🍥｜anime

[lyric zealot]

might just be selecting the None thingy

but not sure tbh

maybe

no idea tbh

[nova remnant]

thats the pattern i see on my end

cause initially i went and tried to make custom settings and got errors

[lyric zealot]

then maybe

[nova remnant]

but after exporting default i can now make custom trt using the advanced settings and all the sizes

[lyric zealot]

well, i think the guide said to do default first

so you might be right

[nova remnant]

yeah it did

where do you delete the trt models you may not want to keep?

so i can save some space just for the ones i want to use

[lyric zealot]

in models > Unet-onnx and Unet-trt

[nova remnant]

sweet thanks

Torq

Now time to build 60 gigabytes of engines

[supple raptor]

You mean just for models right?.... right?

[nova remnant]

Nubby

[supple raptor]

smol vram neko

[nova remnant]

no way, im not doing that 🙂 i have to pick specific models i wanna use

[fiery plank]

/design a vector-style logo for IA2

[nova remnant]

Beep bop

[lyric zealot]

imma wait for nubby to

[supple raptor]

bots are in #1100170312106127410 - #1101178553900478464

[nova remnant]

@lyric zealot where's the image, bot?

[supple raptor]

https://tenor.com/view/peter-griffin-nails-nail-computer-keyboard-gif-7886259

[nova remnant]

[nova remnant]

Green girl

[lyric zealot]

I'm trying new model

[nova remnant]

Which one

[lyric zealot]

this is an old model by ctd

[nova remnant]

awsome, highres fix worked with the new trt model i just exported

[lyric zealot]

@nova remnant the model you requested

[nova remnant]

Keep using it

:D

[nova remnant]

This model is very nice

It puts lots of details

[lyric zealot]

i don't know why

i can't deactivate this

it will break

like it will literally stop generating

[nova remnant]

Stuck in the matrix using trt forever

@nova remnant how much faster is it for you

[nova remnant]

Mike

[nova remnant]

pretty fast, takes around 20 sec to generate 2 images

[nova remnant]

And what was it before

[nova remnant]

im exporting two trt models for now....

[lyric zealot]

this model works well with Euler a

[primal wharf]

wasup

[nova remnant]

problly around 40ish sec before

[nova remnant]

Try ddim

[lyric zealot]

this model is a little weird tho

like the starting

the noise is unlike other model

no idea why

unless something broke in my SD

[nova remnant]

How many steps do you have

[lyric zealot]

currently?

150 steps

[nova remnant]

[lyric zealot]

lmao i just realised you can't use ddim with dynamic thresholding

no idea why

but it gaves error

then still generate

[nova remnant]

[nova remnant]

I hate you goodbye and I'll see you tomorrow

[lyric zealot]

[nova remnant]

The meme

[lyric zealot]

im liking the output with Euler a tho

[lyric zealot]

i have no knowledge of that meme

[nova remnant]

Time for another 30 samplers test

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

What are the ancestral samplers

I think it's good for that model

[lyric zealot]

that model might be weird

like i can feel something is weird

but no idea what it is

[nova remnant]

It's not

[lyric zealot]

the generation at the start is slow

and the noise at the start, is really weird compared to others

[nova remnant]

Did you prune it

[lyric zealot]

nope

[nova remnant]

-5 gb

[lyric zealot]

that model is 2gb

[nova remnant]

Hmmmmm

Might be thinking of another one

[nova remnant]

Smol

Smoool

Do a room full of books and some magical stuff

@lyric zealot the shirt pants on that girl

Is why my emote

[lyric zealot]

https://tenor.com/view/frieren-beyond-journeys-end-sousei-no-frieren-soso-no-frieren-anime-gif-15331639985042747369

[nova remnant]

Always bonk me

[lyric zealot]

oh wait

i didn't even see that lmao

[nova remnant]

guys .. just for the thought, SDXL image quality are neither fantasy pleasing nor realistic

i say that with comparison to sd 1.5 community models

[lyric zealot]

[nova remnant]

To me SDXL is just as good with realistic

[lyric zealot]

i mean, there's probably good SDXL model that excels on those genre

[nova remnant]

Anime/drawing colors look stiff on it

Idk how to describe

[lyric zealot]

okay time to delete my engines

[nova remnant]

And build more

[nova remnant]

???

you are deleting your trt engines?

[lyric zealot]

delete some of them

[nova remnant]

oh some..

[lyric zealot]

true

[nova remnant]

@nova remnant where cute girls

[nova remnant]

[nova remnant]

you mean smol sus? i think she is cute

and intelligent

Smol only bonks me

[lyric zealot]

[nova remnant]

Now add magic

[nova remnant]

girls look so cute when they read and study

[lyric zealot]

fact

[nova remnant]

"energy flowing, magic circle, aura, bright" @lyric zealot

In the library prompt

o:

[nova remnant]

Sus images on those books

[lyric zealot]

i don't see it

and this model doens't like following prompt

or need more weight

[nova remnant]

What did it not follow

Do one of those weird looking natural prompts kek

[lyric zealot]

lmao

i don't do natural prompt

[nova remnant]

"as she reads the books energy flows in her body making the room light up, she has an aura of power and is taken a shock by the energy"

Lmao

It's so weird

Telling an AI prompts like that

[lyric zealot]

forgive the fingers

well, that's kinda how you would prompt with dalle3 right?

