#🍥|anime
1 messages · Page 108 of 1
Ghost girl
studio ghibli rulez 🙂
It's been so long I wonder if anyone remembers me
I haven't touched this in almost 4 months
if you can run it, get sdxl based models, be addicted again
I will try it out, I got a week or so before I grind tft again
I can't do both unfortunately, it freezes my laptop if I try to gen while ingame
yeah, you need a very beefy gpu + mem to pull that off actually
Any extensions that you use on the regular? It's gonna take a bit to catch up on some must have addons
haven't been paying attention to that 😦
biggest release would be sdxl
but that's kinda divisive -> people who can run it, and people who cant
I just updated to 1.6 and thankfully the ui is still pretty similar
yeah, that's pretty much all the same
you can do double step refining now
so you can let another model take over generation
Oh thats cool, so you mean like a hires pass using another model? Or something different
not highres
for example, you do 15 steps on a specific model doing the initial layout, then finigh up the last 10 steps on another model you know has better quality
11
12
13
50
Hmm I think the most interesting is going to be looking through civit to see any cool stuff I may have missed
lots of new models and loras obviously
Thankfully they seem to have updated the searching and filters since I last used it
I took a looong looong break too
shit, don't make me get my... things from yesterday from the closet XD
we need more love and happiness!
Hey I think I remember you from back then
I think the mermaid pow
I'm just testing yesterday prompt with diff model lmao
yeah glad you remembered
wicked improvement
Yeah I think it was one of the first anime pow winners so it was memorable
I noticed pow took a break too 
i've been gone too for a bit though
Miss the Guizmoose 
Of course we remember you ma man
Did you hear of sdxl?
This good
I've heard of it but haven't gotten around to trying it
you should, it's capable of pretty nutty stuff 🙂
this android comes with a built-in solar panel
she dropped her bag to greet her friends lol
the sun, the moon, the waves the sky. the circle has everything!
Sad
you just awake?
and with this, i'm going to bed, someone do some gens for me to look at tomorrow ❤️ else we should rename this to my channel 😄
I mean isn't this already practically your channel lol
least egomaniac redditor
More ghost lady
Anotha one 
And goodmorning again
Gmorning Edad Eface 
are you calling me old? 😮
Fio and Eri I think
XL?
Cute :O
And another cute one :D
got some more in the pipeline
Also what is that model? :o
ah 😮
im using anmitsumimimi
now i see

Yeee, you told me yours a while back :o
now say that 10 times, REALLY fast
here's both with metadata
Funky name :O
i think it's really the lora tho
gpu goes BRRRRRRR
Whats the LoRA? :o
u guys are using loras?!
Also I'm still waiting for Aurora 2 
I make em xD
i'm still stuckin the prompting age lol
still, getting comparable results to lora 🙂 i think i'm doing just fine without them
is a bit trickier tho
i mean, you are nearly assured of the wanted output with a lora
aoi and taf mainly
ghost girls. very nice idea @magic wave
Taf? :o
Ayyy, glad you liked it 
Ohhh fair then xD
ghost?
Is this SDXL ?
yeah, sdxl based model -> protovision
been using it for the last week, really reignited my sd passion for a bit

@wispy canopy Can you give me the prompt for the image that has her face closer up ?
dunno if they fixed it, before you had to open in browser before
I couldn't just throw it in there, saved first to my pc
yeah, that's what i do
i have a folder called 'stolen sd images'
xD
sometimes, a bit of ❤️ is all you need
hmmm... finally got the part where i was lyrics prompting... about 40% done of my og stable diffusion folder output images
sdxl takes it to the next level for sure
prompting genius, doesn't even need a monitor ❤️
I'm on a rx 580
a minute for my simple 512 768
😮
(go find used 3090's in your area, that's what i did)
I'm from the depths of earth, there's no such things as buying tech stuff here that isn't 10x the price
Might get one for reasonable price from a friend that imports them
I've been trying to create a dataset of top notch gens from some merges(mainly models y'all recommended/use), almost contemplated using AWS
One person tried selling an RX 580 at Lamborghini prices, I understand your pain
and of course, where you're from and your disposable current income
i mean, i can afford a couple of used ones easy, but that's the pro of not having a wife xD
is this the real elsa?!
lets just throw elsa in for fun 😄
lol, here's the prompt with elsa in it 😄
@wispy canopy ghost lady prompt x3
i like the wisps ❤️
shh... i know
why's the rain so....sticky looking?

