🍥｜anime

[lyric zealot]

really liking the lora Nubby recommended

[nova remnant]

Smol

[supple raptor]

drawing LoRA is great. I don't think I've found any models it doesn't look nice on.

[nova remnant]

Nubby

[supple raptor]

Cat

[lyric zealot]

yet to try it on other models, but it's definitely working well for gishiki

[nova remnant]

Steve

Get more light

[supple raptor]

caught in the act of robbing you

[native halo]

i like the darkness 🌚

[nova remnant]

The moon will brighten you

Just wait

[supple raptor]

It's really nice on hell5-dpep

[lyric zealot]

will have to play the model more

[nova remnant]

why smol bonk me

[supple raptor]

make sure you use ctd's negative with that model. it looks pretty bad with my normal negative, but her negative makes it one of my favorites. easynegative, (etone:0.8), (worst quality, low quality, normal quality:1.5), (simple background, white background, abstract, abstract background, uneven, double exposure, odd, disconnected:1.5), holding, teeth, tongue, nsfw, midriff, ((multiple girls, 2girls)), ((watermark)), ((no humans))

[nova remnant]

etone is an embedding right

[supple raptor]

this one is also hell5-dpep with drawing lora.

[lyric zealot]

that might be why im always getting weird result, or not that good result with her models

[supple raptor]

yesh. easynegative and etone are the two negative embeds for that one.

[nova remnant]

what is etone

What does it do

[supple raptor]

yeah that negative is the key to them. I can throw her loras at almost anything without an issue, but need her negative to use her models. some parts can obviously be edited/removed as needed for the prompt.

[lyric zealot]

thanks for the tips

[supple raptor]

adds more gud, removes more bad.

[nova remnant]

I'm having giga bad bodies on my gens

[supple raptor]

You said you started genning at higher res recently instead of Hires Fix right? That will often cause anatomy problems.

[nova remnant]

I only have access to these

[nova remnant]

I'm using a free service right now

I asked them to add the aurora embeddings that I like to use maybe they do

[nova remnant]

Thief, get her! or get me

[supple raptor]

When you get back on your own PC, here's etone since I don't rember where it comes from and I'm too lazy to look right now. Depending on the model, easynegative with negative_hand-neg should be a decent combination from that list. easyneg will force it's own style, but it's not a bad style.

[nova remnant]

also nubby

[supple raptor]

maybe like 0.8 strength for easynegative and 0.8-0.9 for negative_hand-neg.

[nova remnant]

This thing seems to be using an intel xeon platinum thing (expensive as hell I looked up)

can it really gen on it ?

It has the use cpu arguments and no gpu in the specs sheet

[supple raptor]

yeah, you can gen on a CPU. You just shouldn't it'll be slower than your GPU. I think a 10700K is like 45 seconds for a 512x512 image.

[nova remnant]

@lyric zealot

[supple raptor]

the good news is that she seems to have given up on robbing you. the better news is that she seems to have decided to move in.

also, you can tell Miku is Japanese because she took her shoes off to come in and rob you.

[nova remnant]

@supple raptor I can do these and images come out in 1-4 minutes, generally 2, depends on the queue

[nova remnant]

I'll try these

[supple raptor]

Yeah, I'd expect significantly slower speeds on a CPU. Inference is a lot of repetitive math which is what GPUs are especially good at. CPUs are good at random math, but suck at repetitive math.

[nova remnant]

(worst quality, low quality: 1.4), (bad anatomy), extra digit, fewer digits, (extra arms, bad hands, text, watermark:1.2), think some of these are useless if I add the negative TIs?

[nova remnant]

my thief is staying and eating your chips

[supple raptor]

I doubt that basically any training images were ever tagged with bad anatomy, extra digit, or fewer digits. SD doesn't know how to count so extra or fewer digits are essentially meaningless to it. It doesn't know what the right number of fingers is, it just has a basic idea of what it thinks a hand should look like.

[supple raptor]

I will share my chips and offer headpats.

[nova remnant]

SD: Now its time for the fingers
Rolls dice: "4" its above 3 lets add one......

[supple raptor]

[lyric zealot]

[nova remnant]

easy negative or easy negative v2 ?

[nova remnant]

I'd pat her big...

ears and tails

[supple raptor]

the issue is that what SD sees is around this fidelity. it's a miracle that it's as good at hands as it is honestly.

[nova remnant]

[supple raptor]

I think the original is better personally. v2 forces it's style even more without really improving the quality imo

[nova remnant]

[lyric zealot]

surely you know who she is

[native halo]

is that famous anime character vegeta?

[supple raptor]

https://tenor.com/view/neuro-sama-ai-blink-gif-3099804956727570048

[nova remnant]

If she doesn't sell chips I don't know

[lyric zealot]

omg, how did you know

[lyric zealot]

https://tenor.com/view/neuro-neuro-sama-neuro-funny-neuro-sama-funny-neuro-sing-gif-14807952765047839726

[supple raptor]

https://www.youtube.com/watch?v=X3pUEaFtgB4

[lyric zealot]

[nova remnant]

clicks on metal song video

that might not be my style

[lyric zealot]

How is there no Evil Neuro Lora yet

[supple raptor]

she's an AI vtuber. runs multiple models to behave similarly to a real vtuber and also does karaoke

[nova remnant]

AI ?

so thats why the name is neuro

[lyric zealot]

[nova remnant]

interdasting, never watched a vtuber (except one person which doesn't really count lmao)

[supple raptor]

here's a good cover of Bury the Light since that definitely isn't Neuro's best cover. https://www.youtube.com/watch?v=SuwAwrzfUVs

[native halo]

thats not an anime girl ban

[lyric zealot]

i still like this, probably like one of the best, maybe because it's animated with MV and stuff https://www.youtube.com/watch?v=nGmwH2ZYlP4

[supple raptor]

neuro is the only one that I watch aside from random clips.

