both it an a1111
i usually 1.4 everything with maybe more than 1 or 2 parenthesis but in comfy, 1.1 or even 1.2 usually gets it
prompt emphasis is handled differently. A1111 normalizes emphasis, comfy does not. so if you do word1, (word2:1.2), you actually get (word1:0.9), (word2:1.1) in A1111.
i dont like that at all lol
noise generation is also done differently and sampler parameters are likely different between the two so you won't get the same image even if you were to correct the emphasis.
make it be word1, (word2:1.2) 
A1111's method makes prompting significantly more forgiving.
If you put something at 1.5 emphasis in Comfy, it's probably going to break the image. A1111 will more than likely handle it fine though.
It should be a toggle in settings though so that people who know what they're doing and don't want that behavior could turn it off.
i did try this earlier today to see really, interesting (left is A1 and right is comfy)
there is a node that changes it to a1111 weights and also mean, length (idk the exact word) and etc. but its an extension
but usually people stick to one or the other for gens
because of that weirdness
i dont think comfy is worse at images obviouisly, its the same backend, its just different and you might just want to pick one ui and stick with it
idk, i like comfy so far and just started using it after using a1111 for so long
pretty cool how people can share workflows too
Comfy has a lot of cool things going for it, I've just never liked node based UIs. They certainly have their uses, but if given the choice I'd almost always rather have a traditional UI.
yeah. I don't think there's anything you can't do with either platform either. only thing I can think of is the controlnet "reference". but comfy being barebones back end, it's up to the user to figure out stuff like masks and after detailing. gives you more control
which is also why I think the arguing that was on Reddit for a while is weird, like just pick one lol. I don't think any one is better they're just different
how it takes a million years to figure out how to do cn in comfy and actually get it to work compared to hitting that dropdown in a1111 
or you get lucky and find someone who's generated an image with that workflow and just drop it in
yeah, controlnet and inpainting are definitely not strong points for comfy from an ease of use standpoint.
so I'm thinking of trying a1 again after months just because of adetailer etc
see if a 8gb GPU can do it
because I could try with comfy but ..... maybe when I have more time
Yeah I've done it before but iirc it crashed when I tried to hires fix it
which makes sense but then I went to comfy and could do it <30sec
oh then you might have to share your webui args 

I'm pretty sure she is not safe
"its just a cute mask I got from a store"

I'm not reading all of that, have a cute @lyric zealot 
@lyric zealot
don't mind that
I was trying to get clothes hanging up in the air, kinda when someone leaves them drying up I couldn't gen anything like that 
now lift her skirt

now bigger milkers
Is it horny time?
auto1 is just as fast as comfy but LORD the saturation lmao what happened
what in the curse is that
that's the horni train
judging by the green splotches, it looks like it didn't properly load your vae
@lyric zealot I think I found a hack
I stopped controlnet from upscaling midway through and the image was good lmao
wtf even happened here
that train can lead to really bad stations like "timeout plaza" and "banned entities" 😜
i turned vae off because counterfeit has one i think
controlnet upscaling? 
but it wouldve been desaturated with no vae
have you guys played around with the freeU Unet stuff?
@lyric zealot tiles model, only way I can do big resolution 
ahhhh
no what is it
I'm really only using multidiffusion, didn't even know controlnet have that as well
I've been using some settings I yoinked off someone else for most of my gens and it seems to help a bit. I haven't really messed around with the settings to figure out what they do though.
2 frames and my gpu crashes I bet
This is what I'm using
now this is one of the scenes, with different samplers and freeU values.. same seed and prompts
that is really cool
how do you do that with multidiffusion ? :o maybe it works for me
wait
Does it do by "tiles"
yup
show settings 
this
im using what nubby told me tho


basically, from what i can tell, use the highest settings you can, for your vram
so probably lesser than mine
that's Tiled VAE. I'm pretty sure you already tried it and it didn't work.
woohoo, lora is still burnt but at least webui is working lol
@lyric zealot
free

I'm curious
why does her hair get blue

I have no prompts for that
maybe for most bob cut, their undershade of the hair is usually blue
or at least in training data
some models do that for some reason
what you squishing? 
the pillow I swear
she has asthma so im doing a CPR
its round... white...