🥵
when you sneeze too hard
I wonder how good SDXL can make vehicles that isn't car
That one with rainbow blocks is really cool :o
bismuth 🙂
Makes sense :O
Did you do like a ruby, aquamarine, diamond, emerald, bismuth and like obsidian? :O
^^
ruby, saphire, quartz, jade, bismuth, onyx
Ay, I got two :D
diamond would be a good one too
jade is a often not carved in angles, so i thought it'd be better for wind theme (flowing, instead of angular)
bug girls
pretty sure that's not my prompt 😛
why u no have metadata? 😮
here's your prompt
that's a good prompt 😄
yea
created image, then fed that image to deepbooru
fixed it up again, and that's the actual prompt
interesting
just going through the old prompts
ah
got like 70% of my old prompting stash left lol
I go back to my 1.5 prompt a lot too
it just turn out that half of them just don't work anymore

make completely diff thing
for me they kinda work better
this one have metadata
oh fuck... where did this come from?!
ooo nice oil painting
Makes sense :o
have you tried make military vehicle yet?
Eva
like tank or fighter jet
accidental output from... couple days ago?
damn that actually look quite nice
except it still use primer as paint
clearly rushed production
🤣
xD
"WE COME IN PEACE... oops"
i had random locations for that prompt , and it selected tank interior
didn't get tank interior, but not dissapointed regardless
Choo choo
Thanks XD
Exactly that
Chu
I literally just bought this
Wanted to get the chu 2 originally but none was left in stock
What is even a Chu 
Those are cute :o
Very good
Rn just mainly going in Akihabara and stuff
True :o
I fundamentally have a thing against like busy cities, it just makes me feel bad and anxious being there surrounded by many busy people with busy lives. Though it is Akihabara so I had to acclimate myself to the situation
Though tomorrow I am gonna go to Sendai so that will hopefully be even better
u there solo, or with friends?
Only with my step brother
cool, having company makes trips a lot better i suppose
Yeah
I have a fren from the anime land, every product he sends pictures is like that
Yup x3
Ooo I haven't tried Restart, you think its the best one of the new samplers?
I'll have to try it :O
it slaps dpm2m around hard
Fair xD
10 steps with coherent images already close to the 50 steps output, is not bad in any metric
50 steps is like HUGE slow, but the amount of correctly resolved details is incredible
Restart makes the details more detailed and the coherent more coherent
At least from this short comparison
Fair fair :o
yeah, it really does, i mean look at the stuff i posted last couple of days... if sd launched with that detail... oh boy, midjourney would be already out of business
lol
but that's teh cost of... 1 year of r&d i suppose
Does Restart work with hi-res? :o
It do be like that
1 minute 55 secs for me with DPM 2M Karras, 40 steps normal, 20 hires, 40 adetailer at 768x768, and 2 minutes 36 secs with Restart
or sdxl slow i suppose
Its not too bad, seems kinda like a inbetween of SDE and those others
restart does add extra steps (it reverts after a certain amount of steps, adding extra steps)
but i like it a lot ❤️
Model (+LoRAs)?
Oh so it is like SDE?
very clean ❤️
actually, it's quite not like sde
if i read the paper correctly
it re-noises as part of the sampling process
Ayy thanks 
Oh, thats pretty interesting, guess the name makes sense 
Oh wow
i think that makes it so good at correcting issues
Just made me realize
where the initial layout has some errors, but resolving already too far gone to do something else, with restart it can re-attempt and fix those issues with more probable outcomes
Yeee probably 
Yeee :o
And also forcing model biases
More strongly
At least theoretically
On what GPU?

