🍥｜anime

[supple raptor]

https://tenor.com/view/kana-bonk-hell-gif-26401048

[brittle jay]

https://stackoverflow.com/questions/48146970/python-install-error-a-newer-version-of-the-python-launcher-is-already-installe

You can try stuff from here

[brittle jay]

That we haven't already tried

[native halo]

water girl vs fire girl

[lyric zealot]

[brittle jay]

I afk for a bit

[supple raptor]

Why can't they just be frens tho?

[native halo]

if they touch each other they die 😔 sad story

[supple raptor]

Like being frens with Karlach

[brittle jay]

Ok I back

Who is karlach?

[lyric zealot]

https://tenor.com/view/gif-gif-19494014

[brittle jay]

Bg3?

[lyric zealot]

yup

[brittle jay]

Ah

[wispy canopy]

games i'll probably not play

[brittle jay]

I think I was just at the point before meeting them

[supple raptor]

It's no spoiler that Karlach is hot

[brittle jay]

And then my whole partition corrupted

[lyric zealot]

true true

I didn't manage to befriend her this run

[brittle jay]

Githyanki are underrated

Frfr

[supple raptor]

In my current playthrough I'm trying to make gale hate me as much as possible.

[brittle jay]

Whst

I liked gale for the small bit I played with him

[lyric zealot]

in my next play thru, imma kill everyone i see

[brittle jay]

Just like in that one video

4 barbarian halflings

[supple raptor]

I very much dislike Gale. Won't give reasons because I can't really do so without significant spoilers.

[wispy canopy]

4 barbarian halflings video is pretty epic

[brittle jay]

Well I could see him becoming a worse character down the line

Oh ok

[wispy canopy]

i got elemental girls for all ocasions!

[brittle jay]

@hoary pike so can you install now?

[lyric zealot]

i didn't even bother doing quest for others except for my main 3 companions

[supple raptor]

I'll just say there are both personality and game design reasons for me disliking Gale.

[brittle jay]

Damn

Do you mind formatting your PC?

lol

[wispy canopy]

and that one dude going for any%sex records XD

[brittle jay]

Jkjk that's a last resort

Um

[brittle jay]

That "one"?

[lyric zealot]

bro the world record for the whole game is about the same as any%sex run

[brittle jay]

I think there is a bit more than one

[lyric zealot]

like 1 minute or 2 difference

[wispy canopy]

2 minutes or so?

[lyric zealot]

yup

the massive launch your waifu glitch

[brittle jay]

@hoary pike also that guy that said to delete everything in your appdata. You can skip that part

He stupid

[supple raptor]

"shadowboxing" is a hilarious speed running glitch

[brittle jay]

What is that

[supple raptor]

It goes something like, kill Shadowheart, put her corpse in a box, set the box on fire, jiggle the box around, bam you're now at the final boss of act 2.

[brittle jay]

Bro

[lyric zealot]

https://youtu.be/wGBtcCBw-U8?t=104

[gusty palm]

does anyone wanna start a youtube channel

[brittle jay]

I can send you magnet if you want eface

Only weighs like 100+ gb

[wispy canopy]

no need. it's the same with divinity, divinity 2

can't keep my interest long enough

[gusty palm]

ah i always wanted to paly divinity

[brittle jay]

I gave up when I became 14 and a half

[wispy canopy]

very good games. just can't keep my interest

[gusty palm]

im only 35 im not ready

[brittle jay]

Ok become youtuber

Just need to sell your soul

Jk

[gusty palm]

im ok w that

[supple raptor]

Being based on actual D&D, it is quite a bit different from the Divinity games, but CRPGs aren't for everyone.

[brittle jay]

What you gonna be youtuber about

[gusty palm]

something like a cross between like. 3blue1brown and hbomberguy and funhaus

[brittle jay]

Math?

Video essays?

[lyric zealot]

well, if you intend to give it a go, you can get ||the not so legal copy|| and play for like an hour, and if you like it, then buy it

[brittle jay]

That's what I said

[lyric zealot]

the save file is compatible

[gusty palm]

yeah but not math more like gaming improv with politics video essays and computer art

[wispy canopy]

that's what i do with all more expensive games, actually

gave armored core 6 a go, decided: yeah, imma buy this, i like this

[brittle jay]

So like those that were more popular a couple of years ago

@hoary pike it worked????

Damn

[lyric zealot]

that's actually what i do for most game

[brittle jay]

Let's go

[gusty palm]

yeah ;\

[brittle jay]

@hoary pike default is ok

[gusty palm]

ive been watching a lot of video essays and vaush videos lately and hassan

[brittle jay]

Also maybe sharescreen

[gusty palm]

did u fix ur it/s

[brittle jay]

You need to add to path yeah

What

This might complicate things a bit

[wispy canopy]

more variations of magic girl!

[brittle jay]

But whatever

[gusty palm]

woah

[lyric zealot]

[brittle jay]

Can you send screenshot

[gusty palm]

u could make like a whole gotcha mobile game with that method lol

[brittle jay]

I can't read

[hoary pike]

OSError: [WinError 126] The specified module could not be found. Error loading "G:\Stable
Diffusion\automatic\venv\lib\site-packages\torch\lib\torch_python.dll" or one of its dependencies.

[brittle jay]

He managed to install python

[brittle jay]

Oh um

[gusty palm]

ur gettin there

run update.bat

[brittle jay]

It there one

[brittle jay]

Yeah maybe try reinstalling cuda

Though a1111 normal one is good

Why people not use it

Idk

Like unless they use comfy

Then I get it

[gusty palm]

--update-check cmd line or something similar

[brittle jay]

Ok I will have to go in a bit

So in case something happens you might wanna reinstall a1111 using the tutorial here https://github.com/AUTOMATIC1111/stable-diffusion-webui#installation-and-running

[wispy canopy]

now it's demon queens on a throne variations!

