#🍥|anime
1 messages · Page 123 of 1
yup, but it didn't work for my regional prompter for prompt mode. Since it's not divisible by 64. I was also trying to gen at that ratio for discord server banner
ahh ok, so far i haven't noticed any glitchy output with 960x640
yup, natively in auto1111, should work fine, but with certain extensions, might not
ahh ok .. yet to test on my end
My favorite model goes nuts with anything above 512 width 
Yess
yo kid you want drugs?
I love dynamic pose!!!
forgive the fingers

i assume this is sdxl?
This is MY statue!
One day we'll use SDXL, hires 2x, and do 8 images batch 1024*1024 each
yes but I do modded lol
cat with 4gb should start it 
for dnd campaign with AI art as asset
yup, modded for sure 
Wait
that sounds like a lot of work
hard work
Terraria > Minecraft
it would've been real fun when it does work tho
They are different
that's true
I play Minecraft to explore and look at pretty things
terraria is for min maxing and fighting stuff
And making pretty stuff :<
Building is the most important
hard to make pretty stuff when you've got brain of Cthulhu on your a** at all times
lol

me going to my friend's world with drill containment unit

Double trouble
Powerful stuff
lmao
X3
welp
It's not

I think
i don't think so as well, since i dm that person before

Put that on the choker
Now you have AI girlfriend

:huh:

white hair, streaked hair

bring the supression unit

🤣
Preparing for counter attack
I made oni horns in the past!
Looks cool I think

Butts are safe to post riiight
cursed

i mean, if it's like straight up nude butt
i think not?

when i forgot to turn off ADDCOMM, and tried using prompt mode


dm just to be safe 
surely im just checking for your behalf
I see you like eye

cool rng
Changed prompt and now RP crashes when I'm using highres. fix

she looks like she's not feeling well, idk
originally from this pic
can you test this pose, just for testing purpose 
sure, gimme a min
that image is pretty small
i will load it up after the curent render
might have to edit the pose in editor lol
but let me share what came up by default ....

it's kinda working 
I got it working
holy, that's good 
I missed paw pose seed because gen crashed 
bit of a anatomical anomally
looking great 
you can press this button to get back the seed
provided you haven't press gen after that
this is the editor i was using but the original pic was a bit improportionate
it's been like forever since i've used openpose
or even controlnet
i got an update for it few mins ago
I tried your oni prompt and all my gen come out black
HOW?
there isn't even dark skin in there

also the latest dw openpose works better than the default open pose
are you using like the same cfg?
cos i was using that high of a cfg due to an extension
but not sure
got to read up on controlnet stuff soon
or is it the horror \(theme\)?
I was using like 13 cfg
maybe that's too high without that extension?
here's the entire grid
also this link gives all the CN models https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main
and also might be sdxl stuff, so not sure
maybe
i think i have those old controlnet models
or maybe yellow eyes trigger dark skin?
maybe try lowering cfg?
that will be weird
@lyric zealot 
<
/
when're you gonna make 64x64 stuff
she nod so fast it leave after images
so you aren't cat anymore
now you're Bed with 4gb


fun fact I accidentally managed abortion pic

wanna see?
🤣

what cfg u using?

7
But you're smol

u using too much extension, lora
it required more CFG to held it together
seriously, i'm already too used to it
depends, i'm really just using the cfg extension, cos it's my default now

probably a little
is that horn or halo?
good ideas I'll see if I can make fish hook in eyes