[nova remnant]

Very calm library, would read books in it

[nova remnant]

Dalle is smort

[lyric zealot]

this model is weird

[nova remnant]

You can do straight up just tags

[lyric zealot]

or my sd is broken

now it's genning at 2it/sec

and after interrupting and starting again

[nova remnant]

Is the prompt size weird

[lyric zealot]

now back to 12it/sec

literally didn't change prompt

[nova remnant]

I mean

Is it staying near the amount you set in the engine

[lyric zealot]

i make it dynamic

[nova remnant]

You guys said prompt size matters

[lyric zealot]

from 75 token to 300

so should be fine

well

https://tenor.com/view/lion-king-toss-bye-gif-4963915

throws model

[nova remnant]

More engines time

[lyric zealot]

https://tenor.com/view/suisei-hololive-train-suisei-train-hoshimachi-suisei-gif-21843410

[nova remnant]

https://tenor.com/view/cat-table-gif-10661286

[nova remnant]

smol sus so before you can export a model profile with advanced setttings you'd first have to export a default onnx file for each

[lyric zealot]

yup, it will create one ONNX file for each model

but the trt is per engine

[nova remnant]

i was trying to switch model and go directly into exporting a profile with advanced setting and it woudl throw error

[nova remnant]

Woof

Dangerous wolf

[lyric zealot]

then that's probably why

[nova remnant]

hmmm, im thinking about the pros and cons of trt

considering im using a 12gb card and with small 500gb ssd, and trt mods taking up a lot of space which kinda overlaps with how i can use the same mods normally i think it sucks to have to delete collection of good mods in my ssd to make room for trt

Buy a 1tb ssd

[primal wharf]

Yuki when she becomes royal guard.

VIP protection is her secondary duty as maid.

[nova remnant]

[nova remnant]

i think i will but just not yet, apart from that this trt tech sucks

Trt like nubby said it's really nice for bots

That have one setting

nvidia had to keep consumerism in their objective

[lyric zealot]

well, you probably only wanna have engines for those commonly used models

[nova remnant]

i know my ocd gets the best of me often

i kinda feel like for all the collection of models i like i want to keep them all organized and ready for trt

[lyric zealot]

i totally get how you feel

[nova remnant]

lol

[nova remnant]

(zoom in, in my DMs:1.6)

so far these are my favorite in collection

[nova remnant]

Kakigori too cute it's out of place

[stable crane]

there are too many good styles

[nova remnant]

lol... naw its fine whre it is

[stable crane]

you also need @zinc flax's ultra cute mixes

[lyric zealot]

wait Sleepti released it?

[nova remnant]

which one are those mixes?

[zinc flax]

Which one lol

[stable crane]

he released a few a while back

[lyric zealot]

Your latest?

[nova remnant]

share the link lol

[lyric zealot]

wait, did you release any? I didn't even know

[nova remnant]

Smol

You can't afford it

No more engines

[lyric zealot]

still have spaces for like 30 engines

[zinc flax]

https://civitai.com/models/132842/sweetparfait

[lyric zealot]

https://tenor.com/view/nascar-racing-start-your-engines-gif-24794262

[nova remnant]

i have 190gb free space left

[nova remnant]

too cute

[lyric zealot]

https://tenor.com/view/error-wait-computer-download-cat-gif-10480536

[stable crane]

its very good

I want to go back to generating cute now

I wish I could have every gen I do just output on like, 5 models simultaneously

[lyric zealot]

you just need to have 5 gpu for that, no big deal

[zinc flax]

There are also some unreleased ones on pixai

"Unreleased" lol
I just disabled the download

[nova remnant]

nice big mecha

very round

[lyric zealot]

didn't you said that you are gonan release that soon?

[zinc flax]

Probably just gonna upload it to hf

soon™️

If I remember

[lyric zealot]

[nova remnant]

(medium shot), 1girl, cute, pink hair, lofi, messy room, high tech gadgets, headphones

used adetailer LoRA too

[stable crane]

with (cute:1.3)

[stable crane]

gn

[wooden coral]

Has anyone here tried this? https://github.com/avatechai/avatar-graph-comfyui/tree/main

[nova remnant]

bored

[wind girder]

I'll investigate, this seems cooooool

[wooden coral]

Let me know how it goes, I wanna try it too

[lyric zealot]

im cooking

[nova remnant]

https://tenor.com/view/skeleton-waiting-hmmm-think-gif-16408853

[lyric zealot]

i will get there for sure

The funny thing is, the generation isn’t even that good

[nova remnant]

Are you sure about that

for 10000 steps you might win de oscar

Clueless

[lyric zealot]

[nova remnant]

[nova remnant]

I did it

I deleted 50,000 generations

[zinc flax]

Oo what model did you use for this

Love the eyes

[lyric zealot]

I’ll send the metadata later

[zinc flax]

Awesome

[lyric zealot]

here

the weird double "," is because i used style for the vaporwave stuff

[nova remnant]

[nova remnant]

Can u make nsfw images in the system?

[dawn jasper]

Like what, specifically?

