#🍥|anime
1 messages · Page 239 of 1
\(darling in the franxx\),\n\nwhite t-shirt, \n\nhead tilt, \n\nmasterpiece
all o those
zero two \(darling in the franxx\) = zero two (darling in the franxx)
why the / then?
I'm using comfy btw, not a1111/forge
is comfy different?
yes
ahh
okay, the \n
is basically newline
Where did you even find that \n\
Might wanna use this to check metadata
https://supagruen.github.io/StableDiffusion-CheatSheet/
comfy is comfy
what is "CFG"
long story short, it's basically how "creative" SD can be vs how closely it follows your prompt
long story short story short: leave it at 7-10
Style loras
defines what style it will be?
Yes
most of the time yes, unless you are using some other samplers
stupid samplers like you

smuggered
i still remember i used to have it on cfg 22 all the time
umm if i make a prompt and post it here can u guys help me make it better?
but that's with dynamicthresholding enabled
if its booba, post the image with png info instead
👍

no its my negative prompt

👍👍
(score_4,score_3,score_2,score_1) extra legs, extra arms, extra fingers, bad anatomy, (bad quality), 3 arms, 3 legs, 6 fingers thighband pantyhose
First line is the mandatory, second is the less so but important and last line is optional
you are using autismmix right?
let me fix for you: thighband pantyhose
here

those negatives don't do a whole lot
they dont?
you can find some stuff better in some seeds but kinda of rng, they just add some noise
they help a lil bit
Winner's pov
its like better to have than not to
Real
Truly the jujutsu kaisen of all time
prompt:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe,
{prompt}
source_anime
neg:
(worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow
tf is artistic glow
this is what i used last time for autismmix
like scores on negative literally don't do anything, unless the merged model trained on it somehow, they aren't a pony thing

u read dis right?
@lyric zealot is stoooopid
From what I've tested negatives in Pony models mess up a lot of things, whether it's style, image composition or even some concepts. That's why I never use them
try (white background, simple background) on negative
It kinda of do stuff, I like it
prompt:
1girl, {prompt}
(bee \(deadflow\):0.8), ilya kuvshinov \(style\), wlop, realistic, depth of field, ```
neg:
```bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, (worst quality, low quality:1.1), jpeg artifacts, signature, watermark, username, ```
doesnt do a lot if u specify the background
unless you're genning white background then its gonna be bad
currently what i am using
what is extra digits supposed to be

it does, on base pony backgrounds are a lot clearer and colorful with it in the negatives
he has no idea bois
Truly my specialz
negative prompts in sdxl in general are mostly pointless, they should only be used to fix a specific problem
Every little thing mess it up lol
well, it's supposed to mean fingers, but not sure if it works or not
LMAO
Animagine's negatives work in my case
It really does, even the model description on Civitai says not to use negative prompts, they know it's messed up
here's my confetti's metadata

grabby 
Don't check the negatives lol
It's so unhinged but it works lol

specific models sometimes have specific prompts (positive or negative) that they respond well to, but in most cases the negatives aren't actually doing anything
["(gpo, aca, aer, api, fla, gcx, hll, hnj, gpc, fii, fey, fbv, evg, iew, ifl, igh, iwj, iwp, ixb, ixe, ixz, jaf, jbm, jfb, jsf, jyk, kmz, ksh, kxg, kzg, lbv, xsl:0.5), navel, fingers"]
tf
Well, only in comfy though
It breaks in a111
Do NOT prompt for gpo oh my lord 😭

how tf do i make her thighs even thicker
THICK
no seriously don't prompt gpo lol
Those negatives are some of the cursed hashes found in pony
or something
what... what is that ? 
Yeah I've seen a lot of people use those tags in their negatives, those are the hashed tags or whatever it's called
a111 also works
I think weights break stuff in pony

Seriously do not prompt it 😭
what am i missing
(from below:3.0) works amazing
👍
Indeed
You can fix that by going to "Emphasis" and turning the "No norm" setting on or off in the stable diffusion settings
Basically cursed stuff found in the hashes lol
hmm
I forget which one it is but I'm pretty sure you're supposed to turn off the no norm setting
watching youtube and generating images with a 4060 ti does not go well
Definitely
She's doing the
face

Cute
cute but no
loli
no
pls
guys
how do i find names of woman hair styles
like the one side style or the one which they tie a little bit of their hair at the back and let the rest go

wat
side ponytail?
