🍥｜anime

[nova remnant]

this is from nubby

[steel arrow]

I dod this in 3 seconds with autism
pony

[keen hollow]

this is humungo lol

[nova remnant]

using lightning version?

[steel arrow]

Not that I know of

[nova remnant]

yeah it is lol, i narrowed that down to 10 maybe that i like for pretty faces and angles

[keen hollow]

you have a 4090 or what? lol

[steel arrow]

4070ti

And I am running at 4 steps

[steel arrow]

5 seconds 20 steps

[steel arrow]

I'm not at my pc so I'll download this and try it later

[nova remnant]

found more interesting artists

[keen hollow]

im doing an artist run.. prompt s/r

[nova remnant]

Hayao Miyazaki

i havent seen SDXL does feet rendering this good

i mean this is sdxl too but not without artist specification

[keen hollow]

frankly, I don't render many feet LOL

[nova remnant]

me neither, lol, but i saw some amazing image texture with nubby posting them

[keen hollow]

my x/y grid messed up... re-running

[pure oyster]

you also probably didn't see the 300 terrible ones before it

[nova remnant]

hardly the terrible ones, once you get artist style in place 🙂

[nova remnant]

yes those are the facial symmetry i was going for

looks like female version of conan

[keen hollow]

adetailer OFF for this one...

i dunno what's best lol

[nova remnant]

i kept them all

[keen hollow]

and no artist...

[nova remnant]

and i can randomly pick one and see which style fits my scene better

[pure oyster]

I vote for last one

[keen hollow]

not sure it is having much impact

[nova remnant]

CLAMP

Shinichiro Watanabe

[native halo]

one from kantoku for comparison

[nova remnant]

if that's regular sdxl im not sure that will work with artist style

but let me share my version of it ...

from autism ofc

[native halo]

no thats the human kantoku

heres the human CLAMP

[nova remnant]

kantoku

[native halo]

looks nothing like him 🗿

[nova remnant]

lol ok i get it

but joke aside regular sdxl wont interpret styles as good as pony/autism

[native halo]

try kantoku with pony

[nova remnant]

im using autism

didnt wanna download pony cause of style swing

[nova remnant]

Akiko Matsumoto

this artist is known for graceful feminine figure and flowy hair

[keen hollow]

ill try it

[nova remnant]

all four looks good

[keen hollow]

I put "close-up" in my prompt and it went crazy lol

[nova remnant]

Junichi Sato

Makoto Shinkai

Tadahiro Uesugi

Akira Matsushima

[junior quest]

model? animagine?

[astral jasper]

Yep

[junior quest]

style...

[nova remnant]

💤

[near elbow]

@wooden coral @bold ether @rose tapir sorry for pinging all three of you, but you seem available and we have a spammer in several channels

have some color

[lyric zealot]

holy new way of scam

[bold ether]

might want to delete the screenshot now, just in case 😉

[nova remnant]

768 x 1152

[cerulean cosmos]

is that animagine? @nova remnant

i though it was some pony stuff

[nova remnant]

these image i just posted?

[cerulean cosmos]

yeah, most of what u have posted

[nova remnant]

lol i was posting tons different models too but these now are lightning bloom

[cerulean cosmos]

but the girl sitting on the tree stump?

[nova remnant]

that's autism mix

[cerulean cosmos]

AutismMix SDXL?

[nova remnant]

yeah

[cerulean cosmos]

it seems to follow prompt a whole other way then sd 1.5 does

[nova remnant]

yeah very different from sd1.5

[oak spoke]

AI is getting good at furry it seems

[nova remnant]

is that pony v6?

[oak spoke]

it's autismsdxl

interesting name ik

[nova remnant]

ahh ok, yeah i use that too 🙂

[nova remnant]

Furry was always better than 1.5 lmao

Easy fluff model is kinda like pony for 1.5
And now there's pony which is basically a furry model too

[nova remnant]

[keen hollow]

mornin'

[keen hollow]

trying out that autismmix lightning

[nova remnant]

hi how is it working?

this pony/autism models are the bomb

[nova remnant]

goblin

[keen hollow]

um, it's ok... getting somewhat muted images

[lyric zealot]

white skin goblin?

[keen hollow]

gotta go to 8 steps

i'm using the '4-step' version but... I dunno, that's a lie lol

[nova remnant]

ooooh, thats what i was thinking, ... if there is mismatch between sampler, cfg and steps itll produce muted images

[keen hollow]

gonna try some other samplers that apparently are recommended

here's an 8-step, 2 CFG set

[nova remnant]

if that's 8 steps it looks pretty good

[keen hollow]

yeah that's 8

[nova remnant]

try a hardocre test to see if the steps misses any detail, generate feet

[keen hollow]

gotta get to 4

that's with adetailer, I should clarify

[nova remnant]

the images i got here ... i can see the veins on feet clearly

[keen hollow]

cfg 2, steps 4

not great

that's with euler a sgmuniform

dropping to cfg 1 is too muted

[nova remnant]

how did you get sgmuniform on a1111?