[nova remnant]

Idk

My friend wants to know

But they ain’t saying what :/

[wooden coral]

What system are you talking about

[nova remnant]

Idk

I think for the entire platform in itself

[wooden coral]

On this server with the bot, hard no. What you do on your private machine, we wont judge

[light moat]

Yeah, your friend 😉

[wooden coral]

Also that lol

At least have the balls to tell us the truth

[nova remnant]

Bro

Nvm

[nova remnant]

No honestly

[bitter latch]

if you are on a local install you can make whatever you wish. that being said... shrugs. not sure what else to say

[nova remnant]

I ain’t using the app

Cuz I like drawing irl

[wooden coral]

Good for you

[crystal isle]

i'm using the stability api and when i try to create an anime picture they are always totally blurry, anyone has an idea?

[bitter latch]

hmm they added tensorRT to the ext in auto 1111 but I get this when trying to do the generate engine thing

[ornate echo]

That's not an issue and if you look at the terminal, it should still be progressing.

[bitter latch]

yeaqh bunch of errors. wow

[ornate echo]

The stuff in the console matters more.

What's it say?

[bitter latch]

but it seemed to have done what it should. so lets restart

says it expected one machine found two gpu , cpu. checked settings it's set to GPU only so shrugs

hmm prob cause I have onboard graphics as well as the dedicated GPU. hmmms

ah ha! removed the medvram in bat file

and it worked

now let's see how it works

hmm taking a long time to load it tho

[zinc flax]

Nice

Clean model

[bitter latch]

hmm still doing it's switch to the SD unet. wow hope it only does it once, otherwise this is gonna be so not worth it

leaving it on automatic

with ... 34 secs. tho felt like less

without 17 secs

diff model 9s

hmmm

not sure if this thing is making any difference

[wooden coral]

https://tenor.com/view/cats-catitude-cattitude-cat-attitude-ew-gif-22504376

[bitter latch]

yeah I think it sped it up without the medvram

ok it doesn't like nonsame x nonsame

only does squares? what?

or hmm anything above 768

odd

meh not worth it

takes 40 or more secs to load the tensorRT sd net and then you have to make one for each model. meh.. no thanks

[nova remnant]

lol ......

[nova remnant]

hey tora .. i deleted the tensor rt extension

my render speed isn't bad to begin with but i can't settle with all that garbled strategy to use trt models

[sharp swan]

first time doing something in 3594808934598345894tr89u456wuytr8u978j7u9fhn8g99 years

[bitter latch]

yeah I deleted as well. no really seeing any diff

[nova remnant]

speed wasn't a big point for me to lose space in ssd

and with every model i'd have to waste double the space using trt models, that's just plain stupid

[bitter latch]

yep

[sour shuttle]

guys if anyone can help me it would be much appreciate

i can stream you my screen , i need to know why image generating are so ugly , i dont have your results !

[supple raptor]

if you're using WebUi then just drag some of your images that you are having a hard time with from your txt2img-images folder into here, then we can drag them to our own WebUI and see your prompt and all of the parameters.

[nova remnant]

hi nubby

did you bother to run tensor rt?

there was a guide to fix the common error many were having and i got it to work on pc too

but the thing about tensor rt is just whacky by wasting huge space

and even if space wasnt a factor it doesn't make sense to run a model with twice as much resources

so i removed tensor rt altogether

[supple raptor]

I tried a few times yesterday, and got a different error than we were having before. I gave up for now. I'll probably try again in a week or two after they've hopefully figured out the problems with installing it.

[nova remnant]

ok, i felt like wanting to try it once too just for curiosity but after i was able to actually run it and see what it does i kinda lost interest to keep using it when my speed isn't terrible w/o it

there is a sweet feature update for chrome browser

[supple raptor]

Yeah, my speed is fine without it as well. If I had a slower card I'd probably be more interested in trying to make it work. It's less than 30 seconds for a 1080p image including an adetailer pass with my GPU.

[dawn jasper]

What a lovely witch lady. 😛

[nova remnant]

hey, yeah she is a cutie

dark sushi model generates a lot of anatomical deformity

[autumn violet]

I might have to try it at some point. It is harder to keep interest when a good image takes so long for me.

[supple raptor]

yeah, that's completely fair.

[bitter latch]

We were both young when I first saw you
I close my eyes and the flashback starts
I'm standing there on a balcony in summer air

[nova remnant]

my interest level gets drained because of spaghetti fingers

[autumn violet]

Oh no

[dawn jasper]

Who is this beautiful woman?

[nova remnant]

saylor twift

[nova remnant]

beautiful young witch in attractive and stylish makeup and black dress, witch hat, fair skin, light brown hair, green eyes, slender tall body, bare legs, before a large wooden table with candles, smoke trails, potions, herbs, gems, large window in the background

[gritty spruce]

i hate the navel on some generations, it's too much to have it recessed like that voer clothing

happens a lot on anime gens

we have to inpaint etc

[nova remnant]

the dress clings so tight

kinda like pasting over skinb

but honestly im not too fussed about it

[dusk seal]

She's ready to command capital punishment

[supple raptor]

any mixes that contain any version of AOM are especially bad about that which is a lot of the anime mixes.

[dusk seal]

shinji help me

[nova remnant]

is that sdxl by any chance?