Go to danbooru, type in "long_hair" for example, click on the question mark next to the tag and you'll see a description of what that tag is.
Under it there should be a group for hair styles
OHHHH
this is side ponytail
no dat one
I think it was more the samurai style
Wait I think I get it
that lets the side hang
I have no clue honestly 😭
this ? @noble flame
ye
left is more grabby
2nd one: i included this to neg prompt: (gpo, aca, aer, api, fla, gcx, hll, hnj, gpc, fii, fey, fbv, evg, iew, ifl, igh, iwj, iwp, ixb, ixe, ixz, jaf, jbm, jfb, jsf, jyk, kmz, ksh, kxg, kzg, lbv, xsl:0.5) credit @zinc flax
2nd one looks cleaner
Single hair bun?
Left one is too saturated for me lol
both's great
I like em soft
but i kinda prefer right

let me see the soft part
i tried adding that to my negative and it doesn't change anything
must be artist specific
I'd definitely get the bonk if I send it here lol
although i'm not using tponynai either
It doesn't change much for confetti
If you use something like pony anime however
i thought it works for pony-based model?
The right is just nicer to look at honestly, left is way too saturated
agreed
@lyric zealot are you using fatal anime ?
i mean, if i only have the left pic, i won't mind, but comparing them, i do prefer the right more
yes
You're saturated
ok google set timer 10 min
I just noticed this but, why do most of the anime girls on local models have this hairstyle in the shape of an M? 😭
Like, their bangs
is it a miku thing?

Style bias, probably

Mcdonalds hairstyle
since in my prompt, i'm using miku as based then removing it after 30% of the steps
NAI also has that, though again probably depends on the style lol
Even this one, at the top of her head where the sidelocks begin, also an M
how do u do dat?
well cos the prompt is....
thats the most basic girl hairstyle in anime 
stupid
meaning after 30 percent of the steps, it will change from miku to which is nothing
(something:1.5) does wat?
that's the weight
are you using forge?
for what res?
it generally takes 30 seconds
1024
and you said you are on 4060, so you have 12gb vram?
with hires .fix to 1.5x
No wonder lol
4060 ti
then you should not take that long
4060 is either 8 or 16
even on my 3070, it's only around 40 sec for 768x1024 and 1.5x
oh 8gb vram
sadge
cat invasion 
I guess that makes sense, I never really noticed how all or most of them have the same hairstyle with that tag, like this for example
hair vent 
so ugly 
😄
LMAO
actually useless plot holy shit
i didn't really notice that lmao

@nova remnant
I think you also get that by prompting for hair intakes 😭
You asked for it didint u?
"hair_intakes" is the actual tag
lmao even worse

This is 2000s hairstyle lol
I'll stand there and she can sit back
👍
It's not just an AI thing, it's a trope in a lot of anime so I guess no need to blame it on the AI models
Those eyes are something else tho, actually cursed 😭
huuuge eyes
yeah bro
like 1/3 of the face
Still love the right one lol
only fatal anime looks different from the others lmao
It's the anime moe eyes
useless maid

Rikasa taking a break 
Fancy coloring. How'd you prompt this?
Horror and injuries warning
banned
It should have metadata
Metadata got wiped lol
Oh yeah I forgot that nai uses alpha layer when I copu from ui
This should work
Ooo nice artists
I think something in that negative affects style alot
no wonder i can see it on this one
See what?
why she place her sneaker on the sofa 
i have no idea blame the AI
I think that's just the basic negs provided by nai, no?
I copied it from somewhere
there
Nai is struggling 
Yep, same negatives lol
nsfw, lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]
I think animagine also uses this one
u liked it?