[keen hollow]

I'm using Forge

[nova remnant]

oh

[keen hollow]

yeah, ran some benchmarks and it truly was faster in all categories

ok the images are garbage.. but next example...

[nova remnant]

im basically using the dpm++ sde karras for most stuff

[keen hollow]

6 steps, cfg 2, DPM++ 2M SGMUniform

gettin' closer

[nova remnant]

third boot

[keen hollow]

yeah you gotta have the extra boot

[nova remnant]

so you have both 8 step version and 4 step version downloaded?

[keen hollow]

nah just 4

the 8 step is probably higher quality but always requires 8 steps or more I guess

and I want a 4-stepper 🙂

[nova remnant]

i wanted to apply the 4step lora on the regular autism mix

but wasnt getting decent results

6 steps and cfg 2.5 looked kinda off

and going lower made image lack detail

[keen hollow]

4 steps, cfg 2

[nova remnant]

didnt get the rigth balance

[nova remnant]

not bad

looks clean

[keen hollow]

yeah getting closer... running a sampler grid now

oh geez... and it went dirty

sigh

wasn't there a keyword to prevent that?

[nova remnant]

rating_safe

[vital raptor]

using 'nsfw' in the negatives can help too, depends on what's causing it

[nova remnant]

neither sd or sdxl can do feet properly

[keen hollow]

definitely some clear winners here...

[nova remnant]

well by sdxl i mean regular

how do you like the sde karras?

[keen hollow]

I use that almost exclusively with your model.

[nova remnant]

omnia ... example of autism and how well it does feet

[keen hollow]

I don't really generate a lot of feet LOL

[nova remnant]

yeah i know lol

but in context of the whole subject that adds to beauty

[keen hollow]

getting easier to pick out the combinations that work...

[nova remnant]

i like dpm++ sde karras, looks more crisp than the rest in there

[nova remnant]

bad anatomy, asymmetric, deformed, conjoined, bad hands, bad legs, bad feet

[keen hollow]

bumping to 6 steps gets all the samplers in a better state..

yes I know the images are garbage, just an example. and I added a foot for you.

[nova remnant]

sde on the left looks like her position is normal

[keen hollow]

gonna go back to 4 step and turn of adetailer as I think that's skewing my perception...

[keen hollow]

most of them have something unique i like... a couple duds haha

[nova remnant]

the farthest left looks most aesthetic

[keen hollow]

ok, this is tough...

running a grid of 60 😛

[nova remnant]

no luck here with 4 step lora

the quality i get with 24 steps and 7 cfg i can't complain

[keen hollow]

wow i can' tbelieve that uploaded.. it's like 80 megs

[nova remnant]

dpm 2m turbo is a sampler?

btw some of those results could be off from the prompt

and complex prompts may differ more

[keen hollow]

with adetailer...

4 steps has obvious duds

[nova remnant]

i couldn't get it to work right either with lora

[keen hollow]

ugh i give up.. I dunno how to choose lol

[nova remnant]

pick the one closest to prompt

euler a sgmuniform looks good but might be missing some detail compared to dpm++ sde karras

@keen hollow their lightning model is obviously not fine tuned .. it'd be better to use their regular model for quality and prompt coherence

[keen hollow]

I dunno, I like the speed of this. 🙂

[nova remnant]

is it responding to your prompt closely?

[keen hollow]

yeah pretty much

it's a very basic prompt

I like the crispness on the last two samplers but it is struggling with the hands a lot.

[ivory pebble]

That's a model based on PonyV6, there's other models based on Pony but that one is by far my favorite one, mainly cause it does human anatomy better than any other model, not sure how but it's very impressive

[keen hollow]

@nova remnant - yeah so .. I added a bit of complexity to the prompt and it does pretty much fall apart.
"1girl, tifa lockhart, skirt, thigh highs, on stage, holding a cat in arms"

[void galleon]

Nice Kitty at the bottom right edge heh

[keen hollow]

yeah that's weird haha

[nova remnant]

This model doesn't really understand natural language

[void galleon]

Multiple Cat words, hahahahaha, lol, NSFW (P)Cats

[nova remnant]

You have to look for ways through tags

[keen hollow]

yeah i'm not great with tags 😐

[void galleon]

i use the word kitten in cases like this

[keen hollow]

then I get monstrosities lol

oho my gosh

[cerulean cosmos]

alright, great, thanks for telling me

[nova remnant]

score_9, score_8_up, score_7_up, score_6_up,

tifa lockhart on stage, holding a cat in her arms, wearing skirt, thighhighs, cinematic, spotlight,

source_anime, Tetsuya Nomura, Kazushige Nojima

looked up artists name who created tifa

not sure if that worked

A very buggy tifa xD @keen hollow

[void galleon]

Cursed Pingers

[ivory pebble]

The model was trained on images from various booru sites (danbooru, gelbooru...) so I recommend using the tags from those sites for prompts instead of natural language

[nova remnant]

Pingers

Pringles

To do like "on a stage holding a cat"

Try the tags "stage, holding animal, cat"

[void galleon]

I generally use "Hugging a kitten"

[nova remnant]

Animal hug is a tag

Sometimes pony doesn't really know how to translate things that are close to the tags and you have to look for the tag exactly

Like "crossed legs" won't work and you have to do "legs crossed"

Big tifa

👍

[void galleon]

Let me try using a odd pose

[nova remnant]

Falling, open arms, action pose

[void galleon]

yeah, my computer didint like that, i got something totally random

The reason i cut the image

[nova remnant]

What even

[steady grail]

i missed something or the key elements of sd just dissolved?