[dusk seal]

D..dall e

🤕

[nova remnant]

ahh ok, similar to sdxl anime

[dusk seal]

Yeah tbh the gens I saw from sdxl anime are... Not bad

[nova remnant]

i wanted to play with sdxl MORE but it runs slow on my system

[dusk seal]

F

[nova remnant]

i have 16gb ram which makes it hard to run sdxl smoothly but gpu is fair enough with 12gb

[dusk seal]

Breh

I'm 8gb vramlet

If it's too slow for you

[nova remnant]

not too slow 🙂

8gb meets the requirement

and if you are running tensor rt with nvidia cards you could speed it up even more

[dusk seal]

I was told the best method for Ai art is brute force

Generate 11 pick the best 2 and fix them

[nova remnant]

oh wait... not sure if rtx has 8gb variant

[supple raptor]

yeah, tensorrt works on 8gb

[nova remnant]

nice

[supple raptor]

I saw someone say it will work on a 1660

[nova remnant]

something just ran through my mind... i could use tensor rt for one thing..... sdxl

[dusk seal]

Now she's staring with extra menace

[dusk seal]

How do I increase the weight with dall e

Want more menace

🧐

[dusk seal]

Is that realism in my anime channel

[nova remnant]

im going for it... gonna install tensor rt again but just for SDXL only

also found out there's new update on adetailer just now

im gonna grind the default base and refiner sdxl

not gonna bother with community models

[dusk seal]

Fingers botched

Waaaaaaa

[nova remnant]

lol dalle has to fix that too

[dusk seal]

It doesn't help that they must nerf it for censorship

[gritty spruce]

hahah trying the voluptous Asuka eating burgers there too? 😜

[dusk seal]

You bet

[gritty spruce]

[dusk seal]

And I get racially ambiguous asuka for it

Life's cruel

[gritty spruce]

go local

the thing is that on public servers and services it's not allowed, dalle3 is pretty bad with censorship, also MJ

[dawn jasper]

Humanoid Angel ladies.

[nova remnant]

if i wanted to render images with 800x1024 resolution on SDXL is that gonna come out distorted?

[dusk seal]

I understand, public and corporate reasons

[nova remnant]

since sdxl recommends 1024 as base standards .. i wonder about setting the resolution differently

[gritty spruce]

if you maintain the total pixel count, any res should work fine

1024 x 1024 = 1.048.576

[nova remnant]

how do you calculate the pixel count?

oh

interesting

[gritty spruce]

so if you go 1920 in one side, have to rest to the other one

[nova remnant]

yeah im going to try a lower gap

[gritty spruce]

1024 x 1024 is the maximum, you can do lower anyhow u like

those ones are all safe

[nova remnant]

nice 🙂

[gritty spruce]

another important thing is that all those numbers are divisible by 64

which make things easier

[supple raptor]

this one actually looks pretty good if you want better anime style than the base xl model. https://civitai.com/models/162577?modelVersionId=183131

[gritty spruce]

if you do a random 1032 x 734 it will either fit them to the nearest one or do it but will take longer

[sonic portal]

how do make?

[nova remnant]

she learned all the tricks from nubby here

[sonic portal]

im a fool with typing in the cursed box

[nova remnant]

https://civitai.com/models/100505/kakigori

[sonic portal]

dont u have to write like 2 paragraphs of prompts to make good looking images?

[nova remnant]

not at all

[gritty spruce]

nope, you just need to use the correct tokens and be efficient

specially the newer models with a few tokens they do great outputs

[nova remnant]

this was my prompt using that model i posted for that image you saw

beautiful young witch in attractive and stylish makeup and black dress, witch hat, fair skin, light brown hair, green eyes, slender tall body, bare legs, before a large wooden table with candles, smoke trails, potions, herbs, gems, large window in the background

[sonic portal]

im very unknowledgeable, on alot of stuff when it comes to ai art prompts and checkpoints and stuff

i heard u need like a vae and lora's too but idek what those do or are

[gritty spruce]

we're all learning

[nova remnant]

LoRAs are not a must but using the right ones can improve quality

there are also LoRA that can alter images

like character attributes

or scenic look

[sonic portal]

says 2 hours

wifi slow

[gritty spruce]

your internet connection is, where are you located?

[sonic portal]

not exactly close to society

took a whole night time to get sdxl

[nova remnant]

300kb/s is tooooo slow

[sonic portal]

u learn to live with it

[gritty spruce]

that could be anywhere on earth, there are more rural less populated areas than cities 😜

[sonic portal]

well i can look outside and see horse fields sooo

[gritty spruce]

anyhow i feel you, slow internet is the worse

[gritty spruce]

starlink is working on improving that

[sonic portal]

when it do download i can make images like this?

[sonic portal]

very pricey

[nova remnant]

yep you should be able too

[sonic portal]

do i need a vae?

[nova remnant]

yeah use this vae ....

https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless09.ckpt

[gritty spruce]

using the correct vae is important, you get better definition and colors

[nova remnant]

thats for all anime related models

[sonic portal]

ALL?

[nova remnant]

ALL

[nova remnant]

the muffin man

[gritty spruce]

yeah that one works well with most anime

[nova remnant]

lol ALL anime yes

[sonic portal]

i feel like things get much more complicated when u need to add negative prompts

[nova remnant]

nawww

use simple negative prompts and not over do it

[supple raptor]

that particular one is from an experimental version of Kakigori that I haven't uploaded yet. both Kakigori versions have the vae that torqx linked baked into them btw

[sonic portal]

is kakigori like the best looking anime model thingy?

or is it just everyone likes using it

[supple raptor]

it's just the name I gave my model

[sour shuttle]

do you know the simple way to transform myself in a manga anime ( one piece ) ?