Too much 
One I'm using is little bit longer
Vibe transfer is so fun to play around with
Adding a bit of noise while enhacing also gives the image a neat look sometimes
It's one of those negative presets that you can use, I'm pretty sure it's from the "quality" or the "human focus" preset
Apparently you're not supposed to use those presets when using artist tags but they usually give better results for me

why>

the horns
ahhh, well, it makes sense for it to block the hair i guess
all ugly
lol

It says pony needs to be used with clip skip 2. How do I do that in comfyUI?
should be set by default
In a1111 at least setting it to another clip skip doesn't even do anything
Oh ok. All my gens are trash on pony. I copied off what I saw on civitai except the Lora; I haven't used any Lora. Is lora needed to make good images with pony?
no
Ok I'll just experiment more
Send something with PNG info
but comfy cursed PNG info
I'll try a few things tomorrow but if I can't get it working I'll ask for your help 😺 tysm
Also this is about as good as I get pony to look
It's not gonna be mind blowing just a good base model
More than that you'll want merges and loras

Yeah mine look like they have that static look on old TVs and stuff. Something I'm doing majorly wrong. Maybe resolution or something. I haven't tested it too much, I'm sure I'm overlooking something
Send me the prompt
Youre using
I'm not on my PC and about to sleep soon. But I copied random prompts off civitai and did my own too, none good
sort of like scan lines with a chromatic aberation effect when you zoom in? if so, that's a problem with the vae that they released with SDXL 1.0. it's also not compatible with fp16 which will lead to either slower generations due to running the vae in fp32 or black images due to NaN errors in VAE. use this one to solve both the scan lines and the fp16 problem. https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl.vae.safetensors
i need to work but i dont wanna ><
ok what model are you using?
hi hi here! this is AWpainting + two Loras for styling
ahh ok, thanks
My pleasure. nevertheless, it is already in the metadata. I'm not deleting anything.
wow very small model
this is a model for 1.5
damn how do you do this!?!? and how resource-intensive is this?!
it's pretty easy, and not at all
I should take a couple of lessons from you if it wouldn't be too much trouble...
me.. I see .. squirrell.. gives chase, runs into wall
you can't hide!
just goto checkpoint merge tab it explains itself,
I should be asleep but brain not shutting up
... sigh
I actually turned off this tab in the interface. because I was sure that I wouldn’t have to work with it
oh come on!!!
yes she is very nice and all her versions
ok merged my NOMD model (which is a big merge itself so y eah) and awP
same seed and prompt
hmm reminds me of the inkpaint style I used with a lora
and the model?
see above
Have you already done the merge?!
anyway, now I’ll finish with one more generation and restart the server
but nah i'm lazy
but hey @brittle rampart you are doing some amazing art, keep up the good work
I just dabble in random stuff
Yes, you and the Cat are still lazy! 😂
you flatter me. there are much better ones in this regard
but meh my brainz not in art mode
morning

I’m just almost quoting what was said 😉
It does have weird colors but the composition is bad too. But thanks, I'll try that one
my student we'll learn about cats and why they sleep so much
We already know why cats sleep a lot - they are lazy! 🤣
Smol
midori and momoi
about Blue Archive
yeah
Is this Lora!?
ok i will send the lora
not this
this one
👍
8/10 on the smol scale!
Maybe it's the "vintage theme"
Try without weights
👍 but not anime style...
Anyone want a ride 
cut your hair bald! Then all you need is polish!
We'll call you a Sphynx cat!

there we go
kinda.. her hair wtf?!
gets back to doing training course for work (sneaks away)
shhhh
it's just a costume party at school
Witch, summon a nice breakfast for me please

Demons are not allowed!
😉

Is it just me or is it really so. for some reason, in my case, the DPM++ 2M Karras sampler gives the best results.
just you
😄
I like it too
restart is good
now I'll try Restart
ok it's DPM++ 2M Karras
now Restart
I think restart will be on the speed of dpm with 40 steps or so
If you're using 20

I don't remember if it restarts on the hires process too
and how many optimal steps does Restart need?
40 at min
I worked
80 went overboard
food witch
40 steps minimum!?
yep
@brittle rampart 
and on HRfix?
eular
aaaaaaa!!!!
I don't use HRfix most of the time
I got a third party stable diffusion upscaler it goes to like 8k and is fast
from giveawayoftheday
becareful of hidden strings attached to those, read the terms for it
this is in 25 steps + HRFix 14
normally it's like 80 a year or somethng
hmm ok
oh I'll have to look into that
Euler 20
like use limitations?