[void galleon]

i put riding on a turtle (Clearly thats what i got)

[oak spoke]

[oak spoke]

I wonder how effective it'll be to make like actual spider girl like actual arachnid

[nova remnant]

Just type arachnid, monster girl

[void galleon]

omg no

[oak spoke]

I wonder how good will it come out

I'm having smt else in queue atm

so couldn't try yet

but soon™️

[void galleon]

K, im using my other program to get slightly better results

[oak spoke]

it's a little confused but it got a spirit

🤣

[nova remnant]

[oak spoke]

what're you crying for? should I make it eat cat next?

all hail spider MILF

[void galleon]

lol wut?

[oak spoke]

the cat really like her it seems

[nova remnant]

@lyric zealot what has this chat turned to

[lyric zealot]

i see nothing wrong

[oak spoke]

it hasn't turn into anything it shrimply revert back to the biblically lore accurate time of this channel

[keen hollow]

yessssss look at all the Tifas

mooarrr

well that's weird

no idea why that doesn't show up as an image

oh well

[nova remnant]

casual 17mb jpg

[near elbow]

t800 meets anime

more classical

[green temple]

hmmmm... actually nice < what`s this model ?

[nova remnant]

@keen hollow i had to delete the 4 step version of autism, wasnt working out as good as the regular version

[steady grail]

pony with turbo, pretty much gives the same results of the defailt one

[nova remnant]

ive read about pony v6 has more style swing which is bit unpredictable

[steady grail]

never got a good result with lightning

[nova remnant]

im loving the autism regular model

eface talked me into downloading the regular model over 4step one, that was a good recomendation

and i was just starting out autism model, if i had gotten then 4 step ... i'd probably delete the whole thing moved away from pony/autism

[keen hollow]

Yay more Tifas

I got decent results with the lightning one, but I'm not good with tags

Those hands you got are good.

I dunno how to use turbo models.. aren't they the same as lightning? Low cfg?

I dunno how to use lcm ones either 😬

[nova remnant]

yeah but trubo is like 6-8 steps

but very similar

the whole lcm and turbo branched out from the idea of using low steps and cfg

same with lightning

[steady grail]

yeah , but lightning is even more aggressive and degrade the gen even more

not worth the compromise, in my opinion

[nova remnant]

a different tifa

[nova remnant]

the quality that gets compromised depends on how they are integrated, the technology itself is not necessarily compromised

for example the 4step lightning integration with autism mix didnt work out

but there are other decent models with lightning version work very good

[nova remnant]

steven hey

how did you make that image?

[native halo]

dalle

[nova remnant]

but how did you get all those feet aligned like that

[native halo]

it was just rng

[nova remnant]

wow

but you did similar things with other objects too before

like putting things in miku's face

i mean mouth

[native halo]

oh yea i think it was with dalle too

[nova remnant]

amazing i thought that was to do with some controlnet lol

[native halo]

but its rng sometimes u get an img but most of the times u get blocked for unsafe img content

you can do it with controlnet too but takes too long

[nova remnant]

i still have tough time using dalle with anything revealing

[nova remnant]

yeah not going to do that... but i was curious with those images you made

[native halo]

with dalle is easier but u get this most of the time

[nova remnant]

lol

yeah im way too familiar with that pic. btw since you can run sdxl have you looked into autism model?

[native halo]

yea already tried bunch of XL models but i dont use them,less content for XL

[nova remnant]

oh? dont like sdxl style?

[native halo]

no,less loras,less controlnet models,less motion modules for animatediff XL and less performance

[nova remnant]

ohh yes, true, but you are good with prompt ideas, you could do pretty amazing stuff just with prompt using sdxl

and i'd recommend austism mix model... you wont need much lora

[native halo]

i mostly do videos now and motion modules are not good on XL

[nova remnant]

do you upload them on civitai?

[native halo]

no just youtube or tiktok

[nova remnant]

ahh interesting

i havent dealt with videos much

didnt get any good guide to start with

played with animatediff for a bit but the quality wasnt impressive

[native halo]

yea takes a lot of effort to make them look good,if u only do a simple vid2vid it will look bad or foggy

[nova remnant]

yeah and they are pretty short in len

but you know about sora, i guess that technology is finally maturing

[native halo]

gonna be censored like dalle

[nova remnant]

most likely yes

stability ai is releasing txt2vid too

but right now i cant wait to use sd3

altho i should mention after discovering autsim mix (pony style) i love sdxl a lot, also merged a model on sdxl that i think i managed to get consistency

the concept i was trying to reflect for scenic beauty.. 1girl, feet focus, tall grass, closeup, morning dew, sunlight, bokeh.