[sonic portal]

THE ONE PIECE

[nova remnant]

here you can use this as generic negs for your images .... save that
bad anatomy, bad hands, (fused fingers, interlocked fingers:1.2), (low quality, worst quality:1.2), (watermark:1.1)

[supple raptor]

kakigori is like a japanese version of shaved ice.

this is the only negative I use on my models any more (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)

[gritty spruce]

simple like a one click solution no

[sonic portal]

nubby how do u know what to write, i c for some of your pics its like a paragraph of prompts

[nova remnant]

and if you want to extnd that you could use something like ...

bad anatomy, bad proportions, malformed, bad hands, (fused fingers, interlocked fingers:1.2), asymmetric, conjoined, distorted, cloned, (low quality, worst quality:1.2), out of focus, low contrast, oversaturated, monochrome, grainy, pixelated, border, (watermark:1.1)

[sonic portal]

like how do u figure out what to write?

[nova remnant]

you just imagine

use your ideas

[gritty spruce]

like any other language including music, you start with simple stuff, try and error and build up your skill

[supple raptor]

I've got over a year of practice on prompting anime models so it just kind of comes naturally to me at this point. This extension can help a ton. It autocompletes words into the closest danbooru tags which is what our models are trained on. https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

[sonic portal]

lots of info to take in

[sour shuttle]

@gritty sprucedid you install a lot of safetensor ?

@gritty spruceis it a good solution to get good results ?

[sonic portal]

this link will help me not suck?

[bitter latch]

hmm reminds me I need to make me a resource page on my site

[gritty spruce]

that is just a format, newer one for storing tensors

[supple raptor]

it'll help you figure out the right words to use instead of just prompting with natural language.

[gritty spruce]

newer models use it, it's more efficient

[sonic portal]

oh and another thing thats been irking me, whats with the brackets?

and the numbers with the brackets

[gritty spruce]

weights

[sonic portal]

ohnoooooooo

[gritty spruce]

the more weight something has the better probability it shows

a girl in a house is very different that (a girl) in a house, makes the weight of girl higher

you can also do it numerically

a girl (blue hair:1.2)

[nova remnant]

hello

[supple raptor]

(word) = (word:1.1) = more emphasis on "word"

you might also see something like:
[word1:word2:10] this will replace "word1" with "word2" after 10 steps.
[word::10] would remove "word" at 10 steps
[word:10] would add "word" at 10 steps

[gritty spruce]

that's more advanced don't scare them!

[sonic portal]

I feel like im in school again looking at like equations

[nova remnant]

@sonic portal blazingdrive model you type 2 words and the images come out amazing lmao

[sonic portal]

2 words?

[nova remnant]

In the prompt

[gritty spruce]

that's why I told you, we're all learning, a couple years ago none of this existed outside research centers and universities

[nova remnant]

Some models work well with almost nothing in prompt

Gives you ideas :p

This was like "girl, sitting, white hair"

[supple raptor]

Here's a full link to the cheatsheet for emphasis: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis

And here's the one for prompt editing (I would ignore this until you feel pretty comfortable with prompting): https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing

[sour shuttle]

anyone can post a photo of him turned into a manga character ?

[gritty spruce]

also mind that we're talking about stable diffusion models

[sonic portal]

Nubby is a godsend

[gritty spruce]

prompting for Midjourney or Dalle is different

[sonic portal]

Is both them better?

[gritty spruce]

at some things at others no

each has it's flaws

[nova remnant]

Cat typing gif man gives us big text of knowledge

[magic wave]

I'd say Dall-E, Midjourney and Nijijourney are better for like... casual users, or like minimal effort

[gritty spruce]

If you really want control of what you are doing, SD based workflows are the best

[magic wave]

Yeee

[sour shuttle]

midjourney is easier to use

[sonic portal]

Whats the difference between stable and easy diffusion?

[sour shuttle]

we dont need to install of thsi github shit

[gritty spruce]

yeah but most niji stuff looks the same

even the same color palettes

[magic wave]

True too

With SD we got the power of LoRAs

[nova remnant]

I hate people that say stable diffusion images look the same

Like bruh

[gritty spruce]

niji

[magic wave]

I mean, some do, but they just gotta git gud at generating

[supple raptor]

[nova remnant]

It's like a table full of pencils and you chose the one everyone is using

[magic wave]

amdroid

[nova remnant]

X3

[magic wave]

Yeee

[sour shuttle]

if i use a picture of me , it will generate a totally different manga character

[nova remnant]

Is cappuccino maker included

[magic wave]

You gotta use either img2img, or controlnet, or both at once

[nova remnant]

I'm in if so

[magic wave]

Built right in

[nova remnant]

You should try controlnet canny model

[sour shuttle]

@magic wavethnaks , where is controlnet ? i saw a video of somzeone using it but i cant find it on my web-ui

[gritty spruce]

you have to install it first

[magic wave]

Yee, you'll find it in the extensions tab

[gritty spruce]

it's an extension and some models

[supple raptor]

Easy Diffusion is the name of a WebUI that was intended to be a simpler interface than the WebUI by Automatic1111 that most of us use. The problem is that almost everyone uses Automatic1111's WebUI so when beginners run into problems and ask for help nobody can help them much if they're using Easy Diffusion.

[magic wave]

Then you can look on a site like CivitAI for Controlnet models and download the ones you want

[nova remnant]

Does it require less vram by chance

Asking for a friend

[nova remnant]

https://stable-diffusion-art.com/controlnet/
https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main

[sour shuttle]

thanks

[dawn jasper]

Is that Rin Tohsaka?