Am I right or am I right?
another test. it seems to resolve in 10 steps what dpm2m does in 25... with more detail to boot
This another one, lil Adetailer artifact tho
3060 actually
Close enough
12 GB, I don't think its the Ti version, but 12 GB
Then the normal version
thicc
Before it I had a 960 GTX, made a huge improvement xD
Oh shit
Yeee 
Yeee
I could do AI with it tho
Slowly, but I could
I think I had basically the bare minimum for AI
That one can run AI? :O
Oh thats fair
I remember I read lower versions had like issues because of the chipset
It I run on no half or something, I get limited to like 250 resolution lmao
To be able to inpaint
You need to get a new GPU :O
I want a 6800
you want nvidia for ai
Yeah
It's the price of a 3060
cuda being what it all the libraries being programmed aroudn
I play games mainly
AMD is bad for AI tho, real bad
And 3060 kinda... sounds cooler
XD
Surprised by how good the teeth are in tis, usually teeth are bad :o
Maybe by the time I get one Nvidia will have cards at the same price for me per performance :b
Maybe :o
This
Actually very sad
Nvidia is dominating
Also my dad does like stocks and stuff so he told me that their stocks are way too high rn because of that
I was told by the discord professionals that AMD just doesn't really care about AI on the consumer products
But their other cards are good or something
They have to like quadruple their sales or something for their stock prices to be justified
They care a bit
But it's pretty hard to catch up
Because Nvidia started with this shit before everyone
At least I think
Cuda is OP
i mean, in terms of indstrial gpu's i think nvidia's the only real player still left
there's no alternative to a100's or h100's
Tesla v100
Lmao
Also by nvidia
Still funny to me that Intel cards are a thing
Intel cards are not a thing
Sometimes I see them in graphs and looks like an invader
I choose to believe so
intel cards are sadly, not a thing 'enough'
they're VERY good for their price point
but that's the big caveat -> for their price point
i think their most expensive card was € 270 or zs?
Here they probably cost like 2000 bucks
Those images are connecting to each other and I'm going insane
I think restart also feels less "dynamic", is the best I can describe it
You get what I mean?
it's deterministic -> so it's already very close to "final" result from the get go
that's the prompt...
I mean
I feel like Restart
Kinda does what the name says
And like
Adapts to the pic you're generating basically
I'm too not smart to rate the difference between those images
low steps = faster generation
but at low steps, some samplers have artifacs, or resolve into completely different things
but restart shows a lot of right details from the low steps already
for example, if you look closely at the dpm++ 2m 10 steps image, the image looks smudgy at places
Small example
what's the part what you're not liking?
Like the clothes?
Looks more consistent
all the flowy things are artifacts not resolved, the higher steps on dpm2 also fix that... so i kinda not mind actually
This I think is in correlation with its coherence
Yeah I guessed
i do agree that at some level, those unresolved errors can make the images more interesting though
Yes
How is it on hands
And also I was also guessing before that restart might also make model biases stronger
Since now those "artifacts" are less likely
Like even if you use a model that is not really meant for a specific style you could get lucky
But with restart this luck is less ig
Though you should technically just use a fitting model
for me, it's just the best thing ever ^^
as i'm not working with anime models anyway
I kinda enjoy those artifacts sometimes ngl
so i can introduce glitches with prompt editing at any point
Hello I am inviting you to the Euler a team

#@$* Euler a
like switching to a completely different prompt all together at 50%
give restart a try, you'll like it
My card is too slow for anything else
euler A was good... 1.5 years ago
Hahaha
nah, you can get away with less steps on restart
Euler a is a classic tho
isn't euler more classic?
I mean it's still viable
the non "a" one
Idk I had both since the beginning of my usage
yea
1.5 years ago...
Such a difference in AI tech since then
yeah, it went crazy hard
back in my day we don't even have SD yet
man, when the first GAN's came out, everyone lost their minds
Yes yes
and now, those GAN's are just... sad in comparison
gigagan?
?
I saw that
basically a gan with more parameters, but still limited in the end, i thought
it just can't handle the versatility that SD brings
i think all this AI goodness happened in 2019, when the transformers became a reality
I mean isn't it cool interacting with "latent space" (I think you could call it that) in like a graph with points and just dragging/picking your point on the graph and getting an image or an output of some kind
I wonder what was it like
yeah, if you like having only like... a very finite amount of dimensions
You had like 12 different parameters that range from like 0 to 1 and then you get out an image
ah so kinda like slot machine
Ye
Images are 2 dimensions
lol
the gan latent space vs the sd latent space, those dimensions, you troll 😛
I troll
You got me
I mean though aren't these two different kinds of "latent spaces"?
Idk I am not an AI researcher
yeah, gan doesn't use latent space iirc
Need to ask over at projectai
that's the sd thing that makes it so versatile
Yeah but like it has that thing that represents the thing
Like in those graphs people like to showcase
Me when I do not know what I am talking about
Make anime AI this pls
combining all that is good 🙂
right - cats, girls, milkshake AND fingers, what more could you want?
my. cursed. lyrics.
CARs of course!
??
Still milkshake
I think she needs to focus on the road
Take away her milkshake
ok, this is it - cats, cat ears, girls, milk shakes, muscle cars!
Stareeeeeeee
Very good one
Mooscles
Poomp the mooscles
first for a more realistic car
Retro
Does she have a name?
ok, more muscular too (i have so many nice other results but they are not really SWF any more)
If you mean character I call her Yue
last one
Yes, also nice name
Oh dyslexia hit again and I read "She" as "This"
good name for a cat
Are you sure this is ai.
What
fisheater
yue is a good name for a cat
glasspusher
You have seen nothing yet
Idk man I don't have anything on me atm
cat are just vagrant sleep seekers!
Look at her fingers