[native halo]

i too get a bunch 6 toes sometimes

[cerulean bane]

[nova remnant]

yeah kinda rng going on

the good part about autismmix is that if you narrow in on a specifc artist style you like, you dont have to rely on rng too much

this looks neat

[keen hollow]

dang some of these are top notch .. this is the full model?

[nova remnant]

yes, 24 steps // 7 cfg

[keen hollow]

confetti or orig?

[nova remnant]

yes confetti

[nova remnant]

also used dpm++ sde karras

[keen hollow]

ok ill try it

ooof so slow lol

spoiled on lightning 😄

[nova remnant]

yeah lol wish their lightning model worked out

just wasnt getting the right quality with their lightning 😐

[keen hollow]

from coherence you mean?

[nova remnant]

image quality

I want sd3

the images i was getting were too baked or too dull

[keen hollow]

oh

this was lightning

for me it's the hands

[nova remnant]

but was that with sgmuniform?

[keen hollow]

DPM++ SDE

[nova remnant]

hmm that image tone looks quite nice

i wasnt getting anything close to that

[keen hollow]

i can get 'em, but coherence was meh

side objects would be weird.. hands would be bad

versus full confetti.... I get this

[nova remnant]

yeah unfair trade off

gotta have prompt coherence and quality in image

the regular model isnt bad

working ok for me

and on 3060 rtx takes less than a min

[keen hollow]

i have a 3060 also. 12 gig

euler-a takes 20 seconds

[nova remnant]

yep 12gb here too

[keen hollow]

right on

[nova remnant]

euler a takes exactly 20 sec

[keen hollow]

yeah

[nova remnant]

lol

i find that amusing

[keen hollow]

it's a softer output, so just depends on what you're looking to generate

[nova remnant]

yeah

[keen hollow]

i should run a grid with confetti on the samplers I guess...

still get wonky hands sometimes

stable g'on stable I guess

[nova remnant]

[keen hollow]

DPM++ SDE Karras for that one

[nova remnant]

[keen hollow]

it's just Tifa, calm down

[nova remnant]

Too much tifa

[keen hollow]

never too much tifa lol

[nova remnant]

You'll get tifa overload

[keen hollow]

ill see if it can do aerith...

ugh I suck at tag prompting

[nova remnant]

Do you have the extension thing

It kinda helps

[nova remnant]

I just go on danbooru look at similar stuff when I'm not sure

[keen hollow]

that place is landmines lol

[native halo]

not for me

[nova remnant]

dpm++ sde karras despite being it a lil slow, its good for consistency and less anomaly

[native halo]

https://tenor.com/view/civ5-civ4-civ1-civ6-look-at-these-yields-gif-24164271

[keen hollow]

there's a safe one out there I think

[keen hollow]

it's like, double step processing or whatever

[nova remnant]

(latex:3.4) on it

[keen hollow]

nice

when I first started with SD, I was team LMS 😄

[junior quest]

animagine

and dat style is bat

[nova remnant]

@keen hollow

[junior quest]

https://tenor.com/view/gear-5-gear-5-luffy-gear-luffy-gear-5-luffy-gear-5-white-gif-4070656934437701360

[nova remnant]

Little heart

[native halo]

this one?

[nova remnant]

https://tenor.com/view/cat-funny-cat-yoinky-sploinky-gif-22375025

elsa

[keen hollow]

do you wanna build a snowman?

[green temple]

Oh the XL

[native halo]

https://tenor.com/view/pepe-hacker-pog-gif-19786221

[junior quest]

yesh

[green temple]

👍

[nova remnant]

good artist source_anime, ebifurya,

[keen hollow]

upside-down artist? haha

[nova remnant]

lol .. just one of the pose

but look at how well the face is drawn

one of the booru tag that im using tifa lockhart's sporty dress

[keen hollow]

which one is ebifurya?

[nova remnant]

all of them above

[keen hollow]

ok time to run some ultrawides

ugh bad hands

[cerulean bane]

imagine you made your arts degree in 2022 and now SDXL took your job LMAO

[nova remnant]

as AI advances lotta changes will come into effect and society will eventually have to come to terms with doing things they'd enjoy and doing things for money

[native halo]

not yet cause ai cant do certain things thats why here some companies hire an artist that can use AI to combine the best of both worlds

[keen hollow]

running an ultrawide batch... see ya tomorrow

[nova remnant]

have fun

[nova remnant]

LCM sampler

4 steps

[nova remnant]

this is an interesting artist style

[native halo]

i recognize the red spots with a shine in her skin

[nova remnant]

im cataloging images with artist names

people make style based lora

can you imagine the power of pony models that can give you all kinds of different styles

[nova remnant]

man oh man

you can basically switch art style completely w/o needing lora or swapping checkpoints

got 25 artist cataloged with images

i have like 315 total artist count

think i will write a mini article on autism mix model and upload all the images to google drive for easy downloads

[desert island]

So new to this I don't know what any of that means yet... 😆 I just installed sd finally today and am just trying to understand basic prompting atm.