[magic wave]

Seems like Asuka from Evangelion

[nova remnant]

It's 1girl

[supple raptor]

I don't think so. Fooocus might use less though I think. I have no idea what level, if any, of support it has for AMD though. https://github.com/lllyasviel/Fooocus

[gritty spruce]

1girl cyberpunk pink hair futuristic neck choker 😜

[magic wave]

Nun

[gritty spruce]

omg, it started! gotta bring the mod with the electric ban hammer to town

[magic wave]

Then, I present, a this

[gritty spruce]

is the kohakuXL model yours @supple raptor ?

[golden stratus]

can i just ask what SDXL actually is since this is the only active chat, keep seeing mentions of it and have no clue what it is

[gritty spruce]

it's a model, stable difusion XL

[golden stratus]

oh thats it?

[gritty spruce]

trained on 1024x1024 resolution

[golden stratus]

damn

rightttt

alr thanks

[supple raptor]

no, that belongs to Kohaku-Blueleaf aka: kblueleaf (the person behind LyCORIS ). Kakigori is my model.

[nova remnant]

lmao

They are onto us smile

I am

On you*

I'm innocent

[gritty spruce]

oh the report-bot has been on fire with this channel, I think it isn't anyone in particular but all of you. I think even something I posted got reported

[sour shuttle]

i have one more stupid questiion

why they dont do a simple application ? not everyone can use this software

[gritty spruce]

try "foocus"

[sour shuttle]

its because its not a server generating the picture but our own gpu ?

[supple raptor]

@finite herald look what I found hiding in your wifi

[golden stratus]

How do i make shonen like characters, cuz every anime thing ive found just makes hentai, i wanna make smth or smn like ichigo kinda

like what models are good for it

[sour shuttle]

@gritty sprucethx$

[dusk seal]

who

who could possibly mistake asuka for someone else

[gritty spruce]

wifi miku or waifu miku?

[supple raptor]

it's a waifi router

[golden stratus]

anyone?

[dusk seal]

bald miku

[gritty spruce]

oh noeeee

just don't do her like monster teeth alien thingy, we got scared a couple days ago hahaha

[supple raptor]

this is @nova remnant's favorite girl

[dusk seal]

I did unspeakable things to her, I made her... NSFW

the horror

[golden stratus]

swear shes underage

[gritty spruce]

just dont do that here or horny jail 😂

[golden stratus]

reinstated

swear shes underage

[supple raptor]

she's been 16 for 16 years

[dusk seal]

not my problem cope

[golden stratus]

aye nah

[gritty spruce]

she's like 107 years old, but uses ancient techniques to stay at 16 and even going full chibi power

[nova remnant]

https://tenor.com/view/cat-lava-cat-minecraft-cat-gif-13440479435726322249

goodbye cruel world, I had to see IT again

[golden stratus]

lit

now someone please

i need one thing answered

and ill disappear forever

[supple raptor]

so if we don't answer then you'll stay?

quick, nobody answer

[nova remnant]

nice one more friend forever

[golden stratus]

ill stay if you want me to

your choice

okay ill leave if it isnt answered

tricked ya

good model to make shonen-style anime characters like ichigo/naruto/goku

i need it

i beg

someone has to know

[gritty spruce]

nubby is the one for that

[golden stratus]

dear nubby

please

:)

h e l p

[gritty spruce]

kakigori will do that right

[golden stratus]

is that a model or are you speaking japanese

[dusk seal]

loli diffusion very good for making shonen

[golden stratus]

nah you need to be locked up

[dusk seal]

[golden stratus]

AYE AND THE KAKIGORI ONE IS LOLI ASWELL

[supple raptor]

goku is pretty well known on pretty much all models. naruto is not as well known, but using uzimaki naruto should get you fairly close, for ichigo you're probably going to want to find a LoRA.

[nova remnant]

couldn't export a dynamic engine for SDXL, it was throwing errors, but got the static 1024x1024 successfully exported

[golden stratus]

But is there any good models trained specifically for those styles

ion wanna be using like

abyssorange style

for gojo

or sum shit

[supple raptor]

[golden stratus]

im gonna guess thats a no

maybe

😭

[sour shuttle]

this is embarassing ...

[supple raptor]

there probably are some. I don't know which ones off the top of my head though. I'll take a look at a few.

[golden stratus]

Thank you ive been looking for so long but will continue

[dusk seal]

Tragic

[golden stratus]

is that fucking luffy

[sour shuttle]

i just want to be drawn as an anime

and this is the result

[golden stratus]

i tried doing smth like that once

it takes a decade of fine tuning

but youll get a good result when you figure that shi out

[gritty spruce]

you can concatenate multiple controlnets, canny and depth are good for overall facial features

and then img2img too

[keen hollow]

evenin'

[gritty spruce]

if you want video of that maybe runway Gen-1 or something like warpFusion

[keen hollow]

how's everyone doin'?

[supple raptor]

have you tried AAM? https://civitai.com/models/84586/aam-anylora-anime-mix-anime-screencap-style-model

[golden stratus]

OOOH THAT LOOKS GOOD

I SHALL TRY IT

THANK YOOU

[supple raptor]

those two images are in the previews for the model and include loras, which you can mostly likely also find on CivitAI.

[keen hollow]

what's up Zeke?