I mean you could check out the latest things on my Instagram but I rarely upload there and it's all pretty old
Ik but damn. It's sorta hard to tell
Me when nai came out:
Im into the type of stuff @wispy canopy generates
I got gud fingers on the cat chibi
Schizo prompting?
lol jk
i'm trying to use comfy ui because is less performance hugry, but damn, how hard is to use
damn no one liked my stuff

Idk what that is but i love the backgrounds/landscaped he does
Wait is it?
Show Fr
I also love him them
literally scroll up by like 2 scroll lmao
Both
I love this.
another app to use ai models

@brittle jay is this the kind of catgirl you desire?
I didn't say "what is it" but rather "wait... is it?"
Orange
wdym why?
You can???
yep
How
Nothin im scared of insects that's all.
oh lmao
what how? setting the right nodes
I got more insect filled pic if you wanna see
I suggest Tryna add some boom shakalaka💀
What is it doing that allows you such high res with such low vram?
bugs
swampcat
oh, you can set just the nodes that you need, working on a barebone workload, reducing performance issues
Can you try to make it 2D like how smile does?
Can you screenshot node setup for 2k by 2k that works on 4 vram?
but like this, you need to know what do you need to get a good job, or you get images burned pretty easily
it already is 2D tho
Sorta like a manwha style
Luna
this is already hard...
Make it more 2D like naruto Fr
Luna what
That's smile's name
or at wort, you get crappy results
Hard?
Oh make sense ty
i'm not using an anime model
Can someone please link @glacial bane like a good starter anime model or a model they are using?
if no one minds though. i just use anything 4.5
Grindset
Guess
Sounds like an only girl's type
No
Ill check the pic's
Is base wd even that good?
the 1.4 is the newest one now
the earlier version was pretty mid
1.4 is decent and easy to use I think
except if your name is eface, then you can use anything but anime models to make anime lol
🤣
i hate anything that have to do with anything, that model traumatized me
average
Ye i really do need to change.
True and real
you can feel my pain, bro 
What's Luna's model? Or does he/she use different models at the same time
I feel like that'll suit me FR
Aurora
Oh wait that's the other luna
I think?
Just go to projectai and copy people's models
let's try to upscale from 2045 to 4090 lol
I just pick models that I like and merge them
Smile's
First they need to pick the models they like
let's see how long it takeds to bake a 4k
Takes Atleast few hours.
True and real
these were impegnative days
Which Luna are you
Not bad for anything 4.5?
Had to re-read this a couple of times
The second Luna right?
Same
I'm the 2nd one
Oh ok
yeah, eng is not my strong point
WELL
Why can't we call the 2nd belle?
And
i suppose i whould have used busy days
Pregananant
And I am crazyluna
well, would have been a cool way to describe this period lol
The first one of these is a LoRA I trained that is not published, but I can publish it if you want @glacial bane
Whats does greenucorm do?
Oh wait
Maybe I did publish it
Lemme check
O ye I did
Aurora + Kenshi creator is first luna
Its a style
oh, can't post this one...
https://civitai.com/models/70635/cyancapsule-style-loconlycoris This is my Cyancapsule one

wow, milk
No thx im just curious (in all honesty I'm absolutely obsessed with your generating including the one i sent)
Well, its already published
I just forgor I published it
xD
Oh welp im installing it
You know how to use lycoris and what the differences are?
Haha .no.
I reccomend the other 3 LoRAs too tho, and any LoRA by Tsukisuwa Nana 
https://civitai.com/user/2vXpSwA7/models They make great slider LoRAs
Im gonna add negative hand embeddings
just add the extension to use lycoris
they are loaded like lora and pretty much gives the same results as lora
Btw what are the loras trained for?
Where is sd compendium when I need it
You know what loras are?
Yes i use them frequently
But technically speaking
Like this
Mine are trained for styles, the other 3, one is trained for a specific skin texture, which when you make it negative basically gives like a flat color look, add detail in negative reduces detail, makes it more simplistic in style, and kasshokuhada makes skin darker in positive numbers and pale in negative numbers
lora force determined results on the image, like a model expansion
Imagine you take model, slice into pieces and only train them, then you take only the changes you made to these parts and that's a lora model for you
upscaled image to 2024, let's see how long it takes for double it
Lycoris is basically a collection of different ways to do LoRAs, so for example locon I think uses more slices or deeper slices or something like that (didn't really look much into it) and each lycoris has it's own ups and downs
Now you know
Ye i will in a couple of days
Read here for more details
Ye sure thx
Basically, LyCoris are cool bois 
Im staying up for some research
Better then lora!?
Depends
Are these the ones that make flat colors m
?
I wonder how that prompt will fare in diff model
I think so, depends but it gud
share your prompt vii
Yeeee
Hey i dont use a negative hand embedding yet my negative prompt was accidentally train to sorta give good hands
Well, you can just use uhhh, detailer one at -1, or hadanuri2 at -1, whichever you like most, instead of using both
only like 30% as active as it used to be in it's peak
Okay TY
lol fr