[supple raptor]

yesh this is a good one

[lyric zealot]

i was watching that live on stream when they were revealing Neuro V2, and the concurrent viewers at that moment is crazy high compared to normal stream

[nova remnant]

How does it work

[supple raptor]

nice! i missed the live stream, but watched the vod as soon as it was up.

[lyric zealot]

how does neuro works?

[nova remnant]

yeah

[supple raptor]

lots of AI models all working together.

[lyric zealot]

yup, a lot of them

[supple raptor]

the core is a txt gen that feeds into txt2voice, but that is an extreme simplification.

[lyric zealot]

well, also depending on a lot of stuff. Neuro can play osu, and it's the top player in the world(not sure if currently still is), is learning about among us(unsure if done), and can react to video, pause video on her own, react to messages from stream, and collab with other vtubers, and talk to them

[nova remnant]

huh

[nova remnant]

Wait

AI playing among us

that seems funny lmao

[supple raptor]

sus

[lyric zealot]

well, I think Vedel(the creator of Neuro) is training the AI to play that

and or coding

[supple raptor]

this one is one of my favorites covers of hers https://www.youtube.com/watch?v=mhuQ_-UCbsg

[lyric zealot]

whenever i watched their duet, I always see Neuro bouncing around, like a hyperactive child, and evil just really shy but wanna be like her sister.

[supple raptor]

I see it as evil neuro being the one taking it seriously and forcing her hyperactive sister to perform with her.

[lyric zealot]

that's another way to look at it

[supple raptor]

which xl model are you using these days?

[supple raptor]

is this the lora that you're using? https://civitai.com/models/80754/neuro-sama-v2

[lyric zealot]

hold on let me find it

[slate narwhal]

KohakuBeta7, is the main one I'm using, a model trained over 1.5m images. I'm doing refining tests with MergeHeaven (just released), really nice model on it's own. Honorable mention to AnimeIllustDiff, a model I keep coming back to, as they've trained a LOT of different styles of modern Pixiv anime artists separately inside the model (and it works just by prompting), problem is it has really bad quality, but it has a very promising future. I'm trying to mix it with better quality models, but didn't get any decent results yet.

[lyric zealot]

help i have no idea which one is mine

[lyric zealot]

yup, that's the one, I don't know mine is named something like neuroTEST

[slate narwhal]

MergeHeaven includes NijiXL_SE; Niji is a good model on it's own, but it has problems

[lyric zealot]

but i checked the sha256 and it's the same as 1.1 version of that

here's one with metadata

bad fingers tho so

[supple raptor]

Nice. I was going to recommend you try KohakuBeta7 if you hadn't already. It's looked pretty nice in all of the gens I've seen from it. I'm still having a hard time convincing myself to give up all the 1.5 loras that I like to use and take the hit on it/s to give XL a real try.

Now that I have Kakigori_v3 out maybe I'll start experimenting some more with XL.

[supple raptor]

nice, I'll grab that one then. yours looks a lot better than the one that I had before.

[lyric zealot]

i still have one for V1 Neuro, but it's bad

[supple raptor]

I think that's the one I have probably.

It can make her, but it's overfitted AF

[lyric zealot]

yup, i vaguely remember something like that happening with that lora

[slate narwhal]

It is slowly getting there, but SD 1.5 is still better I think. I'm doing a lot of extra tricks to get decent generations without any upscales, it's not easy. The recent released XLOutline lora helps by improving a lot of models

[supple raptor]

I'm really hoping that WDXL won't be a letdown. Both for all of our sake, and the team that's working on it. They've put an insane amount of work into figuring out the right training parameters and fixing the dataset.

Previous WD models were pretty much just tagged by AI and sent with a few runs to find acceptable parameters, but they've actually put work into modifying the captions this time around.

I can only imagine how disheartening it would be for them if they can't get something good out of it this time around.

[slate narwhal]

In previous posts I was doing some tests using the Russ Mills art, it blurs everything, ruins the eyes, basically ruins everything, but using the XLOutline Lora + refining tricks with other models, (plus all the other stuff like SAG), it does the trick and works, making it almost acceptable (eye problems still persist though)

[slate narwhal]

Yeah let's hope they can make it work. SDXL training is really difficult. Even after training, several stabilization merges are needed to get quality

[supple raptor]

I've been keeping an eye on your gens. Yours are some of the only ones that I know are XL gens that catch my eye. Most of the other XL gens I see still look like either a mediocre 1.5 model or just bad style.

[slate narwhal]

Yeah, but it's like slowly sculpting a rock, not easy; there's so many stuff and tricks going on behind the scenes that it gets difficult to explain

[supple raptor]

It seems that way. that's been what's turned me off everytime I've tried it so far. It's a hard sell to convince myself to dive in when I know I could just jump back to a 1.5 model and get a better image (in most cases) in less time.

[slate narwhal]

but you are right, most SDXL gens are mediocre. Eyes are the main problem for every single anime gen out there, and if it's eyes at distances SDXL fails miserably

[supple raptor]

I guess I'll just have to wait for the NAI XL model to come out and then I'll roll up my sleeves and antagonize TabloidA again and see if we get another leak

last time I did that, the NAI model leaked less than 4 hours later

[zinc flax]

NAI 2.0 leak would be huge

They improved on the hands big time

[supple raptor]

2.0 is just an sd1.5 finetune. it's a lot better than NAI 1.0, but it can't really compete with our models that we've already finetuned on 1.0

[nova remnant]

@lyric zealot give hug

@supple raptor something is weird

not even my pc does bad hands as some of these

I'm using breakdomain atm, but most of the ones I tried are just as bad

[supple raptor]

hmm.. try getting rid o the negative-hand embed and see if they improve

in the meantime, have a raphtalia

[nova remnant]

I was genning without them before, about 20% luck with the hands

I'm trying a prompt s/r with the negatives in various weights to see

hopefully I did it right lmao

[supple raptor]

just do a prompt matrix with like 25 different entries. it'll be fine, I promise

[nova remnant]

kek

[nova remnant]

https://tenor.com/view/hello-oh-hello-fancy-cat-cat-kitty-gif-7364490

[native halo]

i wanna lift

her hat

[nova remnant]

leave it alone

[nova remnant]

more hat

[nova remnant]

scawy eyes

[magic wave]

But cool eyes

[magic wave]

Kobeni

[nova remnant]

needs a hat

[nova remnant]

@silk narwhal morning

why she sad

[silk narwhal]

good morning, perhaps she got thrown into the middle of the ocean as a sacrifice by pirates?