inspired by you

chat suddenly went quiet over a little joke lmao
cfg 7 she's not black anymore
new merge time, this one more focused on anime (i got too much real stuffs)
weird that cfg does that
some of the gen still lil on the dark side
weird
does combination of pink hair and yellow eyes does that?
is it not supposed to do that?
there is an argument to make it start automatically
1.5 doesn't do it without the arg
there's the settings
i forgot the argument method
o:
wait a minute
@lyric zealot YOU SEE THIS
O:
it doesn't crash on 1.6
hm
does that work for every resolution/model ?
the same size in tiled vae
i guess so
im not too sure how it works
but nubby gave me the settings that worked well for me
1.6 is so ugly
nevermind tiled vae or the super much more smoother performance UI 
I can't generate anything past 512x512 it just instantly dies
back to 1.5 crying I go 
3.10.6
i wonder why the pic came out with her sitting when i have the following prompt with no mention of sitting
1girl wearing a sweater, mini skirt, bare foot, posing for camera in a flowery garden
and the canvas height is bigger than width
well
its random
well then
its doing what your prompt says :P
but im not prompting for sitting
try using standing if you don't want her to sit
think the posing is making her sit
not saying that
trying to understand why she is sitting down
probably the posing
garden is busy
:p
probably wants flowers by her side
and is making her pose in that level
yeah could be
im gonna see what happens if i remove posing
1girl wearing a sweater, mini skirt, bare foot, in a flowery garden
she is still sitting
not that i mind her sitting cute.....
hmm it could be because i have (medium shot:1.2) and the prompt bare foot tries to fit her view within that medium shot frame
or not... i replaced medium shot with full shot but she likes to sit
I wonder to make a pokemon gijinka would I need a Lora or just generate something with (character design)
I'm curious
same with Minecraft
or other creature gijinkas
probably
like something like the Pokemon mikus
idk
I want to make sure before I'd try and fail to make a Lora for it or something
tried making one... Yeah
you'll definitely need a LoRA for this
no idea lol
try looking for "gijinka" or "擬人化" on civitai or tusiart
maybe someone made one 🤷♂️
she look mad
she hates that ketchup isn't accepted as makeup
ah I figured something out! just applying a Pokemon Lora or one of not very high strength it seems to know what you're trying to do
it didn't work with Gardevoir however
must be made differently
Oh, I made something similar one day
is thaat a bootymon?
i just tried junk in the trunk
but... alas, SDXL not smart enough to know what it means
I'll have to google that

oh
😄
is controlnet kindof slower by nature
yes -> it's an extra model
so it's going to really heavy if you're already up against max vram
🍑, sorceress, anime
works everytim
I think I will make a Sylveon gijinka first
one of my favoritesb
it seems it has to be for a specific character unless I trained on multiple concepts but...concept bleed
wait, it understand emojis?
some models do, some models don't

🍑💦 😅
i noticed some wildcards have stuff like :> but never knew if it works
your model is too spicy 🌶
im kindof trying controlnet settings at random
dm me 
k
Are you using SDXL
sent
i have a fruit with a ... debatable middle XD (not posted obviously )
belle delphine oranges
https://antfu.me/posts/ai-qrcode-101
https://civitai.com/models/90940/controlnet-qr-pattern-qr-codes
using info from these 2 sources trying ot get something like this
look at my name
I once had it and was causing blue screens on my pc
PTSD 
the controlnet qr code model doesnt seem to do anything
Is that a bad thing? 
wait the emoji in the prompt did that?
yes
thats coooooool
very nice ❤️
You gave me a new addiction...
try 🥸
is there any extension to like analyze a model to see if it recognizes certain words or emojis other than testing
hands are fucked tho
seems like it shoukld be possible to scan for words it was trained on .
The emoji turned into this when I pasted it
I don't think discord emojis work
masterpiece,high resolution,highly detailed,best quality,pbr material,Intersteller Dust,Gradient Noise,
cute girl,,,shy smile,idol makeup,short hair,playful,
pastel,saturated,abstract,
Idk if they are from discord, I never use emojis
its doing something. this is canny controlnet though not the qr code one
Maybe you should try depth ?
but I guess it wouldn't just recognize letters
maybe by messing with the detection sliders thing value
aight, got this locked in! LETS GO SDXL MODEL