[brittle jay]

Just in case

[brittle jay]

This very cool

I like

[brittle jay]

@hoary pike

[hoary pike]

doing it now

[gusty palm]

kagigori

[supple raptor]

[wispy canopy]

moving onto witch apprentices

[wispy canopy]

niji?

one of these witches is not like the others

[slate narwhal]

sdxlnijispecial-edition, author made a really nice quality upgrade in this version

[gusty palm]

what artist or tokens to get those cool rainbow paint splashes

[wispy canopy]

just add rainbow both in positive and negative

or go with bismuth

[native halo]

in my case i used rainbow and iridiscent in positive

[gusty palm]

oo iridescent

[slate narwhal]

ink wash is nice but then you can get something like this:

[gusty palm]

minimal vector lora works so well with kakigori

which are the essential controlnet models to download

i got the inpainting one

it took a long time though

[native halo]

prob canny,openpose,depth,tile,reference

[gusty palm]

ty

https://civitai.com/models/143636/400gb-lora-xl-repository

whats this

[native halo]

extracted models for testing or experimenting

[gusty palm]

idk what that means ;p

[brittle rampart]

Guys, can anyone tell me where in Auto1111 there is a parameter that is responsible for the time it takes to update the preview of the generated image?

[wispy canopy]

settings -> live previews?

[brittle jay]

This is actually so good wow

Damn

[raven oasis]

can i send a yandere art with a bit of blood or is it forbidden?

[sweet bramble]

thanks 🙂

[magic wave]

A bit of blood is fine

[dawn jasper]

So yeah, still fabulous.

[bitter latch]

can ya guess who she is?

[wispy canopy]

should we know her?

[bitter latch]

mm hmmm

hold on let me make one of her NON furry

or this

Emma Watson

😄

[steady grail]

but you must put a bottle of Ketchup in the image XD

[raven oasis]

what? why?

[steady grail]

it's an old inside joke lol

[nova remnant]

hewwo people

[steady grail]

ohy

[raven oasis]

wait a min, i'll send them but gimme a few mins to upscale them in 8K (sadly, 32K is too heavy for discord and it won't show)

[steady grail]

converte them in jpeg

[nova remnant]

Bro discord shows like 1000x1000

[raven oasis]

oh okay
but i think that 8K is fine

[nova remnant]

Why so much kek

[raven oasis]

it's more when you click on them

how could 1080p images be rendered otherwise?

[steady grail]

8k unless impainted, look bad

details becomes inaccurate

@nova remnant 8k are 8000x8000 and 32k are 32000x32000

[raven oasis]

nah, 8k is 7680 × 4320

[nova remnant]

You'd need to open them in browser to even see that much

[steady grail]

k stands for a thousand

[nova remnant]

No one does that >.>

[nova remnant]

Needs ice

[nova remnant]

Make more ice girls :<

[steady grail]

i like colors

[steady grail]

what a messy girls lol

[nova remnant]

Someone gotta katch her before she does anything bad

[steady grail]

it's always funny XD

[raven oasis]

too lazy to calculate how to upscale them real 4k so i just did 2048x3072

BLOOD warning

[wispy canopy]

i gotchu fam, is just food coloring

[steady grail]

that's not blood, that's red paint, and in her hand, we have an inaccurate brush

[raven oasis]

😂
the knife was a pure pain in the ass to do

[steady grail]

inaccurate brush! lol

[nova remnant]

She was about to make food

Red sauce on the knife

[steady grail]

naturally lol

[nova remnant]

Yandere katchup #ad

[wispy canopy]

oh. i should've removed the ugly

that prevents a lot

[steady grail]

ehehehe

how to open a sausage bottle with a knife lol

[nova remnant]

How to make art using only kitchen tools must watch

[wispy canopy]

ketchup incident

[steady grail]

here we are again, the sausagapocalipse lol

[wispy canopy]

you mean sausage sword?

[nova remnant]

🥖

[wispy canopy]

oh well

model doesn't know what heterochromia is

[steady grail]

just do it by hand

way faster

[magic wave]

Office lady hours

[magic wave]

Damn that looks dope tho :o

[wispy canopy]

surging vitality!

[magic wave]

Office lady always good

[steady grail]

marry me... o damn, it's just a generate image lol

[steady grail]

this character is so spot on!

so, that was my type of woman, i didn't know

[wispy canopy]

"come see me after class"

[magic wave]

Understandable reaction xD

[steady grail]

eheheeh

[wispy canopy]

ok, class dismissed!

[magic wave]

OC

[wet elm]

Damn what model is that? i really like the kinda flat style

[magic wave]

https://civitai.com/models/40199/aurora Aurora (model) +
https://civitai.com/models/139632/greenucorm-style-or-locon-lycoris 1 style LoRA for the base prompt +
https://civitai.com/models/81395?modelVersionId=86360 These two LoRAs at -1
https://civitai.com/models/58390/detail-tweaker-lora-lora +
https://civitai.com/models/70635/cyancapsule-style-loconlycoris 1 style LoRA for the Adetailer prompt (face inpaint)

Big ol list of LoRAs

Also I just realized the Adetailer prompt has GreenuCorm at 0.5 and Cyancapsule at 1.0

[magic wave]

Slavsuka

[wet elm]

holy, tysm

[magic wave]

You welcome

[steady grail]

i'm not good with creepy stuff...

[brittle jay]

Me when 4:40am

[brittle jay]

Correct totle

Title

[brittle jay]

Cute

[brittle jay]

Very good image

[brittle jay]

step on me please I will do anything she looks cool ig

[brittle jay]

insert pp joke here

[steady grail]

charming

[brittle jay]

Bad upscaling. Look at all the artifacts.

Also very little actual detail

It's just artifacts all over the place

[raven oasis]

i'm just using the "extra" tab

[brittle jay]

Oh

Perhaps that's why

[raven oasis]

what do you recommend then? i need upscaling to at least 4K

[brittle jay]

Try upscaling through img2img tab using SD upscale script

[raven oasis]

doesn't img2img change the image?