[steady grail]

"why i'm wet?!? that makes no sense!" the klutz
"my hair!!! why the hair? they where innocent!" the fashion girl that just left 3 hours of salon a minute ago.
"the water is cold !" the picky girl

[magic wave]

Maybe

[nova remnant]

Where the smart people

how bad is clip skip 3

[silk narwhal]

experiment with a matrix i would say

[silk narwhal]

found the exit?

[nova remnant]

🐺 woman

[nova remnant]

I was going to ask if this has PNG info

But might be a dumb question, I'll check later

Are you using controlnet for the pose?

[silk narwhal]

nah..was using the same seed as previous gens

[nova remnant]

That's very dark

this is like

The cutest vampire I've seen

[nova remnant]

[near elbow]

"a generic anime image" with sdxl 🙂

arcane style

[nova remnant]

Main guy, spiky hair, uses a sword, tries to act cool

generic indeed

Meow on the side

[near elbow]

"Meow on the side" as a prompt with sdxl anime style 😸

[nova remnant]

[dawn jasper]

This is a neat pic.

[untold glacier]

https://tenor.com/view/sunmanosdetijeras-gif-23377892

[nova remnant]

Am I invited for cheese

[bitter latch]

https://tenor.com/bhLMl.gif off phones and omg

[bitter latch]

doing these using XL, next one's gonna take a moment or 30

[native halo]

clip skip 20 👍

[stable crane]

she's your new mousewife, ofc you're invited for cheese

[nova remnant]

Creepy

[lyric zealot]

https://www.reddit.com/r/StableDiffusion/s/CiYeqO5hy3

^ might be interesting

[nova remnant]

smol my friend

What is that

[lyric zealot]

A different type of model that only requires little steps for good generation?

[nova remnant]

Note that LCMs are a completely different class of models than Stable Diffusion, and the only available checkpoint currently is LCM_Dreamshaper_v7.

[native halo]

i see no anime here

7 week mute / ban imo

[nova remnant]

@lyric zealot gonna keep an eye on that one, great share btw!

and thanks ~

[lyric zealot]

Yup, might be interesting if what they claimed is true

[nova remnant]

its going to evolve

[lyric zealot]

im thinking what's the downside tho, if there's any

[nova remnant]

Steven

I'll save you and this chat

with the power of booba

she came to give you breakfast

[native halo]

id like something to drink

[untold glacier]

[nova remnant]

its just the bartender why bonk

[untold glacier]

The juxtaposition of "booba" and "breakfast".

[nova remnant]

@untold glacier maybe you need to go outside more, look at that view in the window

[native halo]

https://tenor.com/view/anime-milk-cat-anime-milk-gif-23177860

[untold glacier]

[nova remnant]

the outside

[native halo]

i love the massive trees

[nova remnant]

Spooky scary skeletons

[supple raptor]

dead chat

[native halo]

https://tenor.com/view/shocked-shook-onizuka-whats-happening-gif-17248402

[nova remnant]

@native halo since its so empty here

we need big things to fill it

@native halo forgot vae

many pumpkins to fill the room

what you think of the idea

[native halo]

i like the big juicy pumpkins

[nova remnant]

@upper slate

[upper slate]

[nova remnant]

did you like the pumpkin I brought

[upper slate]

Yes

The pumpkin

[nova remnant]

Make ice wing girl happy

[nova remnant]

Christmas wing girl happy too

[supple raptor]

can't, her name is チルノ which is pronounced chi-ru-no... cheer no... no cheer... only sad

[nova remnant]

Cheer no stop

[upper slate]

Why you guys making underage lolis

[native halo]

https://tenor.com/view/kurono-fireforce-anime-mooneyes-gif-26035604

[gusty palm]

cd tuner is still doing weird stuff

in img2img

[native halo]

yea same here sometimes noise appears even when disabled i guess its still buggy or maybe another extension conflicts with it

[gusty palm]

ill try genning without cd tuner than img2img with cd tuner

[supple raptor]

Cirno, Flandre, and Remilia aren't underaged, but lewds aren't even allowed here so what's the problem with genning an underaged character even if they were?

[upper slate]

Why the underage character has b~ big melons

Sussy

[native halo]

id have to hit my head against the concrete in order to reply to w/e u are asking

[supple raptor]

ah, just a troll. k then. welcome to the blocked list.

[magic wave]

(it's a boy)

[bitter latch]

trying some new prompts

Miku beautiful abstract image side shot close up of a woman's face, classic neurographic art design on her face and hair, detailed; ink 4k UHD, Hi-Res

tho I think by adding miku it changes it

[dawn jasper]

Dunno who she is, she's blessed though.

[gusty palm]

dont get mad shes actually 666 years old

i gave up on img2img upscaling for now

[nova remnant]

I want cappuccino

[supple raptor]

The first Coffee Girl brought an extra for you

[nova remnant]

She looks like

Wait

4'9 (I think this is pretty short, murica numbers hard")

angry and smol

I was doing the good things in #🔆｜dailies 👍

[supple raptor]

Yesh, 4'9 is smol. I think like 5'6" is average female height.

5'4" is average apparently.

[nova remnant]

Whatever the average is I like it

[supple raptor]

In the US at least, I'm sure average height varies

[nova remnant]

short girls

@supple raptor do you think clip skip 3 would have anything to do with the bad anatomy

[supple raptor]

In general I would expect it to marginally improve anatomy at the cost of following the prompt less. It'll vary on different models though, you can run an x/y/z to compare different clip skip values on the same seed.