im using wildcards for the first line
masterpiece,high resolution,highly detailed,best quality,__autowp/3d-term__,__autowp/cosmic-term__,__term_graphics__,
sits in a quirentine room off to the side, sneezing, watering rihgt eye thats red and itchy, oh and swolen, with a runny nose.. yeah i'm gonna stay in this room
made the mistake of going outside yesterday, worse.. walmart. blah
this is a prompt. you just don't know know it yet
uhh ok
ah this one with the qrcode model it worked almost
model
a what
Clyde?
Bro... Where's the spoiler tag?
this is clearly not safe for this chat
go to prison 
see's nothing wrong
you too 
sounds like you got the dirty mind
This is the least
I got
no, get bonked harder

me irl
for the tag-autocomplete thingy is there any way to use both or would it make sense to combine them into a single csv file or is there ane better than the other
see, perfectly fine prompt!
rest of him is missing
it happens to the best of us
bet her names "candy"
Inpainting is hard... Had to throw the result into krita to fix some small things
I'm never prompting 🍼 again
jk, I will
is there any plans for a gimp plugin or should i just learn krita better
damn, regional prompter makes this.. VERY easy
gimpfussion plugin
works with auto1111 looks like
GIMP (GNU Image Manipulation Program) is a powerful open-source image editing software that offers a wide range of features and functionality. One of the notable features of GIMP is its ability to integrate with various plugins and scripts, allowing users to enhance their editing capabilities. In this article, we will guide you through the proce...
On a side note, I'm finally getting better results

now how do I make regional prompter easy
photo of a jar on a desk in a living room BREAK anime magical fairy, jar

ah neat
tiled vae works in 1.6 but regional still doesn't
replaces bubble with jar in his prompt
and 1.6 just doesn't work in the end, I can't generate anything past 512x512

that really fucking sucks 😦
and everything you want to do takes more vram
upscale it 😄
what is cfg at?
i like how i can define different styles on different subjects 🙂
ok let's switch gears
could be a lora that is a little strong or try 4 cfg
not using a lora
just fairy in a jar works for me
i know it works 🙂
but it gets significantly harder if you want to apply multiple styles at the same time
i hate i can't upscale it with highres
swinir is also really good cleaning up images
fuck, restarting webui XD

oops
uhhh
dont'fill your jars with water if they have fairies in them
unless you need to preserve the .. thing
the cool part about the regional prompter is just that you can specify whatever you want INSIDE the jar, and it will not leak to the other parts of the image 🙂
unless... the jar was destroyed
which just happened.
got it
some things are better kept inside a jar
make a little hole so it makes the room smell nice
i got a idea
looks like it hurt
when i was like 12 i drew the best thing ive ever drawn it was a half life military grunt dude in a glass jar with like pen or pencil on paper
i ve been trying to find it for ages so i could img2img it
illegal technologies
it was kinda like that
vill, join the rage, put your waifu in a glass jar!
ok NOT what I was going for but .. ok
clipping issues have nto been resolved yet
Is chibi fine?
ghost so it makes sense
somehow i feel this is a pre-baked one 😮
but at some point, everyone has images for EVERY occasion
these new subway attendants getting more advanced by the day
reminds me of that movie the cell with jennifer lopez and a horse in an enclosure like that
Anyone want some Chibi tea?
closer
uhh... that flavor looks kinda... dank
new firmware, we got rid of the ugly cubes, got the clipping issues back.
Miku under glass
fulltime subway ghost
Is there part-time ghosts also?
not after they escape tho
don't engage with the subwaifus tho
blink blink

i can't look at one-point perspective anymore, it's driving me crazy
just used this for a prompt, added anime waifu tho 😄
bee-movie?

aight, imma go play some games ... but why does butterfly waifu have bigger... assets as cat waifu 🤔
the ones where its like viewed from a straight down angle?

ye, tired of seeing it in my gens lol
sorry for the single point perspective
Best girl candidate in terms of hair is there, but seriously this expression is powerful.
Space Miku Hatsune. 😮
oh my gawd , came out better than I thoght
now it's anime
so much.......artifact
just changed models
🤣
seriously tho the outlined is so messed up
prob not best model
Lava cake
;3
im gonna try pizza next
@lyric zealot
ok its time a supreme pizza,with extra toppings,melted cheese,viewed from raised angle,
i hope in the future matter printers take text prompts
Just testing one of my merges
What is getting scary?
@lyric zealot 