10%
And Kasshokukada is just to control the color of the skin, 1 is really dark and -1 is really pale
you could hardly follow the last 5 minutes in the highest moments
Omg very cool model idea
lol yea
I'ma do that rn. (It will take 1 hour due to poop wifi)
I just put bad anatomy in negative and be done with it
my neg are quite short
Its pretty good to have, cause AI sometimes struggles to make black people in my experience, with like AI model it sometimes like just makes them tan, or like not too dark of a color, so that one helps
cuz it's limiting
I doubt it will
Mine is 300+ words.
one of the reasons i like sdxl -> no more need for negatives
Hadanuri is pretty small in size
Because Tsukisuwa Nana is a gigachad
And they train small
when neg prompt are too long it limit the result and make it less interesting
Just put like 2-5 embeds
Finna try get an SSD hard drive 2TB from the future ( all i need is a good reward in school and bam my sunt buys me it)
grinch catto
Is the meta data here btw?
Uhhh, I don't think so
Darn.
no
Ty
You welcom 
I may not be as smart as yall with SD but im getting there FR
🇫🇷
It's not smartness it's experience
it's about performance
Francis
what i need hours, with a good pc, you can do in seconds
High res upscaling is only good up to a certain point
or other data for that matter just the prompt
Unless you inpaint almost everything
yeah, with 1.5 it's goot till 4k, 8k gets crappy results
thats why im gonna use them forever

but with a good gc you can generate tousands of results and handpick the best
Even then it's overkill for most images
pc you mean?
graphic card
I think the common abreviation for that is GPU :o
over 3k download
Ayyy, thats nice then :D
We call those GPUs
I just train a bunch of styles
uhh... how does that double viewport work?
Got a bunch that are unreleased
grapic card unit i suppose
Graphics processing unit
Graphics processing unit 
oh

even weirder translation
Not weird at all
I don't like to rely on lora it feels like everything else is overshadowed by lora
I mean, I rely on the LoRAs for style
CPU = central processing unit
in english is called grapic card, or graphics processing unit? lol
What's the difference between a CPU and GPU? And what the heck is a TPU, DPU, or QPU? Learn the how computers actually compute things in this quick lesson.
#computerscience #tech #programming
💬 Chat with Me on Discord
🔗 Resources
Learn more about CPUs https://cpu.land
CPU in 100 Seconds https://youtu.be/vqs_0W-...
am i doing this right
Both graphics card and GPU
bussin
Doing what right?
Oh
You need to download it
And if you on a1111 you can put it in the png info tab
If not then you can like google "metadata image viewer" or something like that
Then put it there
oh aight leme try
oh, this upscaling is giving cool results
imagine if you can use link as a prompt lmao
Technically you could host a txt file
Then just make SD download the txt file and use the contents as a prompt
i use sd prompt reader
kazoo it only shows the prompt right not the loras and other stuff?\
It shows everything
Well technically they can hide whatever they want
even the measure of underwear
That perhaps not
then this is bs
I train small 
Make them fall
which meta data viewer do i evem goooooo
I wanna see them falling

didn't show whole thing lmao
Doesn't matter
Where you putting that?
Put it here
jimp lmao
he didn't use a1111
This is A1111
A1111 best
Omg
This is whit aurora
still no metadata 
A1111 is the way to go, any other way and you doin it wrong
(unless its comfy for SDXL)
Nai lora? You mean is the nai model good as a base? Yeah it's fine
this is why i sometimes wanna quit ai generating
a1111 sdxl masterrace reporting in
For training
No no
thx time to mass produce chibi boba
lets mnake this simple how do i apply it
Very ez to setup a1111
Go grab A1111 ma boy
ik yet where?
Copy someone's config
Its eazy, and you'll be happy
Online
do i install it like the clip skip
for some reason that prompt thinks that "White hoodie" is little bit different than what I think
Get what I sent
I tried using link as prompt lmao
it also got metadata if you want to try the link
@glacial bane to install this follow the guide on the link
Make a new folder wherever you want, and then click on like the top bar, type CMD like tis
There is a full install guide on the repo
Yes