I just typed the numbers and got gudder images 
I just tested 1.4 reccommended settings and got a mess and decide to try again another day
This is powerful.
Powerful Booba
hi
wat this
This extension. How it works is largely a mystery to me. I just grabbed those values from a post from someone else and it has generally worked pretty well for me. https://github.com/ljleb/sd-webui-freeu.git
ooohh
some kind of magic quality enhancement magic
my water was shut off this morning and ive been dealing with that allday
i will install
this is an image
do u install it from the extension url page with https://github.com/ljleb/sd-webui-freeu
a1111 implementation of https://github.com/ChenyangSi/FreeU - GitHub - ljleb/sd-webui-freeu: a1111 implementation of https://github.com/ChenyangSi/FreeU
ok that seemd to work
any update on kakigori or any new models from u @supple raptor
I have an experimental version that I've been messing around with a bit. It looks fairly promising for improved composition. I've just been taking it slow this time around.

how ru alls
feels
that freeu thing definitely does something
ii couldnt get the other controlnet qr code model to do much.
qrCodeMonster_v20 im sure it worked for something i may try later some more
controlnetQRPatternQR_v2Sd15 works pretty consistently though
u have to generate at 4k or 8k for that
Lord of archives flexing on us
huh not sure how it did or why it did a pic in pic.
live peach reaction
no suppose to be princess peach with pink eye
😄
so yeah.. I got to take eyedrops in my eye. no fun and not easy to do
but at least I can look at the screen without cringing from the bright lights
o.. k then
Mario anthro fox
to many tails and wtf is that on the left?
ok this model is so fecking random. thats not what I put!
hmm almost?
lost plushie
Man I fr appreciate you. I felt kinda bad last night and this made me feel so much better
This one very good
😳😳😳
Is this even anime at this point
anyone tried the firefly text to image site?
Whazzat?
a couple more improvements to the software and we can get full fledged AI animations 😮
Im still exploring, people can already do way better than this 
distributed extension was the way to get it working: 4x A100 doing parallel inference lol
Damn
Utilizing your resources to their fullest
Just like how windows utilizes all your ram
Haha
nah, i've got some left
14x16 = 224 images at 1024x1024 in 9 minutes XD
aight, imma close those instances down tho
whoops, had to censor 2 images
5th image 
can't see any genitalia tho, so it should be ok
it's uhh.. very tight fitting fabric around her navel
looks skin-colored, that's the optical illusion
yeah. that's it.


Also, thicc mommy and skinny mommy
when you stumble into high level area on accident
the one piece is truly real 😮
really, stop entering those endgame areas when you just start the game!
i really wonder what happened in the dataset where there's top and bottom water
more people wandering into high level areas
brain filled with uhh... cats?
i suppose i should look into replacing the clip i suppose
i mean, getting my model #🍥|anime certified is a thing too
dunno how to do it, as ever, but... first imma need some inspiration
i couldnt eep
I'm blue ba da bee
oooa
why not just upscale in img2img
i actually use swinir to clean up the image at the same time
prompt and result are quite different
so that would mean i'd have to do 3-4 manual steps instead of evertyhing all at once
🤣
oo
it works for me tho

AI is very racist against every other skin
that's closer tho
i'm using different samplers too, perhaps that's the thing it doesn't like?
maybe (colored skin, red devil skin) maru or st
my first attempt at the red skin xD, didn't work out
maybe
or it could be due to sdxl model?
negative white skin black skin
it can completely change the whole image if it deems it needed to make the latents from the regional prompt fit
LMFAO it's making LGBQT demon

based
trying to get the treshhold right
the dark knight of rainbow
and else just fuggit, i suppose
interesting theme
just demon skin, or complete demon girl?
just the skin
that's awesome
it's hard to do any color other than brown or white on SDXL
maybe look for a model that features red skin
why do it the easy way
identity disorder or havoc
i wanna do it the hard way -> prompt for it 😄
I mean I have 1.5 model that can do any color
I'm tryna do it on SDXL
which's very racist atm

trv
AI need to stop doing these black face fr
imagine if it just give actual burn scar

well we'll see
sunburn did not show up
[cute anime:0.1] 
hey, the dude is really doing his best looking cute!
uh... sunburn? XD
... i temporarily give up
well now that's authentic slap mark
look good for different kind of thing
this model has made some disturbing generations of that kind...

















(oh and multiple moons)