Thought this was a gen
The whole thing

lol
have u seen those 'ai' generated advertisements
reminds me of that
the shaq shaker shamillion
Yes kek
What are you doing with controlnet
doing the qrcode thing
but with text
trying to think of other examples that would look cool
I
u
huh
i can see it 
The border is the UI
I thought it was like a painting on a table
oh yeah 
think she is gushing tears lol
:<
No crying allowed
👍
@nova remnant
do you know how to make loras
i do yes
wait, for 1.5 or sdxl? idk how to sdxl
I've seen people post maven, you know her, right?
i have that lora lol
O:
Really
From where
Did you make it
i did not, this one right? https://civitai.com/models/126245/mevis-hotel-transylvania
Anime channel is friendly channel

it's Maven not mevis
The poe boss :b
Happy! 
oh lol
I would rather see your gens than some other people's bloody gens here 
;3
I have idea
i miss wildstar
What if you do the QR thing with a shape like a flower/rose
And prompt a bismuth crystal

ooooo
Almost a request cause I wanna see 
I need to test that model to see if it's possible for my vram

aaaaa
mogus

it seems to use almost extra vram n theres also a low vram op[tion
Hmmm
need a lucky seed
i can see him actively and always saying :3
reflectivity, refraction
Too many ideas 
i went with a wingding
I think your gens are plenty anime enough for here. https://discord.com/channels/1002292111942635562/1072017138447036486 might also appreciate them if you want another place to share them.
she read what you did
O:
Kek the ear texture
i forgot i had the cfg at like 15
oops, i linked the archived one instead of the current one on accident. I edited it with the right link now.
anthromorphic waifu with a __nouns__ theme, anime shading -> this better be gud

Aspic is a savory, jellied dish made by suspending pieces of food (usually meat, seafood, or vegetables) in a clear, gelatin-like substance.
you doing xl in A1?
ye
do you think it performs the same as comfy?
i forget if it's slower or faster now
Would
add in aitemplates if you can run it -> not even a question
What's that
a node that can speed up generations costing a lot vram
now it's becoming interesting
nvidia only
anthromorphic cria waifu, feature habitat , anime shading
uhhh... i'm sure someone would

my nouns wildcard being just that... nouns
i got a fetus waifu
uhhh
yeah.
it's about as bad as anyone can expect
Not an expert here but wouldn't it be better to remove that

technically, it's just extreme loli
How does that make it any better

it's fckn horror.
dependign on batch size
lemme change it to 1024x1024 euler a
2.3it/s
1mp euler a vs 1.3mp restart (1it/s)
(used) 3090
1070
xl is beeg and slow

hell 1024x1024 20 steps euler a is 8 seconds on my 3080 12gb on SD1.5 with a lora
run restart 🙂
restart + sdxl is slow
real slow
image quality is not even up for question tho
should be roughly double
how many steps?
on sdxl restart 1024x1024
i like having it at 35 tho -> nearly full deterministic
although restart gets really close in 10 steps, that's just plain amazing
What does this gif mean
my honest reaction
I just switched back to 960x544 batch size and count 2 with hires 2x upscaling and wasn't really paying attention. 3 minutes later I look over and see it's still working and think "why is this going so slow"
restart, like the name implies, moves the generation some steps back during generation (renoises), and then goes forward again, but this re-noising actually improves image quality output by a lot
people with more brains than i have 🙂
yeah, it's basically the same impact on generation speed as running DPM++ SDE or DPM++ 2S a, just felt slow because I almost always run 2M Karras
Tried out restart
those are ancestrals tho -> restart is non-ancestral
and resolves to the "ground truth" like in %50 less steps than the others
10 steps restart is faster than 20 steps euler, by far
yeah, different reason for the speed, I was just saying the actual gen time is roughly the same.
you can run restart at 10 steps and you're already very close to whatever it shoudl be
euler/ddim at 10 steps is taking a BIG gamble
however, i noticed it's REALLY bad when you're going in with low confidence prompts (contradicting tokens)
so i'm trying to figure out how to fix that 🙂
I don't like Euler below 35 or DDIM below 50 in most cases.
And I was wondering why eta was over an hour
Is euler good compared to Euler a?
euler = deterministic
euler a = non-deterministic
I usually like Euler a better. Results tend to be more interesting IMO. I prefer DPM++ 2M Karras over both of them though.
omg, i just had a brainfart idea 😮
restart is bad in low confidence, but resolves quick. dpm++ 2m is pretty good in low confidence, it just screws up in that case, but it's clean errors
highresfix with no upscale -> just do a small resample with dpm++ 2m 🙂
(inspect my previous comparison shot -> restart tends to be a bit less clean in the lines)
Yeah
That got me thinking
all the A type samplers are non-deterministic
Isn't stochastic very bad in the end if you want a crisp image
restart with 20 steps... plus hirest fix 10 steps
the output is comparable with dpm++ sde karras 30 steps and hires