[brittle jay]

And you should choose a good upscaler if you don't already have one

[raven oasis]

pretty sure it can't go further than 2048x2048

[brittle jay]

Upscaling is supposed to change your image you know yeah

It adds more detail

[raven oasis]

nah i don't want this
i want to generate something like, 720p, and upscale it to 4K/8K

[brittle jay]

You can

But it also add more detail

[raven oasis]

yeah but sometimes it changes the whole scene, or add whole objects

[brittle jay]

Ok listen you can just go to Photoshop or even Ms paint and just make it 4k, but it will be the same image but with wider pixels

[brittle jay]

No no

Right settings it just adds a bit of detail

[raven oasis]

i already got that problem, hands changeds and eyes changed too

and how are you supposed to upscale past 2048x2048?

[obsidian flame]

Question: Does anyone use XL?

btw im back

I know nobody cares or remembers me but I am back xd

[brittle jay]

Anyways I tell you what to do ok.

Go to img2img tab, choose size 512x512, put your image there, choose denoising like 0.25~, go to scripts and choose SD upscale script, download an upscaling model from the link I will send in a bit, put it in the "a1111/models/esrgan" folder and then in the UI select the upscaler you downloaded in the SD upscale script, then press generate

[raven oasis]

okay, i'll try that, can you send it in dm cuz i have to go and may not find the link here tomorrow lol

[brittle jay]

Actually nvm they moved the website and now it's harder to find good models there

https://openmodeldb.info/

Just ask someone here to send you a good upscaler for anime

I am just not home

Man I am way too sleepy

Fucking 5am

I go slep

[raven oasis]

10p.m in france

[brittle jay]

GN y'all

[raven oasis]

good night :p

[brittle jay]

If you need help you can feel free to harass eface or nubby

[raven oasis]

xD

i'll just ask here for a good anime upscaler that works well with 4K and i'll wait until someone answers

[brittle jay]

Vil perhaps you can help them

I slep now anyways

[gloomy scroll]

I think this is what I'm using
https://openmodeldb.info/models/4x-Remacri

[raven oasis]

I'll look at this tomorrow, thanks
What's the max resolution it can support?

[gloomy scroll]

Largest I have done is 2k

[raven oasis]

Oh okay, i planned to make 4K & 8K, i'll try myself then i'll even try 16k and 32k but just for fun, not for everything

[gloomy scroll]

This is largest I have done

[runic vault]

side note use google drive link

[gloomy scroll]

I downloaded everything from old site so I have no idea how that site works

[steady grail]

that's what i had in mid

not perfect, but cool

[raven oasis]

Noice
But when you upscale, does it add more content around the image to fill the gaps or does it just upgrade the resolution ?

[native halo]

it adds more content if u set denoising strenght too high

[gloomy scroll]

Upscaler just upscales

Highres. fix is -> upscale -> i2i

[raven oasis]

Perfect
So i can generate pictures in 720p and get 4K

[gloomy scroll]

Yeah

[raven oasis]

That's perfect ^^

[native halo]

yea this one went from 1080p to 4k

[gloomy scroll]

This can help if you have memory problems ||like my 1070 has||
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111

[raven oasis]

Wait you manage to run sd on a 1070 ??? How much vram do you even have ? 8Gb?

[gloomy scroll]

8GB

I believe you can run sd with 4GB and sdxl with 6GB

[steady grail]

i'm running sd with a 4g

[hoary fractal]

I'm going crazy soon, wildcards are trolling me. That's "1girl, {angry|happy|amused|disgust|disdain|smug|grin|shy|tired}" without hiresfix and then with it. Looks like with hiresfix all options used at the same time, but in pnginfo it's just "1girl, disdain", "1girl, disgust", "1girl, angry" and "1girl, tired"

[gusty palm]

i feel like wildcards and stylepile random make things confusing

i notice when going from txt2img > img2img with wildcards and stuff in the prompt you have to manually edit the prompt cuz the wildcards will stay random

i always copy-paste the prompt from the result manually to ensure the same prompt

im not sure if hirez fix uses the generation prompt from the result or the prompt from the textbox which contains wildcards, i know theres an option to let u manually adjust the hirez fix prompt though

[gusty palm]

if that makes sense

[wispy canopy]

a1111?

uses the same

highresfix uses the actual used prompt

i've never had issues with it

[gusty palm]

ya so if theres a wildcard in it, it will consolidate to a selection

[wispy canopy]

it survived upscale?!

[gusty palm]

im trying to recreate thie effect today

ive extracted the prompt and using the same model but so far not getting very close

[wispy canopy]

you're not on an sdxl model with too low resolution right?

[pale monolith]

that looks like an overbaked lora

[pale monolith]

sharp

[wispy canopy]

found the things the model likes, noticed i can run swinir without oom

2560x1536 😮

2:30 per image ⏱️ 😦

[gusty palm]

the lora is the only part i havent included so u might be right

[wispy canopy]

well peoples, gotta go to bed

👋

[gusty palm]

goonight

[pale monolith]

👋

[gusty palm]

they are livelooping a harsh noise performance

wats the "restart" sampling method

[nova remnant]

@gusty palm

People were talking about restart here

[wet elm]

how do yall get inspiration for prompts and stuff? im out of ideas rn

[gusty palm]

woa thats awesome

[gusty palm]

feels

stylepile has a 'inspire me' button but it mostly just gives me stuff i dont want

wildcards can be good though

if u have the denamic promps extension theres also a chant fancy-watermagic and firemagic that u can modify

it made a perfect tumblr logo unprompted

ok restart sampler is nice

[nova remnant]

meow / 10

Gaming on those arcades

[nova remnant]

just learned about stylepile from your conversation here, looks pretty interesting ..

https://youtu.be/iXP9hE2Vn_s

[nova remnant]

although i go for specific ideas that i come up on my own instead of just generating images for the sake of using sd

and my ideas are mostly around female aesthetics

im not a woman lol

[native halo]

im girl

[nova remnant]

Wait a minute

Why is woman in there

Lmao

im a guy who finds females the most fascinating

I'll try again

[nova remnant]

https://tenor.com/view/culture-man-of-culture-cultured-gif-10903367

👍

haha

well ty

[nova remnant]

I'm intrigued by this helmet crown

It's like half of each

What is it

[nova remnant]

Thats what an AI would say

.>

I need to gen some Justina Gu

Peak beautifulness

restart sampler is pretty neat

She's in very good shape :p

DPM++ 3M samplers tend to add a oily texture

[native halo]

try with unipc

[nova remnant]

DPM++ 3M SDE Karras

unipc is faster i heard

but quality wise not as good as dpm++

Does SDXL have Euler a

i think so

i dont use any ancestral tho

Can you try it on that prompt

I'm curious

im using SD 1.5 btw

Ohh

you want me to try sdxl on that image with euler a?