[nova remnant]

very nice, its a lot of fun to play with sd, and a lot to explore

[nova remnant]

@keen hollow

[wraith crescent]

https://tenor.com/bSqKH.gif

[supple raptor]

Pony already has this power. Literally hundreds of artist tags work on it.

[steady grail]

artist tags weren't removed?

[supple raptor]

Apparently some were but a huge number of them work. I've only seen evidence of a couple having been changed to random letters.

[steady grail]

oh

[supple raptor]

As per usual the internet played telephone based on the research/testing someone else did and greatly exaggerated the truth.

[steady grail]

pony is so popular on civitai

[steady grail]

cute

[haughty lynx]

is this some sort of control net technique?

[gloomy lodge]

its autism at work and not the model

[oak spoke]

it's the SDXL ppl were hoping for

[nova remnant]

You sure some of them were removed from the data?

typing

[supple raptor]

I remember actually seeing evidence of at least 1 popular artist who's tag doesn't work but some random combination of like 3 letters produces their style accurately and I've seen claims of a couple more that are the same deal.

[oak spoke]

weeeeird

[nova remnant]

You're weird

[oak spoke]

I'll drown you in tomato juice

[nova remnant]

I know you can't use a lot of artists names in the pony discord

The bot blocks you

@oak spoke the bot blocks you from typing tomato

[oak spoke]

it does?

[nova remnant]

It's apparently some artist name

Very stupid

[oak spoke]

🤣

that is very funny ngl

why're they blocking artist name anyway?

not like it really matter to AI server

[nova remnant]

So you don't just use people's art styles

And be creative~

[oak spoke]

tomato

[nova remnant]

The fun police is here

[oak spoke]

that's why you don't use online one

and use localized instead

[nova remnant]

Give me a 4090

No just a 4080 will work

[junior quest]

nah just jump to 4090

[oak spoke]

I don't even have that

🤣

I'm using casual 3060

here more tomato

[nova remnant]

[pastel fern]

Product shoot, American aromatherapy on washbasin ::4
Beautiful young woman, washing up next to it ::3
Bathroom with warm lighting, high definition details ::2
Soft focus, aesthetic scene ::3
--s 250

[haughty lynx]

#1047610792226340935 bots currently offline, you can run it locally or use online services like https://seaart.ai

[oak spoke]

this 2 is just perfect lmao the one giving speech and the one recieving it

man I love how far AI has come since I was last here

[nova remnant]

[nova remnant]

Have you seen sd3

[oak spoke]

yea waiting for it I still think it didn't really beat pony in term of style but it def got better lighting

[nova remnant]

What you mean

SD3 will literally fix everything

We just have to wait comfy is holding it hostage so he can gen all fennec girls before us

general FYI

in short ... autismmix brings more recent advancement

[nova remnant]

What am I even reading

Is this English

[haughty lynx]

Is this AI hallucination result

[nova remnant]

@supple raptor does the words in that picture make sense?

[nova remnant]

thats about the differnce between pony and autism mix models from huggingface chat bot

[haughty lynx]

the model treated "AutismMix" as "Autism" + "Mix" lol

[supple raptor]

The words form ideas that make sense, but they were clearly written by an LLM that had absolutely no clue what either model is so it just hallucinated something.

[nova remnant]

i wonder...

[haughty lynx]

and thought the two models are about autism detecting

[nova remnant]

huggingface chat bot failed?

how could it lol

the writing sounds very scintifcy lol

[supple raptor]

I don't think hugging face has their own chatbot. I'm assuming you're using a chatbot on a hugging face space which is just a hosted instance that someone set up.

[nova remnant]

yes hosted there, this one https://huggingface.co/chat/

[supple raptor]

And this is the biggest danger with trying to replace search engines with chat bots. LLMs function by calculating the most likely next tokens, which leads them to sound confident about things they have no clue about. "I don't know" isn't really part of their vocabulary in most contexts.

[nova remnant]

btw can you explain the key differenece between pony and autism, i'm thinking they are both the same model type?

[lyric zealot]

This is definitely hallucination by llm

[supple raptor]

Pony was fine tuned on top of SDXL.
Autismmix-pony is simply Pony with some LoRAs merged into it.
Autismmix-confetti is the above with a model called confetti also merged into it.

[supple raptor]

Again almost entirely made up.

[nova remnant]

ahh ok, so im now thinking i should download pony model lol

[supple raptor]

Autismmix-pony is better in every way. The LoRA merging fixed a lot of problems that Pony itself has.

[nova remnant]

oh wow ...

[nova remnant]

What problems

thanks for explaining @supple raptor .. guess i should be happy with autismmix for now

[nova remnant]

style swings? bit unstable?

[supple raptor]

"Jpeg artifacts" are lessened, the base style when no artist is prompted is more stable, body proportions are generally more correct, it's images are less desaturated, score tags are (slightly) less important, first token of the prompt isn't overemphasized as badly.