[golden stratus]

If its anylora i suppose id have to find a lora for the style im going for right?

[keen hollow]

I had a computer named Zeke once.

[golden stratus]

nothing much js tryna make gojo or ichigo on SD

wbu?

[golden stratus]

damn nice lol

[keen hollow]

watchin' youtube and fartin' around

[golden stratus]

lovely lmao

[supple raptor]

the default style of the model is fairly similar to screencaps from average anime, but the model was intended to be especially compatible with loras.

[golden stratus]

Alright I'll have a play around, Thanks!

[dawn jasper]

Mathematically correct answer is Yes.

[nova remnant]

SDXL base and refiner

only thing on top of that besides prompts is adetailer

[supple raptor]

pretty good, you'd probably be better off using a 1.5 based anime model as the refiner though.

[nova remnant]

oh? i didnt think that was possible to use sd1.5 modes as refiner with SDXL

let me try it out right away

[supple raptor]

you probably have to use comfy to do it, in comfy you just add a decode with xl vae, then encode with 1.5 vae between the switch.

[golden stratus]

Im using two negative prompts with minor changes and they're giving me totally different results its really weird

[supple raptor]

do either of them include negative embeddings?

[golden stratus]

One has easyneg

the other one doesnt

ik embeds do alot

but its like the style changes completely

[supple raptor]

yeah, easyneg has a huge impact on the style and composition of the image.

I don't use it for that reason.

[golden stratus]

It becomes alot more simple and focus' around like a single colour

this is one i made without the easyneg tho and its pretty cool

still waiting on that other model to download cuz i forgot to press start on the download

[nova remnant]

yeah i see what you mean ...

[golden stratus]

thats LIT

[nova remnant]

with a1111 i didnt have the option to use two different vae

[golden stratus]

actually pretty cool imo

creepy style

@supple raptor What negative embeds or prompts would you recommend i use to yield the best results for anime based stuff?

[supple raptor]

this is all I use for almost all of the models I use. I quit using neg embeds for most of them. (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)

[golden stratus]

alr ill give it a try thanks

[nova remnant]

works on 1.5

smol was doing it

[supple raptor]

not if you're using xl for the the low res pass and sd1.5 for the hires pass. the latents aren't compatible.

[nova remnant]

but I think the time to switch is annoying >.>

You're thinking of something more advanced than me

[supple raptor]

poor girl lost half her shirt and 90% of her jeans

[nova remnant]

needs to lose more

[supple raptor]

[nova remnant]

me looking at those jeans

[golden stratus]

https://tenor.com/view/todo-aoi-todo-bruh-ping-pong-table-tennis-gif-11193322147763516631

[nova remnant]

Neg: easynegative, bad-hands-5, (low quality, worst quality:1.2), (watermark:1.1)

[golden stratus]

Pretty coooooooool

[nova remnant]

Neg (but not TI Embeddings):
easynegative, bad-hands-5, (low quality, worst quality:1.2), (watermark:1.1), bad anatomy, bad proportions, malformed, bad hands, (fused fingers, interlocked fingers:1.2), asymmetric, conjoined, distorted, cloned, (low quality, worst quality:1.2), out of focus, low contrast, oversaturated, monochrome, grainy, pixelated, border, (watermark:1.1)

[golden stratus]

ive merged a few things and made a neg prompt which seems to work good for most things

this is what im workin with rn

(worst quality:1.25), (low quality:1.25), (lowres:1.1), border, skimpy, grayscale, multiple_girls, (watermark:1.2), (monochrome:1.1), (greyscale), multiple views, comic, sketch, (blurry:1.05), (bhands-neg:0.9)

[nova remnant]

why skimpy in negative

also

[golden stratus]

just there from when i asked nubby for their negative prompt

ill remove it im sorry 😭

[nova remnant]

1.05 weight :D

[golden stratus]

on which?

[nova remnant]

on blurry

kinda funny

[golden stratus]

is that bad

i just found it somewhere on some civitt image

and thought hey

ill try it

[nova remnant]

its like so minimal compared to just having blurry

0.05 more

[golden stratus]

ion know bruh

[nova remnant]

i mostly prefer to use to avoid the extra hassle Neg: easynegative, bad-hands-5, (low quality, worst quality:1.2), (watermark:1.1)

I use negative_hand-neg, (KHFB, AuroraNegative:1.2), (Worst Quality, Low Quality:1.2), border, grayscale, (watermark, multiple views:1.2)

just make sure you have those TI files

can you share the links for them, id like to test them out rightaway

[nova remnant]

very sweet

[nova remnant]

https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings
https://civitai.com/models/56519/negativehand-negative-embedding

[nova remnant]

thanks!

[nova remnant]

thighs

the 2nd link you meant to share negative hand neg?

i think i have the link for negative hand neg...

[iron star]

she looks perfect for

[nova remnant]

https://huggingface.co/nolanaatama/embeddings/tree/main

[nova remnant]

mistakes have been made I think

https://civitai.com/models/56519/negativehand-negative-embedding @nova remnant

I'm confused I don't know anymore

I just type words and images come out

[untold glacier]

This one is going in the folder.