Wait

Do you need 1.6 to use restart and stuff

I thought it was sdxl

yes

1.6.0

Oohh

I'm afraid to update it lmao

I'm on 1.5.1

you can update safely .. i havent had any bugs with it

[gusty palm]

i mostly like stylepile for when i already have an idea for a subject and just want to see it with different styles like the 'art movement' selection and artist selections

[nova remnant]

1.5.2 broke everything for me

And I luckily got 1.5.1 on a reinstall

lol 1.5.2 broke mine too, i had to resinstall

but 1.6.0 went smooth

What card are you using

rtx 3060

[nova remnant]

Can you do one on Euler a

👀

ok give me few mins i have to add euler a back in my settings, cause i removed it

It's the only one I can use cause the other ones are suuper slow for me :p

[native halo]

try with unipc or dpm++ 2s a karras

[nova remnant]

you want that on SDXL ?

not a lot of good mods for SDXL i can use the base and refiner tho

Do on 1.5

ok

[nova remnant]

Dpm are giga slow

What is unipc about

[native halo]

for me they are about same speed because its ancestral dpm

[nova remnant]

euler a

[native halo]

idk about you tho

[nova remnant]

dpm++ is not that slow on my system

average speed 1.56s/ it

if i wanted to render 2 batches

What was Euler a speed?

Speed: 5.9ms preprocess, 194.2ms inference, 1.0ms postprocess per image at shape (1, 3, 640, 640)
100%|██████████████████████████████████████████████████████████████████████████████████| 10/10 [00:04<00:00, 2.42it/s]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 30/30 [00:39<00:00, 1.31s/it]
Total progress: 100%|██████████████████████████████████████████████████████████████████| 30/30 [00:39<00:00, 1.52it/s]

I had Adetailer on too

can't live w/o adetailer but if i didnt use it render would be faster

I should try adetailer

I'm pretty sure it won't work

why not?

lol i havent had any issue with running adetailer

my gpu can't inpaint without --no-half

just wanna suggest using these two highlights

i think i was using --no-half on my old card with inpaint

and now i use --no-half-vae

[gusty palm]

does masterpiece and high quality tags go at the front or back of the prompt

[native halo]

front

[nova remnant]

@native halo

euler a
dpm++ 2s a karras
unipc

o:

[native halo]

unipc my beloved 😍

[nova remnant]

also why does it show multiple on each

like unipc looks like it had 3 loading thingies :p

[gusty palm]

https://youtu.be/FQ9l4v7zB3I?list=TLPQMTEwOTIwMjN2lq4eVr7eHA

[nova remnant]

Is there something I should know about unipc ?

is it bad at certain things like low (20ish) steps ? >.>

[native halo]

i think u suppose to use it at low steps

[nova remnant]

Euler A

is it the perfect thing for me then

im the low steps enjoyer x3

DPM++ SDE Karras

man the preview images look so weird with unipc

[native halo]

yea they look deepfried 🤣

[nova remnant]

2b from a ps1 game

haha

[nova remnant]

how is yours faster

thats illegal

not sure, i wanted to see the differeence between euler and dpm++

got update for adetailer 😄

they added a new detection script

(full body shot:1.5) of a woman out in the garden sun bathing, long brown hair, (detailed eyes, lips:1.2), wearing tinted sunglass, straw hat, outdoors, beautiful bright day, vibrant colors, highly detailed, ultra quality, dynamic angle,

wasnt supposed to be sunbathing standing tho

[nova remnant]

humans are the weak ones, ai just stands there and gets some sun on the skin

[supple raptor]

DPM++ 2S a and DPM++ SDE (as well as their Karras versions) are the only ones that are slower than typical samplers and that's because they're effectively doing 2 steps for every one. Because of that, you can also usually do roughly half the number of steps compared to other similar samplers. All of the DPM++ 2m and 3m variants (including the SDE ones) are basically the same speed as Euler A.

I believe Restart is another double step sampler like sort of like DPM++ 2S a and DPM++ SDE.

[nova remnant]

im liking the restart

its not slow and render quality is decent

[supple raptor]

Yeah, from the little bit I've played with it, Restart seems pretty interesting.

[nova remnant]

did you see how the dpm++ 3m versions come out?

every images with 3m looks very oily

not sure if its got anything to do with sd vs sdxl

[gusty palm]

it rainy here

[nova remnant]

rain is nice

[supple raptor]

They can look pretty good. This one is DPM++ 3M SDE. The low res pass always looks like garbage to me though. I've heard there might be something wrong with the implementation in Auto1111 and SD.Next.

[nova remnant]

im gonna try a high res image ...

[supple raptor]

I basically never gen without Hires Fix.

[native halo]

yea 1.5 is useless without hires fix

[nova remnant]

gonna use this prompt in 1024 res.. this one is restart at 768

but im using SD 1.5 mod

[supple raptor]

I haven't found an XL model that understands common anime concepts well enough for me to even consider switching over.

[nova remnant]

there is one that looks interesting but has issues with full body rendering

https://civitai.com/models/118729/reproduction-or-sdxl

[supple raptor]

Yeah, all of the ones I've seen are overfitted on a small handful of concepts, and doesn't even really do those appreciably better than good SD1.5 models.

If it can't do the top 50-100 characters on Danbooru fairly reliably then it isn't worth my time.

[nova remnant]

totally agree

im hoping next version of sdxl would make things lot more easy for the devs and users

i also hope as a user it would be optimized in using resources as well

DPM++ 3M SDE Karras with hires fix upscale by 2 from 768 ... the color tone looks vibrant and not oily

the sampler changed the image tho

i used the same seed

[supple raptor]

Yeah, that's pretty common. Some samplers will be basically the same image on the same seed, but others will change the image significantly.