[nova remnant]

I can't change clip skip

[native halo]

clip skip 20 when

[supple raptor]

There aren't 20 layers of clip to skip

[native halo]

😔

then 12 at least do it for the fans

[nova remnant]

https://tenor.com/view/small-cat-angry-cat-cute-gif-24160051

[supple raptor]

You can do 12 in WebUI. That leaves only a single clip layer being adhered to though.

[nova remnant]

with clip skip 3 should I prompt more or less ?

euler a at 50 steps is saving most of the super weird the anatomy issues

euler a

[native halo]

PLMS when

[nova remnant]

PLMS is if you're going abstract x3

lora built-in

[native halo]

[nova remnant]

I made myself hungry

[gusty palm]

humgwy for bwains

blood isnt in my prompt, thats goop or slime or something

rendering at 512*1024 only takes about 75 seconds

[wet belfry]

Wow, this is insane!

[native halo]

whats your gpu

[wooden coral]

"only"

[gusty palm]

1060

im using AND_PERP Though so that makes it slower

[native halo]

🫡

i salute fellow gtx owners

[gusty palm]

:3

i have a 1050ti in a box on the shelf im eventually one day gonna put into a secondary dedicated server maybe

rtx stole the spotlight from gtx and i wish they could put some of those resources back into gtx progress

[native halo]

no love for us gtx owners 😔

[gusty palm]

frfr

[nova remnant]

what you mean ?

You got the 1600 series

👍

[native halo]

1060

[nova remnant]

Is any of the 16 like 1660 ti

Even better than a 1080?

Why do they exist

[native halo]

in SD even the 1060 is faster than the 1660

[gusty palm]

idek i dont udnerstand the 16xx series compared to the 10xx series

when i type it like that it sounds like the 16xx would be better but i thought 16xx were more of a 10xx revision series

[nova remnant]

@native halo green clothes like the nvidia logo

[native halo]

i am now green team

[nova remnant]

@native halo is the small amount of food like that on a date a deal breaker ?

[native halo]

doesnt matter i will eat regardless

[nova remnant]

lmao

Sometimes I wish this extension didn't exist

[native halo]

https://tenor.com/view/kirino-oreimo-gif-7514170

[nova remnant]

candy

[native halo]

thats not my sister

[silk narwhal]

it is now

[native halo]

https://tenor.com/view/shocked-shook-onizuka-whats-happening-gif-17248402

[nova remnant]

my reaction

[native halo]

needs more dark

[nova remnant]

other than black background I don't have the power of darkness

[lyric zealot]

Red hair miku

[nova remnant]

@lyric zealot I remember red hair girl 👍 very nice

[nova remnant]

miku's looking to buy some almost questionably named fruits

[upper slate]

Is that the Frick fruit

[nova remnant]

yes... frick

[upper slate]

So like

is it used as a Frick toy

Or what

[silk narwhal]

Wut

[steady grail]

i weirdly love these

i little creepy but they i don't know what that tickle me

[nova remnant]

Would you buy dolls like that and put in your room

[steady grail]

no! i don't want more nightmares!

[steady grail]

my god, so cool, how did you do that?

[lyric zealot]

here's the metadata, credit nubby for recommending the drawing lora

[steady grail]

this lora is fantastic

[lyric zealot]

[willow canyon]

Is anyone here has webcolab link that work well with anime style, could you share me the link, please?

[shadow estuary]

[silk narwhal]

sorry .. i run everything local

[nova remnant]

I think I have this lora available

lets see

[lyric zealot]

use it and have fun

[nova remnant]

Asuna got in a bad state after those games and is now selling fruits

@lyric zealot probably not the same as your lora

[lyric zealot]

probably not

unless it's because the weight?

[nova remnant]

I put 1

same prompt as that market gen above

[lyric zealot]

I'm using at 0.6

[nova remnant]

I just click generate at 0.6

we have the same brain

[nova remnant]

[lyric zealot]

well, it's working great for me at 0.6

and the trained token is <lora:Drawing1:0.6> drawing, style by NTY

[nova remnant]

I snagged that from your png

[lyric zealot]

[nova remnant]

[nova remnant]

@lyric zealot wait I can check the lora hash

doesn't look like the same effect >.>

[lyric zealot]

yup probably not the same

or

it's the model

when she saw you browsing her r34

[nova remnant]

[nova remnant]

this is cool

very cute model and some ink style lora at low weight

[nova remnant]

[nova remnant]

is this a famous character

[lyric zealot]

not very famous but it's from manga/anime

[nova remnant]

I've seen her before but I'm bad with names

[lyric zealot]

https://tenor.com/view/yotsuba-nakano-nakano-yotsuba-yotsuba-nakano-muri-muri-gif-21901727

it's from the anime called 5-toubun no Hanayome

[nova remnant]

@primal wharf cute hanfu girl

[silk narwhal]

You get some candy and you get some candy ... everyone gets some candy

[nova remnant]

interesting clothes

[silk narwhal]

wha

[nova remnant]

this one looks like shes wearing panties over the bunny outfit thing x3

[nova remnant]

hello boto, good place to meditate

[wooden coral]

Have some landscape in all the waifus

[primal wharf]

Nice

[primal wharf]

Nice pencil style

[lyric zealot]

Thanks to nubby recommendation of the Lora

[nova remnant]

bloo

[nova remnant]

Reminds me I need to gen some wallpapers!!

[wooden coral]

make some in 5120x1440

[nova remnant]

I have the power of nvidia clusters

that will take 30 seconds

https://tenor.com/view/chaos-gif-22919457

[wooden coral]

I tried before but in that res the images get very repetative

[nova remnant]

lmao

I can imagine

What if you do a really big prompt, would that help ?

[wooden coral]

idk you gotta try, maybe with regional prompting

[nova remnant]

Prompt ?