[nova remnant]

What are jpeg artifacts

I don't know what they look like in images

[nova remnant]

I don't mind that, it just sometimes goes into weird 3d stuff

[supple raptor]

Zoom in close to the outlines around her face. See those weird colored speckles in the blue of the sky and water just outside of the face outline? You get those same type of artifacts when compressing jpeg images too much so that's the best way to describe it even though the model is just generating them like that.

[nova remnant]

Hmmm

@lyric zealot

It's the glowing stuff

It also happens on autism

[nova remnant]

naww not so often

[supple raptor]

It does somewhat, it's less than on Pony though.

[nova remnant]

Ye

I found autism much more boring to use

Kinda of need artist tags on it

[supple raptor]

Just throw an artist wildcard at the beginning of your prompt.

[lyric zealot]

Yup yup, happens on autismconfetti

Sorry I’m eating so will late reply

[nova remnant]

Banned

[supple raptor]

Time for me to go sleebgy now

[nova remnant]

I'm not using my PC

And also

That would just do people's art styles

I don't find that fun x3

[nova remnant]

i picked an artist thats more sketch oriented, but you can see the colors noise or outlines are not as harsh

[supple raptor]

Don't be a poor

[nova remnant]

I'm reporting you for bully

[supple raptor]

Throw it on there 3-4 times. Boom unique style on every gen.

[nova remnant]

https://tenor.com/view/but-wait-nope-gif-14217892

That's why we need sd3

It will fix everything

[lyric zealot]

[nova remnant]

different artist, a lot less artifacts

[lyric zealot]

Report denied

[nova remnant]

We must bring down the stupid oppressor nubby that types slow and likes ugly tohou

im not even running hiresfix

[nova remnant]

I like pony style more

I have access to autism (the model)

[nova remnant]

that's cool, but for me pony images could be bit finer .. they look like hard sketch

[lyric zealot]

when nubby read what you typed

[nova remnant]

how clean is that?

[nova remnant]

Well yeah it's more artsy

Something something

yeah its more fun that way when drawing girls

The nose always looks weird when you mix realistic into it cx

i dont think pony is bad, its just bit different

@lyric zealot stupid

[lyric zealot]

as stoopid as you

[nova remnant]

Booba

some artifacts on her hair

[nova remnant]

uhh lol

see girls have booba and boys dont have booba .. that's the reality we must accpet

btw some of the realistic magazine cover photos done with pony are amazing, but not sure if thats due to lora they used

[nova remnant]

[junior quest]

@supple raptor is it fine to add tons of captions? like is it bad or good?

captions like this? will it ruin the model or lora whatever we create?

[native halo]

thats just bad dont add too many words when training a lora

https://tenor.com/view/the-hell-is-that-gif-27667584

[steady grail]

guys, there is a way to know what the wildcard picks?

[lyric zealot]

if you mean for that specific generation, then maybe check the pnginfo?

[steady grail]

nag, still show the wildcard

[lyric zealot]

it shouldn't

post pic with pnginfo

[steady grail]

i'll gen something with wildcard

[lyric zealot]

[steady grail]

oh right, the gen, sorry, i'm a little sluggish today because allergy

[keen hollow]

do ya'll use Kohya's hires fix?

[steady grail]

no

[lyric zealot]

i don't but i think some of the people here do, like nubby

[keen hollow]

The "✨" button has me confused

was thinking this was a Forge feature, but looks like straight Auto has it, too

the output results are just different than running hires straight away, and that makes no sense to me 😐

[steady grail]

maybe is a simple latent upscale without all the fixes of highres fix

[keen hollow]

it's supposed to use the same settings

i can set it to latent and see if it comes out the same

[steady grail]

on comfy latentupscale is just a single step

[lyric zealot]

is the differences like huge

or

is it like xformer kinds of differences

[keen hollow]

it's minimal difference, BUT, consistent repeatable difference

At a high level, it should just be taking the output and putting it through img2img, the same process in both instances.

[steady grail]

yeah, that's what latent upscale does

[keen hollow]

but the hires fix comes out one way (consistently) and the button comes out another way

tried latent and it was way off, so I don't think that's it

lemme generate an example

[keen hollow]

Original

[steady grail]

looks like the extension didn't load the wildcard

[keen hollow]

Button upscale

hires fix during generation

probably could have chosen an example with more obvious differences lol

anyways, you can run either process multiple times, and the differences are consistent

[lyric zealot]

yup, that's what i was thinking about

[lyric zealot]

i see your point

[steady grail]

i suppose i need a different extension

[lyric zealot]

i think it might be how denoising works for hires fix

maybe using the button, it's using a separate random seed for denoising

and such, will never be the same

[keen hollow]

not sure why they would implement a separate denoising workflow.

[lyric zealot]

it might be because

wait

[keen hollow]

looks like it should be same seed and all... https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14598

[lyric zealot]

i was literally looking at this lmao

[keen hollow]

the differences are so minute, hard to get a good example to share

it's easy to see in my image viewer flipping between them but .., yeah

anyways it was just weird, curious i guess

[steady grail]

there are no working wildcard extensions on forge?