[nova remnant]

Do samplers matter on color ?

yes i think so

in fact im impressed with 3M samplers on high res

samplers can affect the color of images generated by Stable Diffusion. Different samplers use different algorithms to generate images, and this can lead to different results in terms of color accuracy, saturation, and contrast.

For example, the Euler sampler is known for producing images with vibrant colors, while the DPM++ 2M sampler is known for producing images with more muted colors. The choice of sampler will also depend on the specific model being used and the desired output.

In general, it is recommended to experiment with different samplers to see which one produces the best results for your specific needs.

that is what bard has to say

kinda obvious how textures are more prominent with 3M

and rounder

I mean

haha

[nova remnant]

that little rock there must catch a lot of people coming out of the sea, very dangerous :p

[steady grail]

I always thought that the sampler to use depended from how the model was trained

[nova remnant]

yes on the depiction part of it, but textures depend a lot on samplers too

Are samplers shown in a specific order?

Euler a is the first one in my list, I always imagined it was because it's the oldest one or so :p

And unipc is the last >.>

i think they are shown in the first / second / third order of how images are rendered

More wind!!!

Gigantic hair what the hell xd

[nova remnant]

One day I will be like her

used a different mod and it completely ignored some of the prompt specification lol, like i prompted sunset but the image looks like a bright day

[nova remnant]

the ground texture is amazing

[hoary pike]

I'm using control net to generate character lighting for my sketches, but I'm having trouble controlling the values across the character.

[hoary pike]

Actually I figured it out, anyone know what the best sampler would be for controlnet lineart generation?

Trying to wash out the lineart in the result and only have the shading values

[nova remnant]

lmao I just clicked on the reaction didn't expect it to work for a bot

[runic vault]

hmm nope hourglass tag bot is still only a project ai thing

[gusty palm]

xl is very slow i dont see why anyone likes it

maybe i need to try -medvram

im well into sysram

[nova remnant]

thats what i found it to be disappointing, sdxl is extremely resource hungry

[nova remnant]

i'd rather prefer to put sdxl behind just like 2.0

and hopefully the next version would be much better

[gusty palm]

i imagine its probably about the same speed for someone with a n 8+gb gpu

[nova remnant]

doubt it

i have 12gb card and its crap

i disliked sdxl from the very start

[gusty palm]

oh shit

=========================================================================================
A tensor with all NaNs was produced in VAE.
Web UI will now convert VAE into 32-bit float and retry.
To disable this behavior, disable the 'Automatically revert VAE to 32-bit floats' setting.
To always start with 32-bit VAE, use --no-half-vae commandline flag.

wabout this could that be whats making it slow

the result is actually not terrible

[nova remnant]

i use that command line

[gusty palm]

i interrupted it before completion

i let that go almost 7 minutes though

[nova remnant]

7 mins ouch

this is the arg im using --medvram --xformers --no-half-vae

dont need the --medvram i suppose but was testing it out, but didnt seem to slow down with 1.5 high resolutions

[supple raptor]

--no-half-vae increases vram consumption and slows down the vram encode/decode stages. Which VAE do you use?

[nova remnant]

oh man..

well im using basically two

kl-f8 anime

vae-ft-mse-840000

but let me take your suggstion and remove the no half vae and see what happens

[supple raptor]

You're in luck. Neither of those need --no-half-vae. Might I suggest the modified version of kl-f8-anime2 that I uploaded though https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless09.ckpt

[gusty palm]

i dont use vmedram

[supple raptor]

I lowered the contrast level by 10% and applied a patch that helps it to behave correctly in the encode step to help with some color problems that can be caused by the original version.

[gusty palm]

is that new i think i already have that

[supple raptor]

I think I remember having you download that when you were starting. I've had it uploaded for 4 months.

[nova remnant]

going to try a few render with that new vae and having removed no half vae

[supple raptor]

shouldn't be a huge change in speed since it's only effecting the VAE encode/decode stages, but probably a couple seconds on the total time for a Hires Fix image. Lowered vram usage is noticeable though.

[nova remnant]

oddly i think by removing --medvram made my render time go up

[gusty palm]

hmmm

[nova remnant]

they were like 2.56 before

this vae of yours works great

thanks for letting me on it 🙂

[supple raptor]

glad you like it. I very rarely switch to anything else since I hate the NAI VAE textures and I don't really do realistic stuff so don't have much use for mse

[nova remnant]

yeah i will be doing most of the anime with this new vae ... but i'll still use mse for realism

[supple raptor]

yeah, mse has the best colors, it just fails hard at anime eyes.

[nova remnant]

@nova remnant hey

I downloaded adetailer

how do you like it?

anything of note that I should change ?

from default

not really no

the latest update requires 0 tweak

just make sure to choose face 8n

[nova remnant]

face_yolov8n.pt

that one was set by default

i'll see if it works

a few minutes later

ok, adetailer fixes face greatly

no need to use hires fix btw

unless you need hires fix for other factors

@nova remnant look at my name do you think I can use hires

amd diesoutofvram

ohh

i could use hires with my old 4gb card

not too high tho

what does hires do exactly

[supple raptor]

do you have this extension yet? https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git

[nova remnant]

improves the detail in image

[supple raptor]

generates an image at the resolution you set (should be relatively low resolution), then upscales the image, and finally runs img2img on the upscaled image at the new resolution to improve detail.

[nova remnant]

hmm if I'm generating 512x768 without it, do I try lower than that ?

[supple raptor]

Usually that's what I like to use for the starting res for hires fix. You could try starting a little lower and upscaling by like 1.25 or 1.5x with hires fix. I wouldn't go a lot lower though. That multidiffusion extension that I linked a couple messages up will help with vram usage so you can gen higher resolutions (specifically the Tiled VAE feature that it has).

[nova remnant]

how do I know it worked >.>

I think it did, does it save where it would save the image normally ?

[supple raptor]

yep, by default it will just save the improved image in place of the original. you'll see something like this under the main progress bar in the CMD window to show that it worked

[gusty palm]

https://i.imgur.com/32DJzLn.gif

[nova remnant]

yep

[gusty palm]

i should have made it faster shoot

[nova remnant]

that's neat

how did you animate ?