The one with mountains

[wooden coral]

Uhm I think all those have metadata

[supple raptor]

[nova remnant]

tried to grab but isn't showing

hmm

It did now, but on a metadata viewer website, not webui

[wooden coral]

works for me

[nova remnant]

o:

you're using neutral prompt

[supple raptor]

When grabbing metadata, make sure to either download the image or click the open in browser button, then drag and drop it from the folder on your PC or from the browser window into the PNG info tab.

[primal wharf]

What lora is it?

[supple raptor]

https://civitai.com/models/110244?modelVersionId=118849

[wooden coral]

Yeah im still playing around, installed it today after I saw someone mentioning it

gotta need to wrap my head around it

[supple raptor]

@wooden coral you wanted W I D E image?

[wooden coral]

Thats really cool, but more of a fruit thing

Also too small

[supple raptor]

I know, I just saw it because I posted it to that LoRA I just linked

[primal wharf]

Thanks.

[supple raptor]

Upscaling it easy.

[nova remnant]

that extension is awesome

put miku in the prompt and AND_SALT black hole :1.5

watch the world burn

[silk narwhal]

what extension?

[supple raptor]

https://github.com/ljleb/sd-webui-neutral-prompt.git

[primal wharf]

Tree becomes legs lmao

Her legs went through the 4th spatial dimension and came out from the tree.

[nova remnant]

only thing that tree needed was adetailer to work on it for a face x3

meow painting!

[nova remnant]

[silk narwhal]

lol.. trying SDXL again:

Time taken: 10 min. 39.0 sec. :/

[nova remnant]

[silk narwhal]

ok..to be fair it hadn't loaded the models into memory yet

[near elbow]

all relative...anime artwork a cute manga image . anime style, key visual, vibrant, studio anime, highly detailed Negative prompt: photo, deformed, black and white, realism, disfigured, low contrast Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2681507305, Size: 896x1152, Model hash: dce7eb8449, Model: realisticFreedomSFW_alpha, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: Anime, Version: v1.5.2 Time taken: 5.2 sec. A: 10.26 GB, R: 14.20 GB, Sys: 16.9/23.9883 GB (70.3%)

the big thing with SDXL is that you can change the style just with prompting, and the prompt understanding itself is excellent

just try this with SD 1.5 models and watch them mix up the colors and objects 🙂
"a blue dog sitting on a yellow rock, red tree in the background"

and switch the style to old film just with a few keywords (same seed)

and to origami 🙂

ok, enough SDXL rants... back to anime

[silk narwhal]

_<

[near elbow]

guess the prompt 🙂

[silk narwhal]

ugh..don't remember that mecha name

[nova remnant]

beautiful selfie of a girl in a bit of armor

[near elbow]

right, exactly that mecha - "gundam cherry blossom"

[nova remnant]

dalle turns every mecha into that one

unless you specify something about them

lmao

[near elbow]

same seed with "mecha" instead of "gundam" - SDXL knows the nuances 🙂

"neon genesis evangelion cherry blossom" ok... not as i expected

more like it

ok, off to other stuff, back later - cu

[nova remnant]

Mouse girllll

👍

[supple raptor]

Plenty of 1.5 models can do style change and color composition with just basic prompting too. This is AuroraONE. Not sure about the origami one since I've never tried that and don't h ave a prompt for it.

[nova remnant]

What is [monochrome::7]

[supple raptor]

using monochrome in the prompt only for the first 7 steps. It tends to help with sketch like styles to get more muted colors. If I cared enough about the prompt to put effort into it I would increase the steps that monochrome is present until only parts of the image are colored.

would probably increase the emphasis of sketch, graphite \(medium\) a bit too.

my realistic keywords could definitely use some work, that's not something I gen often.

[supple raptor]

This is the sort of effect that's intended with the sketch prompt, though these are using a lora, so they're cheating a bit. It's very possible to get a similar effect on Aurora though just prompting though.

[near elbow]

that's more than i used to get from a single SD 1.5 model, nice
the origami one is origami style a blue dog sitting on a yellow rock, red tree in the background . paper art, pleated paper, folded, origami art, pleats, cut and fold, centered composition Negative prompt: noisy, sloppy, messy, grainy, highly detailed, ultra textured, photo Steps: 30, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 2990599109, Size: 896x1152, Model hash: dce7eb8449, Model: realisticFreedomSFW_alpha, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: Origami, Version: v1.5.2

[nova remnant]

Thighs

[near elbow]

"professional photograph of xyz, highly detailed, background bokeh, cinematic lighting" is usually nice for 1.5

[supple raptor]

I expect origami probably won't work well on Aurora, but I'll give it a try.

[near elbow]

(offline again just in case you post something)

[supple raptor]

@near elbow it tried, but really it just gave me a PS1 graphics dog and rock XD

[nova remnant]

Lmao

That's some nice origami work

[supple raptor]

I'll try it on Kakigori_v3 now. I don't expect the different styles to work nearly as well on this one without some more help.

better than I expected. realistic prompt invented a dog rock though

[nova remnant]

New Pokemon

"dogrock"

[supple raptor]

ngl, better design than most of the pokemon I've seen from the last few generations. first gen is still the best.

[silk narwhal]

there is alway pokemon infinite fusions 😄

[supple raptor]

I think heart gold/soul silver are the last pokemon games I've played

[silk narwhal]

infinite fusions or whatever it is called is a rom hack that added every single gen plus two/three and 4 way fusions (believe 72,000 pokemons in the rom 😄 )

[supple raptor]

gotta catch 'em all

[silk narwhal]

176,000 possible combinations O_O

[supple raptor]

imagine trying to target a shiny version of your favorite pokemon with that many possibilities

[silk narwhal]

😄

[native halo]

now try with a nirvash style mecha instead

[nova remnant]

She's friendly right

https://tenor.com/view/scared-cat-running-funny-run-gif-23668419

[supple raptor]

https://tenor.com/view/anime-crazedlook-yandere-crazylove-gif-21784251

[native halo]

no

exactly

[polar lily]

I really like arts in the night time especially when in hot springs and the color scheme fits perfectly

[supple raptor]

Thank you! I was really pleasantly surprised with how nice that one came out since it was using 3 LoRAs at full strength which usually doesn't work very well.