I like 30 or 35 if its not too slow steps on dpm++ then 10 on hires
If doing restart probably like 20 steps normal
how that came about was during old west they had free food in bars, but you had to purchase drink to get it, and the food made you thirsty.. so you got more
drink
kinda like the "free" pretzals
but based on the result it seems that there were not enough steps
scary 
Very!
haha
You have to find the sweet spot, more steps is good but at some point it becomes too much and not worth the difference it makes o:
W and L resolution can effect it too
what does it mean?
Width and Length
ya know like square, retangle, vertical, horizonal
like 768 x768, 960 x 640 or 640 x 960
oh this...
or ya know 1024 x 256
and this is Restart at 40 steps + 15 HRfix
now try it without the HRfix
then you need to increase the size to at least 768x1152
I don't
But isn’t this fraught with the risk that the model will begin to fail?
has mine?
Well, without HRfix at 40 steps Restart
it's small
thats what she said
who is this!?
shrugs
Weird
~7 min
do that or else it will censor to death
Is that on dalle why censor
yes dall 3
yeah we're havin' a good time
very very cute style 
yeah it's a neat sketchy thing
on various "pony" models... I like them for characters but cannot do backgrounds/scenes with the characters very well
and the other way
prompt: Sharp focus on Two girls, duo, Miku (aqua air) and Yuki (white hair), side by side , field of large yellow flowers, mountains, clouds, sunny day, summer dresses (yellow and white), wide brim straw hat's, sketch painted, anime fine art style, art by mark Miller
doh
air haha
so tired
changes direction
that came out interesting
not what I was going for but interesting
not sure who is who. and where is Miku and where is Yuki 🤪
well that got creepy
yuki on left, and miku on right... maybe
what the feck?! warning
Cool
🙃
and that means Harry!? yes...how his life has been battered!
shaggy and velma at bottom?
By the way, how do you get such dark images:?
Dark, gloomy night and mangledmergeVAE
hmm seems old but it works https://civitai.com/models/88453/mangled-merge-vae
yes already find
tho also (style by Lovecraft:.8)
😄
might also be Two tone colors, red high lights
oh here---> (Scooby Doo and gang Manga Comic Book cover, fancy text:1.5) (Dark, gloomy night:1.4) sharp focus, reflections,, Japanese Castle, 3.5D,, dark moon, no lights, smoke, ruins, after war, ground fog, low dark clouds, sketch painted, anime fine art Two tone colors, red high lights , (art by Akihito Tsukushi:.9) (style by Lovecraft:.8)
haha
changes VAE
still dark
looks like a far cry from Scooby
👋
good pencil drawing came out
horror! this is a zombie cat
yeah oops
nothing behind those eyes
only about 2hrs left to go then i'm gonna go fall into bed.
hehe think I used enough saved styles?
what`s the soft line art prompt u saved ?
he must be tortured with cuteness so that he will respond!
shrugs would have to look it up. ask me later or in dm
ok np
yea
realistic , photo, movie, soft edges ,multi-shaded colors, anime overtones, line art style
hmm need to go thru these and redo maybe...
prob only part that matters?---> soft edges ,multi-shaded colors, anime overtones, line art style
ooh this is retro
vivid pastel saved:---> retro anime style, fine art, cel shaded, elegant, vivid pastel colors and shading, line art overlay, holographic triple shadows
haha
ok i'm laughing at everything now.. brainz to tired
awpainting model, furception vae: prompt: retro anime style, fine art, cel shaded, elegant, vivid pastel colors and shading, line art overlay, holographic triple shadows Miku hatsune in the city
sighs, changed some settings, wondered why it wasn't going.. oh right, hit generate
go to sleep don't torture yourself
still on the clock for work training
or trust me I would if I wasn't
what the!!?