[nova remnant]

I'll try it

I had it before but looked complicated ;-;

Now I'm on a fresh windows since then

[gusty palm]

i generated with a batch count of 4 with super low variation

then just put them in kdenlive and rendered as gif

[nova remnant]

ahh nice

i dunno what kdenlive is but i get its an online service?

[gusty palm]

i wish i could figure out how to like ai morph

its an open source nla video editor

[nova remnant]

ok

[gusty palm]

like vegas or adobe premier

[nova remnant]

neat

have you heard of davinci resolve for video editor?

[gusty palm]

that was the other one i wanna try

[nova remnant]

highly recommned

i love it, even their free version is power packed with features with no water marks or such

i made bunch of edited model videos with music tracks on them at 4k using just their free edition

[gusty palm]

i lvoe doing stuff like that

[supple raptor]

This is the only part that you need to use for saving VRAM and it's relatively simple. Set the checkboxes like in my screenshot but, but you'll need to adjust the two tile sizes down (I have 12GB) I would think around 912 for encoder and 80 for decoder would be good for you. basically you want those to be as high as possible before running out of VRAM. Encoder is the step where the AI turns a generation into an actual image (usually the end of the generation). Decoder is the step where it turns an image into latents that can be interpreted by the model (usually the first step in img2img or inpainting).

[gusty palm]

i think the only reason i picked kden over resolve was because kden is totally opensource and not just freemium but i know resolve is probably better even as freemium

resolve is supposed to be really good for vfx

[nova remnant]

i was skeptic about davinci at first but their free version sets NO limitatons to what you can do except for very high end tasks that you may only need at commercial studio level work

also the ease of editing on it is remarkable

[nova remnant]

I got it :d

I tried with unipc sampler and it didn't work

but euler a did :D

[supple raptor]

You might need to update your WebUI. UniPC only got img2img support in the last version (1.6) which came out 2 weeks ago.

[nova remnant]

I dunno about that

I said before but

[supple raptor]

What Adetailer is doing is using face detection on your image and automatically inpainting the full detection box at a higher resolution. Same as if you selected Only masked for the inpaint area on the inpainting tab.

[nova remnant]

I'm on a 1.5.1 right now, on my other hard drive I have 1.5.2 and it absolutely does not work without no half for some reason

[supple raptor]

Ah, yeah. 1.5 won't support img2img on UniPC so adetailer may not work with it. I would think it should fall back to DDIM for adetailer since that what it did at the time for other img2img operations, but I'm not sure if the Adetailer dev enabled that.

[nova remnant]

ohh

I thought adetailer did inpaint

inpaint doesn't work for me without no-half

[dawn jasper]

Lovely pic there.

[nova remnant]

do I change this upscaler to the ultrasharp one ? >.>

[supple raptor]

yeah adetailer inpaints which is technically just img2img but with only part of the new image blended into the original.

[nova remnant]

thanks!

[supple raptor]

I like 4x-AnimeSharp and 4x_NMKD-YandereNeoXL_200k which are available here: https://openmodeldb.info/

You'll want to lower denoising to around 0.5 as well probably. If you stick with one of the latent upscalers then closer to 0.6 is good.

[gusty palm]

https://i.imgur.com/r54e6wx.gif

[nova remnant]

this is pretty cool

I love how the hair changes too feels like its multiple pictures taken throughout a year or something :p

[gusty palm]

yesss like those "i took a pic everyday for a year"

[nova remnant]

I rarely take pictures

im the "I take a picture a year"

lmao

I should've done some for when my hair was growing

[gusty palm]

same

a few years ago i was super confident for like a few monnths but idk what happened i barely ever feel like taking serlfeis

i feel like if i could understand controlnet better i could make a cool animation though or something

i know how to use it sortof but not sure how to actually incorporate it into a workflow

or like idk how to explain it

i need to watch some tuturoails to see what all it can do

[nova remnant]

Do I throw it in the same folder as my ultrasharp one ?

tried but don't see it, I'll try using hires before reopening the sd to see if it updates

[gusty palm]

img2img upscaler is hidden in the settings but if u use tiled diffusion to upscale it kindof workarounds it

oh hires fix

[supple raptor]

Yeah, it goes in .\models\ESRGAN

Might need to restart WebUI for it to show up.

[nova remnant]

@supple raptor

1.5 with hires

[summer night]

is there any reason to update xformers? i'm on xformers 0.0.17 right now

[nova remnant]

I'll try with tiled vae

[summer night]

i see that the current version of xformers is 0.0.20 but idk what the benefits are of updating

[supple raptor]

yes. though you'll probably need to update pytorch as well. xformers 0.0.21 is latest and it is more efficient, faster, and is deterministic.

[summer night]

man i don't even remember how i installed pytorch

[supple raptor]

simplest way to update will be to just update your WebUI, delete the .\venv and .\repositories\ folders then launch WebUI again. It will rebuild the venv at that point which includes pytorch and xformers.

[gusty palm]

o my xformer is out of date

by 0.0.0.0.1

[summer night]

i like how i can set tags to load with loras now and their default weight, but it would have been nice to have when i downloaded the loras

[gusty palm]

shes silly

[supple raptor]

I got about a 3% speed increase after updating from 0.0.20 to 0.0.21.

[summer night]

i ran --reinstall-xformers and it updated to 0.0.20

[nova remnant]

@supple raptor I got this trying tiled vae "RuntimeError: "upsample_nearest2d_channels_last" not implemented for 'Half'"

[summer night]

unpopular opinion: i like how latent upscaling interpolates things sometimes
even though it isn't the most accurate to the original image, i still like some of the results that come from it

[supple raptor]

Yeah Auto1111 has 0.0.20 pinned. You can update by opening a CMD window in your webui folder then activate your venv by running venv\scripts\activate
it will change the path to say (venv) at the beginning to confirm that it worked like in my screenshot.
Then run this command to update:
pip install -U xformers
once it finishes downloading and installing you can close that CMD window and launch WebUI again.