[polar lily]

Glad it worked idk much about ai but it's one of my favorites now

[unreal ridge]

It gave up on the arm I guess

[lone hazel]

Anyone here who is good at creating high quality anime images? I’m paying, DM me if interested

[native halo]

i am

💸

[polar lily]

The details are something else no way

[stable crane]

been feeling like I've had cursed hands lately

[plain dragon]

What should i do next?

[stable crane]

research how to take screenshots using your computer instead of photographs

[native halo]

https://tenor.com/view/lucky-star-anime-konata-izumi-excited-pumped-gif-14835303

[plain dragon]

I am just watching movie in my pc

[stable crane]

there should be a guide ~somewhere maybe #1072220168534642768

[polar lily]

This crazy the hair and eyes are something else 😭

And the pink highlights made it better

What character is this

Evangelion girl?

[nova remnant]

watermelons are good for your health

[native halo]

gonna eat that melon

[nova remnant]

melons are super good

watermelons*

[stable crane]

what a good shop

[lyric zealot]

I'll take two

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

[nova remnant]

👍

More to come

I'm trying science here

Watermelon print on clothes

[stable crane]

we're blazing a trail for our ancestors

[nova remnant]

Damn a lot of YouTubers must've died ):

There are many thumbstones

I'll see myself out

[lyric zealot]

https://tenor.com/view/anime-sweep-push-getout-goaway-gif-12496440

[stable crane]

about the same in this case

[bitter latch]

Prompt: Hatsune Miku Watermelon Magazine cover, very detailed, watermelon dress
Negative prompt: NSFW, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, one tail, no extra legs, kkw-extreme-neg

neverendingdreamNED model. so yeah prob wrong model

[nova remnant]

@lyric zealot so I heard you liked foxes

👍

[bitter latch]

hmm little burnt

[terse ore]

Anybody use 3d software to make screenshots and then into anime images? I use realistic trees in 3d but after putting it into img2img they look neither realistic nor anime and have lots of artifacts and are just plain ugly

[upbeat wasp]

Probably would be better to use the screenshot as a softedge controlnet in txt2img rather than img2img. By the time you've denoised enough to get rid of the 3D look and artifacts, you might as well have done a full t2i

[terse ore]

I want consistency because I want to make multiple panels though. Perhaps I should try stylized 3d trees instead and see what it looks like

[upbeat wasp]

Get one set with the right style, and use IP Adapter to make more for other panels?

[native halo]

trees are not necessary to make in img2img what you should do is create a few tree images on text2img with a lora to get consistent outputs then if u wanna add characters just photobash the character then another img2img pass

[stable crane]

nice

[terse ore]

Hmm, these are good ideas. I'll have to experiment

[nova remnant]

Steven with the wise words?!

[supple raptor]

she tried cooking and burned the cabin down

prompt for that one was cirno, solo, puffy short sleeves, blue hair, blue bow, hair bow, hair between eyes, blue eyes, neck ribbon, blue dress, ice wings, white shirt, pinafore dress, short hair, red ribbon, (muddy:1.1), dirty, dirty clothes, dirty face, (raining:1.2), terrified, scared, horrified BREAK water drop, (darkness:1.2), night, (ghost:1.1), fog, outdoors, forest, pine tree, spooky, horror \(theme\), masterpiece, best quality <lora:[diffLora] lowexposurenoiseoffset:1>

[nova remnant]

Which sins do I have to commit to get saved by her

[supple raptor]

I already set it as mine

[nova remnant]

Needs hair over the screen

Had a dream, I died

End

Slept for like 10 minutes

[native halo]

you did died but you respawned on your bed

[nova remnant]

Woke up with my Booba generation done

👍

[native halo]

🌚

[near elbow]

voilà (same seed again)

bonus

[supple raptor]

Hmm. Maybe a really long scarf flowing across the screen would work better for this one. Maybe I'll mess around with some controlnet and see what I can come up with.

[bitter latch]

sighs "Tried to sleep, brain said no"

[bitter latch]

preview at about 50% ,Rocky Horror Muppet show as an anime on late night TV in Japan,

hmm not muppets

[bitter latch]

downloads the lora, muppet mania and tries again

[bitter latch]

oh the horror of failure

hmm.. o..k

gives up only breifly

[silk narwhal]

hmm different clip skip values

[oak spoke]

damn holding a sword perfectly

well the handle isn't perfect but eh

[oak spoke]

why's he so grossly muscular I didn't even put any muscular prompt in

[nova remnant]

morning

[dawn jasper]

Who is that person?

[lone hazel]

anyone know if this image is AI generated?

[frozen olive]

it is

[lone hazel]

thanks

i'm looking for people who can create images like these, you know anyone that can make this?

[frozen olive]

what do you mean "like this" in specific

[lone hazel]

this

like this style i guess

[frozen olive]

probably most of the people here can, if u ask me its low quality random generation.

ask here you might get way better pictures as the one u showed

[bitter latch]

it is, hands give it away

[lone hazel]

really?

damn

i want to make images like that, any tips on how i can get started

[frozen olive]

type stable diffusion guide in youtube u will find alot

[zinc flax]

I always look at the eyes to see if it's AI or not

[lone hazel]

how much do you think people will charge for something like that?

[supple raptor]

If they charge anything for that quality of AI art then you're paying too much. It's really not very good. I've seen people making better after a couple days of using SD. If you can't be bothered to learn then there's a plethora of AI art available online for free.

[regal pivot]

, There are millions of things available for free..
If you are not a super recognized artist, it is not a matter of how much you can charge, it is whether or not you find people willing to pay...
A person who has no idea about the subject could pay a lot, a more knowledgeable person would not even be interested in having it for free.