[nova remnant]

why booli me

[nova remnant]

with > AuroraNegative, KHFB, negative_hand-neg, (low quality, worst quality:1.2), (border, grayscale), (watermark:1.1)

something's off

im gonna keep both these quick template

AuroraNegative, KHFB, negative_hand-neg, (low quality, worst quality:1.2), (border, grayscale), (watermark:1.1)

easynegative, bad-hands-5, (low quality, worst quality:1.2), (border, grayscale), (watermark:1.1)

@nova remnant I liked having Auroranegative and khfb at 1.2

[untold glacier]

Auroranegative is 👌🏻 I use it for all anime models, and some RL models.

[nova remnant]

me too,i can see the big difference on the faces im generating

but not using weigh

try with 1.2

not anime, but these faces look different than usual from the ti i was using, its interesting

[nova remnant]

ok

[dawn jasper]

This is a powerful image.

[nova remnant]

@untold glacier why do you have a difficult name to type

anyway

I found the issue

I needed a vae

[untold glacier]

Because Discord forced everyone to change their username.

[untold glacier]

I literally almost suggested that.

[nova remnant]

its so hard to tell

[untold glacier]

Usually when things are bland like that with low saturation, it's either a low CFG or you need a VAE.

[nova remnant]

I also like images not as colorful sometimes

so I thought it was okay lmao

[untold glacier]

Yeah, some models are like that. Especially the earlier ones.

[nova remnant]

these take like 5 minutes to make

its so bad

[untold glacier]

Yikes

Certainly pops a lot more with the color now though.

[nova remnant]

its like 512x768 and 1.5 upscale

I love this model

i wish these embedding came with some explanation for their use case, casue i have no clue in what manner they are influencing images, atlho i like what i see

@nova remnant that would be pretty hard to do, considering ALL the variables :p

You can do a comparison yourself

without, 0.2 strength and +0.2 until 1.0

You'll see whats changing in your image

[nova remnant]

using the xyz plot?

i dont think i fully understand how to use that script

[primal wharf]

Skinny tree branch

[nova remnant]

Those rival your girls

x3

[primal wharf]

Skinny plant stems

[nova remnant]

sweet, let me try

[nova remnant]

And then for example in the box you put "AuroraNegative, (AuroraNegative:0.2), (AuroraNegative:0.4)..... "

it will search for AuroraNegative in your prompt

and replace with the rest

until it does everything

i dont see prompt sr in that section

[nova remnant]

It's a plot

Inside XYZ plot

so something like this?

or do i put the different weights in x y z seperately?

Only one prompt s/r

[nova remnant]

In that x values

It will do each of those prompts

Instead of the first one

ok

what about these

0.5 +0.5 might be too abrupt

For you to see the changes exactly xd

Also

The first one is already 1.0 strength

It will do that one too

[keen hollow]

evenin'

[nova remnant]

I don't know about any of that

Just leave default and it works

ok reducing weight but do i click swap xy axis?

No

Just generate the image

oh

good gosh, i have no idea about this script, thanks for explaining

think im doing it wrong

do i put these into negative prompt on top? (AuroraNegative:0.2), (AuroraNegative:0.4), (AuroraNegative:0.6), (AuroraNegative:0.8), (AuroraNegative:1)

[nova remnant]

what happened

this

im obviously putting those words in wrong places

@nova remnant

and I just left AuroraNegative in the negatives

like normal

[nova remnant]

:/

RuntimeError: Prompt S/R did not find (AuroraNegative) in prompt or negative prompt.
Time taken: 0.0 sec.

i put that word w/o weight in negative box tho

Show me your negatives and the script box

it has to be the exact same

as the first one

i have to include the bracket?

No

The script is looking for (AuroraNegative)

You have AuroraNegative

its not the same :p

[nova remnant]

ok so i put this in negative prompt box (AuroraNegative)

yeah

the script looks for the first prompt you put in it

and it will swap that one for the others after it

https://imgur.com/a/uhgfdYk @nova remnant

open them on a new tab each and swap between to see the difference haha

0.2 aurora up to 1.2

its almost done

is this the aurora negative alone ?

yes

with bunch of images in output folder

I think its better to compare it with all the others too

at least the "low quality, worst quality" etc

stuff

ok, i will try it out later, its starting to give me a headache now

lmao

now you know how to use it :d

yes kind of thanks

[ornate echo]

hi furry four legged feline friend

[nova remnant]

zero legs

one sprout

[ornate echo]

...and only 4gb vram 😭

[nova remnant]

[ornate echo]

[nova remnant]

how do I tell ai to show thighs but not the whole thing sometimes

bare legs

Will try that after

Nerfed her a bit

[untold glacier]

👋🏻

[nova remnant]

[native halo]

she can teach me spells anytime

[nova remnant]

"how did I get here, help"

hello lewds /

[stable crane]

hi

I'm trying to get a huge leg bunnygirl but

it does not wanna cooperate

[native halo]

(gargantuan legs:4.5)

[nova remnant]

This model doesnt do flat chests well

0/10

[nova remnant]

I mean this one

[stable crane]

not even with (...:1.4)?

[nova remnant]

will try more later, it looks really nice

[nova remnant]

did with 1.3 not much change from "small breasts"

can probably try more later

Will game now

[native halo]

try (flat chest : 4.7)

[stable crane]

(flat chest:1.3)

[nova remnant]

Yes

[stable crane]

ah

consider negs too

[nova remnant]

I mean

looking back at it

it didn't do terrible

I will try more later

I wish it took few seconds for each try

[native halo]

are these legs massive enough?

[stable crane]

nice

but she has to be a bunny too

(bunnies have large legs)

[stable crane]

nice