[gusty palm]

i think linear upscale and then renoised can make decent results sometimes as well

[supple raptor]

hmmm... might be a conflict between WebUI version 1.5 and the latest version of the Multidiffusion extension.

[gusty palm]

whats something i can do to help learn to use controlnet

[brittle rampart]

I still can’t understand whether the amplification in the prompt line works for the phrase between commas or not?

[supple raptor]

Here's a guide that seems pretty good at a glance https://stable-diffusion-art.com/controlnet/

[supple raptor]

You mean something like this (worst quality, low quality:1.2)? That does work to increase emphasis of both worst quality and low quality (this example is from my negative prompt)

[brittle rampart]

yes. but in my example (long fluffy gothic lolita dress:1.4) Is it just the word dress or the whole phrase that gets stronger?

[supple raptor]

everything inside the parenthesis

[brittle rampart]

or should I write like this (long fluffy gothic lolita dress):1.4

or for example ,long detailed black hair:1.5, all phrase or just hair?

[supple raptor]

number has to go inside the parenthesis. This example would work the same as (long fluffy gothic lolita dress:1.1),:1.4 which is to say that :1.4 would be treated as normal text and would not modify the emphasis.

[brittle rampart]

there is no comma before 1.4

[nova remnant]

That is just an example of how it is actually working if you don't put the :1.4 inside the () in the prompt

[nova remnant]

think of it this way, to put emphasis on a description you would use value like so (long hair:1.1) and you can do that with multple description sepereated by comma under the same value

like (long hair, brown hair:1.1)

[brittle rampart]

oki-doki

[gusty palm]

i want a suit like that

[gusty palm]

tyyy

[summer night]

do locon/lycoris models still work in the lora folder? or do i have to put them in their special folder

[supple raptor]

In 1.6 (maybe 1.5?) and newer versions, they work in the lora folder.

[summer night]

ok good

because i don't remember which ones are lycoris

[summer night]

i think
you need to play around with the CFG some more

i updated my webui... where did the buttons go

[summer night]

i don't like this, i should not have updated

[nova remnant]

lol you have to update

but yeah i didnt like that silly change

[summer night]

i don't like this either

i was previously using version 1.5.2

how does this refiner thing work

[brittle rampart]

by the way. guys who use the extension civitai hepler?

[nova remnant]

you'd have to download the sdxl refiner to use with that feature, point to note, once you expand it (down arrow) its active, and collapse is inactive

[brittle rampart]

in the new version of Auto1111, link icons in cards are no longer displayed

[brittle rampart]

it was convenient to navigate to the page Checkpoints or LoRa

[gusty palm]

civitai helper preddy good

er civitai shortcut

lets u scan and update model cards

[nova remnant]

gn, i can barely keep my eyes open

[brittle rampart]

yes. but in the old version of Auto1111 there were always icons from Civitai Helper. in Auto111 v1.6.0 they are not here

[gusty palm]

im not sure i notice that

[wispy canopy]

goodmorning all 🙂

[brittle jay]

Good morning

[brittle rampart]

I mean it. When you press the button, the icons do not appear.

[gusty palm]

oh hmm

i dont know if i ever used those icons

but i c now

[brittle rampart]

you are happy. in my case, I got used to the good things which I have now lost

[brittle jay]

This sounds so deep

This is the same thing that happened in Naruto holy shit

You are sasuke

They are naruto

[wispy canopy]

lol, dem toes tho

[gusty palm]

im not happy per say i just didnt notice a change ><

[gusty palm]

is there a way to download these models manually .. the cmd prompt doesnt seem to like the way my wifi sucks at internet

[supple raptor]

Got a link to the extension? I'll take a look at it to see what I can find. There should be a way to download them manually.

[gusty palm]

https://huggingface.co/AUTOMATIC/promptgen-majinai-unsafe
i found this and

oh i think i found where to put em

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

oddly enough thats not the extension im using i think

[brittle rampart]

interesting

[gusty palm]

its dynamic prompts that uses it but also promptgen is a seperate extension

[supple raptor]

Ah, yep. Dynamic prompts probably just incorporated the original magic prompts into their extension.

[bitter latch]

hmm she's missing an arm

[bitter latch]

oh boy put in yakuza tattoo's and well I can't post that here.. the result that is

[supple raptor]

Here's a list of magic prompt models that should include download links looking through code right now to try to find where the models go, but if you think you found it then you could try that https://github.com/adieyal/sd-dynamic-prompts/tree/main#other-models

[bitter latch]

hmm question. if no nipples are shown but covered up by tattoo's , safe or better now?

[native halo]

i have no strong feelings one way or the other

[bitter latch]

yeah I better not

[gusty palm]

they seem to be somehow associated here D:\automatic111\system\transformers-cache

although

[bitter latch]

a safe version :

[gusty palm]

i dont actaully see the .bin files anywhere hmm

[bitter latch]

nani

naneee?

[wispy canopy]

is that a realistic yakuza miku?

[bitter latch]

shrugs

[bitter latch]

it is a Yakuza Miku....

[supple raptor]

Probably failed partial downloads. Transformers has a built in model downloading function which the extension is likely using.

[bitter latch]

Naruto

[gusty palm]

it freaks out while trying to download cuz of wifi

[native halo]

keep trying it doesnt delete the stuff u downloaded so when u open webui.bat again it will continue with next thing

[bitter latch]

hmm lets try last prompt with SDXL

[gusty palm]

ah ok

i switch to my 2.4g wifi and its slower but less unstable

tolerable its only 300mb

[supple raptor]

Yeah, this is what you're going to want to do. This extension is dumb and doesn't put the models somewhere obvious. They get buried in the Huggingface directories somewhere on your PC. I really dislike this https://github.com/adieyal/sd-dynamic-prompts/issues/220

[wispy canopy]

tattoo'd underwear 🤔

[bitter latch]

swinsuit

swimsuit. supposedly

but yeah does look like a speedo

[gusty palm]

oof

[random valley]

Sure... tell me that its free, but I literally have to sign up to their subscription to use it "for free".

[gusty palm]

i got a thing from google today that was gonna give me 300$ of credit for 90 days on some of theri ai service

[random valley]

wdym?

how did that happen?