[nova remnant]

clothes are also a huge giveaway

things don't connect or are asymetrical x3

[lone hazel]

damn i see

[supple raptor]

The huge blurry artifact across the wall and window is another dead giveaway. Looks like some boring, generic anime model, using controlnet to copy a pose, with a low effort prompt, and not bothering to fix the problems.

[lone hazel]

i see

you got an idea of what model it could be? i would like to learn

[supple raptor]

Anything V5 if I had to guess.

[unreal ridge]

Just like the low effort horny gens that often show up on my twitter feed

[lone hazel]

thanks

[nova remnant]

Bunny tail looking like a stress ball

https://tenor.com/view/كسمك-cats-cat-paw-gif-11683988

[nova remnant]

Why are these images so different from each other

I don't understand upscalers

[supple raptor]

the tldr is that GAN upscaling adds it's own noise to the image which can often result in noticeable change, especially when the image is passed over again in img2img (which is exactly what hires fix does: low res diffusion, upscale, img2img diffusion on upscaled image)

[nova remnant]

Is that noise generally good?

All of these are adding noise, right

Hmm

Ultrasharp also seems to make the image different

I'm genning one without hires to see

What the original is

[stable crane]

niji?

I'm looking for some more good ideas to up the randomization of my prompts with wildcards

I've already got hair color, eye color, hairstyle, pose, orientation

[nova remnant]

@supple raptor out of these which one would you use for anime look
Would any of them help with hands by chance

[supple raptor]

in most cases it's better than simply resizing the image.

[supple raptor]

None will help with hands in any meaningful way, and they're not particularly good for anime. This is my current favorite for anime. https://openmodeldb.info/models/4x-NMKD-YandereNeo-XL

If you're stuck with just those options then either 4x-UltraSharp or R-ESRGAN 4x+ Anime6B will be the best in most cases. The R-ESRGAN ones will be slower though.

[gusty palm]

does the krita or inscape extensions basically mimic the adobe generative fill thing?

[nova remnant]

yes, stuck x3

okay I'll try r-esrgan

[frozen olive]

yes

[supple raptor]

pretty much, yeah, the krita extension will give you a txt2img, img2img, and inpaint toolbar. each of them will generate within the area you select with the area selection tool, with the img2img and inpainting ones using underlying layers as the input image. For inpainting you also add a paint layer which you draw the inpainting mask on.

[nova remnant]

And why are they slower

[supple raptor]

R-ESRGAN stands for Real-ESRGAN which is different from ESRGAN. They're different types of GAN upscalers.
Enhanced
Super-
Resolution
Generative
Adversarial
Networks
I'm not sure what the technical differences are, besides that R-ESRGAN is slower at the same task.

[final vault]

Which model is being used in this server?

I mean stable dreamer bot is using which model?

[supple raptor]

sdxl1.0 AFAIK

[steady grail]

the first is just deformed, the second is missing a finger

if not for the missing finger, the second is better

[white zenith]

well it has the right number of fingers...

[steady grail]

near the middle finger, she have a cut off finger and the first 2 finger are just undetailed

is easy to manually edit it

edit it by hand, use a controlnet and remake the image

[final vault]

right one ?

lmao

My mind goes brrrrrr when i see nodes

[supple raptor]

left is after inpainting?

yeah, that's about what I would expect, hand inpainting without additional control is largely a waste of time. it can be used to make a pretty good hand more detailed, but it will almost never make a bad hand good.

If you add controlnet with a hand depthmap to the inpainting process, you can make it a lot better, otherwise it's time to learn to draw so you can fix them yourself.

[supple raptor]

Now the problem with that method is going to be trying to apply any sort of automation to it since the body position will also effect the proper orientation for the depthmap.

It should be possible to build something that would effectively preprocess your image with openpose, match that skeleton the closest match from a pre-determined set of good skeletons and then use that to automatically align either the openpose hand skeleton or depthmap or both. I'm no programmer though so I have no idea how to actually make that or what the compute costs would look like.

[native halo]

controlnet openpose then gen a lot of imgs

it also depends on model some models suck at hands

[keen hollow]

evenin'

what's new?

[stable crane]

pray

[native halo]

cat is pregante

[keen hollow]

erm... ok

[stable crane]

you must be a proud father

[native halo]

https://tenor.com/view/dawg-skull-emoji-gif-24661581

[stable crane]

so cute

[nova remnant]

started reading hs dxd so there's that... although now I want to make my own loras lol
either too many Rias ones that aren't very flexible and only give the style of the anime,
and the clamp style only works better on anylora and 3.0. wanted to make a 2000s style anyways like the LN so idk....so many choices...

although

clamp proportions moment

[native halo]

have u tried this model? https://civitai.com/models/137781/era-esthetic-retro-anime

[supple raptor]

who's ready for math class?

[native halo]

[nova remnant]

omg thank you

After looking, I think it's nice
but I think I'll make a Lora anyways lol. That model seems more cel shaded and like actual anime, I'm looking for LN / virtual novel / manga style

I mean, I might be able to get this with clamp

but I'll have to do more...testing. lol

[nova remnant]

I do not trust her

[native halo]

maybe something like this? or still too colorful?

[nova remnant]

Yeah! how did you get that one

I've tried it on anylora but now it's too baked 😭

[native halo]

its the clamp lora https://civitai.com/models/76069/clamp-style-lora

[nova remnant]

yeah it's what I've been trying

I mean what strength

it could just be the character Lora I'm using

[native halo]

0.8 try with another model

meinamix or camelia

[nova remnant]

which in that case just means I have to make my own lol, but easier than a style one

ah okay

[supple raptor]

@nova remnant this is using epoch 7 at 1.0 strength on Kakigori_v3 with 0 effort into prompting for any retro style. It would work better on models with a more neutral style and with some prompting to help. https://huggingface.co/segole2003/Lora/tree/main/clamp128dim

segole2003 has some really good loras and usually they work at full strength or close to it.