I appreciate your images 2x more now they take a while
cat... try a few render with restart 20 steps
cuts render time
and quality of the output is fairly decent
first tell me how to run 1.6 without running into out of VRAM problems

You using Tiled VAE?
Yes on 1.6 it works
oh.. hm im not sure why you'd have issues with vram with webui version, but might be a good idea to ask in #🤝|tech-support
But still can't do anything above 512*512
because some versions suck and slow down some gpus like when it was slowing down 4090s
I'll do it now, it's not hell temperature here anymore, I can gen some images \o/
I'm just...
I'm glamorously in front of the fan
Drying my hair
ok, Aryetis and CS1o are pretty knowledgable with those kinda technical stuff
not sure if i like it yet, but it's certainly worthwhile to check out 🙂
and with that, imma go to bed
cya tomorrow you lovelies ❤️
bye
that looks like a promo for durarara 😍
Who is the woman in the image, @nova remnant ?
no one in particular, just generated by Ai
It's the incredibly popular 1girl
yes that one girl...
im getting a lot of bad fingers with restart 20 steps
odd cause restart is supposed to produce good quality images
restart 20 steps, anomalys..
could also be some big rng going on here
prompt diff
would 1girl and girl have a different effect on the image, now im curious
1grill
@supple raptor I have question 
my webui (1.5) is laggy, just like slow to compute where I am clicking and stuff, do you have any idea why
if I restart my windows explorer through task manager it fixes it
for a time
yes, 1girl is the actual tag used for images with a single girl in them.
I'll do a prompt matrix to see the difference >.>
.>
girl is left (ear hidden) 1girl is right (ear shown)
its useful if youre trying to get multiple characters if avoid multiple like if u want 2girls kissing u wouldnt wanna use girl or 1girl cuz 1girl is trained on single character images
but not all models may be trained with booru tags
Is there a way to know which models are ?
thats interesting to know theres a visibile difference though
not that i know of unless the publisher says so
id love to find a way to inspect what kind of terms a model has trained into it though
theres a model keyword extension but idk if thats it
the 4 genders
indeed

I'm too dumb for these comparisons
feels like 1girl is trying to do a smaller "cuter" face in all of these
is this 1.5 or xl
XL
they look good
the bikes
I don't use 1.5 anymore.
i would hope ai is heading toward less memory expensive models instead of the opposite
Is XL even good for anime gens?
now try with lady,madam,maiden,dame,lass
good
M'lady
I mean...you see results above, right? Plus, there are a bunch of anime models now.
decent
This?
Getting better every day.
THE TECHNOLOGY
Evolution will expand things.
Yeah, that...among others posted here.
They don't look anime tho (not saying that they're bad)
Anime is a really broad genre.
no nai dataset so its on 2.1 levels
I just feel like there's so many shit for 1.5 released already that it completely kills the XL hype for me
search for eface's images in this chat, hes got some good ones anime style like
That explains it
Yeah, well 1.5 has a limited lifespan and XL is fresh. Give XL the time that 1.5 has an it'll eclipse.