[lyric zealot]

well, i guess they didn't do the seed thingy properly, not sure

[keen hollow]

Yeah dunno... It's so close. I need the Auto guy to explain it to me 😝

[lyric zealot]

wait, is dynamic prompts not working?

[steady grail]

dynamic prompts have anything to do with wildcards? it's all weird

[lyric zealot]

i didn't download that for forge, since i didn't use much with wildcards

there's also that Umi

what were you using previously?

[steady grail]

https://github.com/AUTOMATIC1111/stable-diffusion-webui-wildcards

[lyric zealot]

this is the first time i saw this

[steady grail]

i try with the other one

what a mess, i already hate this extension lol

[lyric zealot]

which one did you go for

[steady grail]

dynamic

the other one looks not working any more

[lyric zealot]

good luck

[steady grail]

still don't understand what's the online function

maybe the problem is that the wildcard is in excell and not txt

[lyric zealot]

online function?

[steady grail]

yeah, the extention say it have an online function, but i have no idea what it means

[lyric zealot]

not sure, never really used that so called online function

[light quail]

you should have multiple sets of captions for each image, and then during training a set is randomly selected each time it runs the image through the model

[bitter latch]

haha big head

[green temple]

Cool

Guys, I have a question , should I delete anything from the c drive to get more quality, cuz I heard something like that

[bitter latch]

shrugs

never heard that

[green temple]

Np

[supple raptor]

Afaik there is no hack to get around CLIP's 75 token limit in training. Meaning, it most of that would be ignored.

[oak spoke]

pretty cute no?

[primal wharf]

Big eyes fit the definition of "cute"

💀

[zinc flax]

Who will fix her

[green temple]

if anyone know , pls help

[zinc flax]

No..? Don't think it'll change the quality

[nova remnant]

@nova remnant pony xl v6 is very good at following prompt, especially with 2 girls interacting intimately , no anatomical deformity 👀

[primal wharf]

Hitman dual wield

[nova remnant]

[nova remnant]

score settings are bit different tho

cant put scores in negative box

[lyric zealot]

what do you mean by that?

[lyric zealot]

holy eyes

[primal wharf]

NovelAI

[nova remnant]

usually with autsimmix i put score_6, score_5, score_4 in negative box but with that in pony the image comes out extremeley contrasty

[lyric zealot]

when SD3 get released, we can hopefully gen at this quality like it's nothing

[lyric zealot]

oh, hmmm, i think for pony model might need a little more negative prompt than autism

but well, im usually on autism so i don't know much about pony tbh

[primal wharf]

Would be cool to have heterochromia combined with gradient eyes. So a total of 4 colours. Imagine the left eye being red and blue, and the right eye being green and yellow.

[lyric zealot]

[nova remnant]

@nova remnant im starting to love the art style with pony .. there is difference between autism and pony even with same artists style...

[primal wharf]

Or technically 6 colours, as the shade between red and blue, and shade between yellow and green/

[lyric zealot]

i think that level of complexity will be close to impossible for just a prompt

but hopefully im wrong

[primal wharf]

custom lora maybe

and we are still far away from true AI

CS style gun skin lol

whatever that is called

[nova remnant]

hmmm

i wonder if pony is at all following arists name mentions

[primal wharf]

@lyric zealot it almost did it, she has heterochromia and her left eye has two colours. Red and blue.

[nova remnant]

cause this is not source_anime, ebifurya

[lyric zealot]

i like the glass tube on her vest

[primal wharf]

A type of weapon similar to molotov

But for use against vehicles

[lyric zealot]

or just a very smelly liquid in it, so she can stay awake

[primal wharf]

"Nutrients"

[lyric zealot]

[primal wharf]

Would you like to buy some drugs, onii-chan?

throws up kitsune gang sign

[zinc flax]

Confetti is nice

[primal wharf]

Red for heal, blue for mana, purple is not a drug, it's an anti vehicle weapon.

[zinc flax]

Though I'd use panime to play around with styles

[lyric zealot]

https://tenor.com/view/kanna-pay-money-gif-24746936

[nova remnant]

i dont get it cat...

pony is not following artist styles that i mention

autismmix on the other hand is spot on

but despite that, pony image results are interesting

[zinc flax]

bwf = lineart style
iun = chibi style
fhy = anime style
zpx = 2.5 ish
fku = thick lineart

Try this lol

[nova remnant]

sleepti i mean actual artist names

[zinc flax]

Maybe try increasing the weights to see if it'll help

[near elbow]

sqwaaaak?

mad miku

[zinc flax]

Though I think most ppl use confetti/panime instead of the actual pony v6 lol

[nova remnant]

love that autsim mix model, very aesthetic but i was curious about pony lol

[zinc flax]

To this day I haven't tried base pony

Started off with confetti lol

[vital raptor]

pony is a base model, the intention is to train/merge/etc with pony as a base, but not really for general image gens

[nova remnant]

ahhhhh man no kidding that explains it

[primal wharf]

@lyric zealot also imagine heterochromia + gradient eyes + star shaped pupils in SD3

Mogu mogu, ~Okayu!