[nova remnant]

Oh awesome

[supple raptor]

too much denoising most likely.

usually around 0.2 for minor corrections, 0.4 for larger corrections, and 0.6 for when you want drastic changes (expect to often need to inpaint along the seams again at lower denoising at anything over 0.5).

Also, @nova remnant if you like inpainting, consider trying this because inpainting in Comfy is about the farthest thing from being comfy. https://github.com/JasonS09/comfy_sd_krita_plugin

Better, but still not as good as inpainting with the Krita Plugin. In Auto you just plop your image onto the webpage and draw the mask on it.

The main benefits of using Krita instead is you have far better control of the inpaint mask because you use the Krita brushes to draw it and everything is done in layers so you can inpaint the face a few times find one that has the mouth you want and another that has the eyes that you want and easily blend them together, and if you can't get the blend quite right you can just inpaint over that again at lower denoise to fix the blending.

Soft edge brushes like the airbrush in Krita are awesome for inpaint masking because it allows you to control the blending better.

[native halo]

who is this beautiful girl

https://tenor.com/view/lucky-star-anime-konata-izumi-happy-excited-gif-14835420

[supple raptor]

Oh that's even more cursed than I thought it was. It's because you have to erase the section of the image that you're inpainting which means that you have to use denoise of 1 because there's nothing there for it to base the image on so it needs the full number of timesteps to arrive at an image. In WebUI, this is the same as choosing fill for the masked content which is almost always worse. I'm not even sure how you you go about doing inpainting the right way in Comfy. You might be able to get some acceptable results by either turning up your grow_mask_by to around 24-30 and also adding a FeatherMask node, or by using a soft edge erasing in your image editing software to manually create a feathered mask.

[nova remnant]

idk it might just be wrong settings? what do the mask settings say

like uhh iirc

"use latent noise"

around that area

I would just mess with it

[nova remnant]

lmao dont mind jacked hand but the rest is nice

ah ai

i really need to get the betterhands lora

[native halo]

idk what happened here

[nova remnant]

finally figured it out lol

[bitter latch]

hmm ok did not expect loki bat girl

[nova remnant]

HEY SHE JUST LOOKS LIKE THAT!

[bitter latch]

hmm she needs a better weapon

[nova remnant]

She looks big and nice I approve

[bitter latch]

hmm?

[bitter latch]

nice but not sure whats going on. not what I prompted.

oops someone made Miku, Thor Daughter Mad

[nova remnant]

anime is much more ecchi but that's all I'll give you

or whatever, bonk

[nova remnant]

hammer

squeaky hammer though

[bitter latch]

time?

[nova remnant]

lol

[bitter latch]

ok

[nova remnant]

What do you mean

[bitter latch]

shhh

hmm squeeky hammer is not easy

[nova remnant]

LOL nah I didn't think so

but the hammer looks nice

[bitter latch]

"i'm ready!" umm ok...

[nova remnant]

too many ...idk even
handles??

[bitter latch]

yeah...

[nova remnant]

Rock hilt

I mean, that WOULD hurt

[bitter latch]

no I think thtas the bonking end

[native halo]

she can hurt me anytime

[nova remnant]

Slaps hilt on a rock
ready

[nova remnant]

We need more skin tight suits

[native halo]

(pink bodysuit,latex,skin tight:1.5)

[bitter latch]

makes one with a ban hammer

tho got a feeling that isn't gonna work either..

blinks

waits

not sure what happened here...

makes a new merge and....

oh yeah

hmm ok it's chaotic

oh if I took out glitch art.. doh!

there we go!!!

[nova remnant]

@supple raptor so the hugging face lora profile you linked earlier, have you used the other loras? could you explain some to me? bc Idk any of the names

[supple raptor]

The majority of them are just named after the artist they're trained on.

[nova remnant]

ah okay ty

[light moat]

https://media.discordapp.net/attachments/1163310633970839632/1166605304901935154/upscaled_image.png?ex=654b18db&is=6538a3db&hm=1e80b5b9b4f5e122957eb4ddfaba340ed949e8eb402e743e7d853eb4fddd0516&=&width=531&height=943

[light moat]

wait, why does Discord not embed its own image links?

[final vault]

kek

[white zenith]

completely failed for what I was going for but interesting style

[dawn jasper]

This is quite the well-endowed pic.

[oak spoke]

nobody

just a rando

[nova remnant]

[gusty palm]

i wanna try it hmmm

i saw somebody use the generative fill in some adobe thing and it was like wow they turned a 1:1 image into like a 16:9 by just expanding the sides

afaik for the upscaler question above. GAN models can effect the shape of faces and things like that whereas something like 4xanimesharp dont do that

i remeber that cuz the first time i heard about GAN it was in a face fixer type web app

i randomly ran into this video titled 'the trial of tim hidecker"

i highly recommend it it im not a fan of tim and eric but its funny but then it sent me down a rabbit hole of a show called live on cinema

or just on cinema

so funny

[cyan burrow]

If you use a good out painter that can analyze the existing image you can do the same in SD. Perhaps with more control and higher resolution.

Firefly is great though, I love it and can't wait for the full res version

[outer tartan]

Maybe I should ask in here. I'm using the membership Stable Diffusion on their main website trying to get some anime pictures. I've noticed the quality dramatically dropped from a day ago. I now get flat 2d images while earlier ones were almost 3d like with better lighting. Anyone know what's up? Or do I have to put in a special keyword to get that quality back?

[cyan burrow]

depends which image you're using and style. There are options toggles for both

I'm not home atm so afraid I can't help with details

[outer tartan]

I'm just generating from the website.

[cyan burrow]

yeah but it has those options

[outer tartan]

ok, I might not understand how to exactly 'conjure' it up. I could literally copy and paste my prompts from one day ago, get one style, do it again today, and get a totally different style.