I do phone customer service from home. going to new project
Do you have training there 24/7?
no no , only till 6pm Eastern time. so umm 1 1/2 more hrs about
keeping my brain busy
hehe
grinz , changes the background and press's generate
ok not big change
hmm that works
hides all miku's from sunny
prompt: retro cartoon style, fine art, illustration, Cel shaded, elegant, vivid Gothic painted pastel color shading, line art overlay, Multi=sampled holographic Shiny triple shadows , Chibi chibi child Miku , Sci-fi Fantasy, complex strawberry background
hehe
realcartoon3d Model: mangledmerge VAE:
ah ha, Hi-res Viivid pastel.. gotta save this one
randomtora over here with the Owakita style
hmm what?
this?
oooh creepy
damn
that ain't creepy bro
dammnnnn
Tifa Hatsune
damn
yeah... damn
moreso the more pastel outdoor ones
leather coveralls
I think i'm mixing some i've seen.. who's the one with white hair and red horns?
guess no one likes my art.. sigh
brain fading, getting grumpy
30 mins to go
go back to the owakita ones 😛
cool
oh she mad
I find it strange how, over time, an expression is more like a symbol of emotion than the emotion itself. She doesn't actually look mad but the expression symbolizes it via the mouth because that's what people have come to agree on.
thud
isn`t that "pouting" ؟
no she about to throw a magic ball at somebody
anyhooooo
❓
I suppose but I wouldn't call it mad, the eyes are basically expressionless
Always thought that girl in demon slayer was playing the harmonica. That's what it reminded me of
because thats not how you flute 
I only recently saw that wasn't the case. But I only saw 1 episode, don't plan to watch more
That's when it's applying a VAE I think
And it takes the longest
6 minutes?
definitely not 6 minutes
what is your vram usage
is it hitting 100%
the thing is that
the cmd says its done
forge page says its at 95%
something like this @nova remnant

Are you using --medvram-sdxl ?
didint change anything on forge config
its all default
recently updated forge?
forge is now a beta testing branch
new update will probably break stuff
try adding that to your arguments
see if it helps
don't update
where?
it will probably break
F:\SD\webui\stable-diffusion-webui-directml something like that
then open the batch file you use to open sd in notepad
Mine: set COMMANDLINE_ARGS=--medvram --opt-sub-quad-attention --opt-split-attention-v1
on yours just add the line I said and restart your sd to see if it does something
Managed to get pony to make images that aren't complete trash. I mainly wanted to use it to make political cartoons but still having more luck with BluePencil XL. But I have more experience with it after all. We'll see how it goes
for forge
you let it decide itself
so you don't need to do that
maybe your webui is lagging
just refresh it

cuute
smash
well, it's just wlop lora

I mean, smash the competition

coool !
does this count 
You now have 1193 coins!
sorry was just testing something ^
Ugly 
Coot, like soosi :O
Is that uhhh, Jemnite's model? :o
1girl, solo, white hair, parted bangs, long hair, blue eyes, sportswear, open jacket, tan, nose blush, heavy breathing, open mouth, panting,
(bee \(deadflow\):0.8), ilya kuvshinov \(style\), wlop, realistic, depth of field, <lora:wlop-000010-pony:1>
Negative prompt: bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, (worst quality, low quality:1.1), jpeg artifacts, signature, watermark, username, white background, simple background, (gpo, aca, aer, api, fla, gcx, hll, hnj, gpc, fii, fey, fbv, evg, iew, ifl, igh, iwj, iwp, ixb, ixe, ixz, jaf, jbm, jfb, jsf, jyk, kmz, ksh, kxg, kzg, lbv, xsl:0.5)
Steps: 24, Sampler: Euler a, CFG scale: 7, Seed: 3312286734, Size: 768x1024, Model hash: 89e7c7518c, Model: tPonynai3_v55, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Lora hashes: "wlop-000010-pony: 72fe85e36a35", Version: f0.0.17v1.8.0rc-latest-269-gef35383b```
so i don't think so? 
O, Tpony, I didn't know it could do look like dat, das fair then :O

here we go again
Wow a free 50$ steam code how thoughtful 
free money glitch 
Whats that now?
😄
model?
trying to pull off that style isnt' easy
might be all the artist tags, and using lora, that's why

no no problems...
trying for a Drow Dark Elf... sigh
gets mad and finds a lora
ooops didn't tag it right, but interesting result
changes to a more anime model
meh
.. gives up
what is it you're trying?