[nova remnant]

this is autism, every image i generate between pony and austim are clearly telling which is which

[nova remnant]

where did that come from

[nova remnant]

she thought art styles

those work too but yeah i was talking about artists names

[zinc flax]

Some hashed styles that I find interesting

[nova remnant]

for me when i generate images girls are top priority or i wouldn't be playing with this lol

[zinc flax]

I think the only model that can actually do artist names is animagine 3.1

[nova remnant]

autism mix is good at following artists too 🙂

[zinc flax]

But the model suuuuucks at anatomy

[lyric zealot]

hopefully

[nova remnant]

@zinc flax this .. #🍥｜anime message

[primal wharf]

why the Kizuna smug lmao

[nova remnant]

i get beautiful poses

[nova remnant]

ugly

[lyric zealot]

why not

[lyric zealot]

[primal wharf]

it's not lewd

[zinc flax]

I only recognize mdastarou, not sure about the rest though

[nova remnant]

its in the faces

i tried to distinguish facial differences

faces are like a major part of bringing rest of the image aesthetics for me

[lyric zealot]

i didn't say it is

[primal wharf]

lmao

[zinc flax]

Animagine gacha lol

Hoping that someone will do another finetune

[nova remnant]

grabby

[zinc flax]

no grabby

[lyric zealot]

[nova remnant]

so i loved this mode animaPencilXL_v260 before discovering autismmix

now i feel badly about deleting animaPencilXL_v260

i don't think its obvious for many how powerful and versatile these pony/autismmix models are

once you've gone in with it there is no turning back

you can't settle with any other anima

seriously after generating each images with autismmix i go into contemplating about sd1.5 models

it should be a general policy to promote pony/autism models on civitai for anyone who wishes to play around with anime

if not its like settling with technological impotence

the level of aesthetics, the artistic allure is just too impressive with autism model

i dont mean just revealing clothes on female subjects, its rather about the whole composition

[supple raptor]

NAI with hires fix+face detailer in kxl-delta got kind of close. it wants one eye to be red and yellow and the other blue and green though.

[nova remnant]

im pausing to breath in awe

[primal wharf]

nice

[zinc flax]

Can't fight NAI

Literally a goated model

[supple raptor]

funny enough, NAI missed the red and kxl fixed that.

[nova remnant]

the kind of default dynamic outputs im getting with autismmix i kinda stopped playing with controlnet

She fell, died

[nova remnant]

she landed on a lake just a few feet under

[steady grail]

she can fly beating her arms

[nova remnant]

👍

[nova remnant]

post them in your civitai page

and give me the link 🙂

I don't post anything there

omigawrsh why not?

Too lazy for anything like that

what do you do with sweet images?

delete or archive locally?

I post them here

And there

but they are not retrievable as easily you could on civitai

all of the images i wanna keep go up on my civitai page.... have something like over 1k images now

[raven rivet]

Hello guys!
I just came across the works of @Kben_ai, and I'm just impressed by how faithful his images are to the anime style; it looks just like a screencap! Does anyone have any ideas about the workflow used? Like on training the lora, the prompt, the model, all that stuff.

Thanks in advance!!

because all my results are absolutely not like a screecap anime style

[silent kraken]

Share your account name

[silent kraken]

Is it a screencap Lora? I'm sure if you try searching screencap Lora workflows you can get

[raven rivet]

all the loras i found for screencaps are clearly not that good

[steady grail]

what's this screencap?

[raven rivet]

i don’t understand your question

[steady grail]

i know about lora, about locon, but is the first time i ear the term screencap

[raven rivet]

oh screencap is just a style that is super faithful with the original anime style, like it was a screenshot took from the anime

[nova remnant]

same as my nick here

[raven rivet]

screencap just means screen capture

but it’s just to talk about the anime style

[steady grail]

if i look about this term on civitai i don't find anything.
but i understand what do you mean

[raven rivet]

sd images are always too detailed, i search to do a minimalist but faithful to the original anime images

[steady grail]

faithful lora are hard to do

[raven rivet]

is there some good comfy ui workflow for anime images?

[nova remnant]

check here https://openart.ai/workflows/academy

its a big hub for comfyui users

however i will admit, comfyui workflow goes over my head

[steady grail]

i should prepare an xl workfloww asap

forge is fast and easy, but lack many things

[nova remnant]

im somewhat happy with a1111

[vital raptor]

A1111 does everything I need. A better inpainting interface would be nice but it’s not worth changing UIs for it

[raven rivet]

i prefer a1111 too but i can’t get to find a good workflow 😭

i can only find shitty videos on YouTube

and the workflow is not adapted for anime images

if someone have a good guide, a good workflow or something, it would be super nice!

[nova remnant]

workflow for a1111 you say? i basically keep a style template that i click and use then start writing prompt

[vital raptor]

It doesn’t matter if it’s anime or not, that doesn’t change the ‘workflow’

[steady grail]

what do you expect from an anime gen workflow?

[raven rivet]

yeah but the image is come out as bad quality, right?

like 512x512

[vital raptor]

Pick a good model, put in a prompt and click generate

[raven rivet]

how do you upscale anime images?