[bitter latch]

sounds like a scam to me

'warning warning'

sorta like I got a prince in africa wanting me to help him transfer some funds

😄

[gusty palm]

https://cloud.google.com/ai/generative-ai

not sure exactly what it is

'
Deliver generative AI powered experiences quickly, efficiently, and responsibly, powered by Google’s most advanced technology. Plus, new customers can start their AI journey today with $300 in free credits.'

i think its a 'space' basically

[gusty palm]

the main selling point is that it wont 'autochange you unless u manually continue the subscription'

[random valley]

what's stable swarm, essentially?

I'm looking for some ways to do this whole free SD business away from my GPU because it's just terrible.

[bitter latch]

now th is I would put on a poster

[gusty palm]

its just running sd on a remote gpu instead of ur own afaik

like a render farm

[bitter latch]

ok went a little nuts

but i like it

yawns, ok I need to get back to bed. but few more 😄

[brittle jay]

Cup noodle is epic

[bitter latch]

here we go

[brittle jay]

The tpu thing?

If yes then that's legit

Anyone can easily sign up for it as well

[bitter latch]

ok challenge accepted

😄

oks sets some running in a batch and goes back to bed, nite. thuds

[supple raptor]

👋

[gusty palm]

yeah dynamic promps downloads the pytorch.bin or model.safetensors and then makes a syslink that points to a blobs/ folder full of files with no file extensions

[gusty palm]

im just not sure exactly hwo to use it

[nova remnant]

2b with some hires and adetailer looks nice but takes a while

How many steps do you guys use on hires

[gusty palm]

15-40

[vestal crypt]

8, 16, 20 or 32

[brittle rampart]

30-50

[gusty palm]

whats the best gradio theme

im using moongoblin rn but its got joker vibes

[brittle jay]

Same

[gusty palm]

i get the sense its a cloud host for like a company to run a ai chatbot on or something

or train model idk

3 d anime render of a decora gyaru kawaii cybergoth emo fashion model vtuber, in a cyberpunk blade runner maximalist city of my melody sanrio plushies, artstation cgsociety moody cinematic neon lighting
i hit feeling lucky and i wish sanrio worked as a token

[brittle jay]

Cool

Or rather

Hot

Hehehe see what I did there

[supple raptor]

Heh

[brittle jay]

@hoary pike did it go well in the end?

[zinc flax]

Holy crap $300

[brittle jay]

I think anyone can get it quite easily

I signed up and they gave it to me in like a day or something

And same for my step brother

And one friend

[zinc flax]

Free anime AI stuff

[brittle jay]

https://cloud.google.com/tpu

Not anime

Just free tensor processing units

For like a month

[zinc flax]

ah shucks

[brittle jay]

I mean 90 days

https://cloud.google.com/free/docs/free-cloud-features

[zinc flax]

Need a colab alternative
Gonna maybe try something like runpod lol

[brittle jay]

Or just buy a GPU?

Save up money or something

[zinc flax]

I have this condition called "broke"

LMAO

[wispy canopy]

Become sysadmin at an ai company 😅

[brittle jay]

Haha

Flexing

[gusty palm]

300 credit is alot i just duno how to make any use of it im hoping the deal is still around when i find a use lol

[brittle jay]

Network admin

In a big company with a spare a100

[zinc flax]

Ez A100

[brittle jay]

Literally eface

[gusty palm]

re civit helper

[wispy canopy]

Single A100 is marginally faster than a 3090... It's the memory that's the real plus

[gusty palm]

i think this setting will revert the change u didnt like

[wispy canopy]

4090 way faster than A100 at inference

[brittle jay]

But training tho

[zinc flax]

I wish I had a pc

[brittle jay]

Wait what

I will rn buy you a 100$ laptop

Jk

[zinc flax]

LMAO

[brittle jay]

But like what

There are very cheap PCs out there

How you not have even one

[zinc flax]

I only have a laptop so that's a start

[grizzled oyster]

yoh anyone not busy to help me figure out why my stable diff is crashing?

[wispy canopy]

I've never had a laptop, only pc

So it's not all bad

[brittle jay]

I mean laptop is like already most of the way there

[gusty palm]

I would recommend a bare bones build that yourself over a cheap pre-built

[wispy canopy]

Lan party soon, can't wait to drag my 20kg behemoth all over town again /s

[brittle jay]

I disagree

[zinc flax]

LMAO

[brittle jay]

Do lan parties still exist?

[zinc flax]

"Heck yeah, time to bring my huge ass setup"

[brittle jay]

What does it say

[brittle jay]

Nice nice

[gusty palm]

ive been dissapointed with every prebuilt ive ever had compared to anything ive built from components thats just my personal thing though i gues

[brittle jay]

Prebuilts especially in the lower bracket are usually more cost effective

I think

[wispy canopy]

Friends lan party, and lan parties should exist, it's more about social than games imho

[gusty palm]

i find that it seems that way, but then u realize they gave u like a reallllllly sub par psu sneakily.. or a motherboard with no pcie slot

[zinc flax]

got a 2gb vram laptop

AMD sadly

[brittle jay]

I wish my friends would have a lan party

[gusty palm]

ive always wanted a lan party

[brittle jay]

I once met someone here with an integrated gpu

[wispy canopy]

I used to organize them for like 10+ years

Nowadays just too expensive 😦

[nova remnant]

I'm half as bad as you

With 4

[zinc flax]

Pain

[brittle jay]

Probably something to do with the model or vae

Try changing them

[brittle jay]

Just read the specification beforehand

[grizzled oyster]

i deleted some of them XD

[wispy canopy]

100% self built... I've never had a pre-built ever

[grizzled oyster]

now its working again

[brittle jay]

Some of what

[grizzled oyster]

i had all the first ones in the second folder

[brittle jay]

Eh

[gusty palm]

problem is they rarely list the mobo model numbers and almost never list the psu model

[nova remnant]

give me potatoes

[gusty palm]

especially if its like an hp

[nova remnant]

Gaming computer 1tb HD i7

[brittle jay]

You can return the vae one to the second folder

[nova remnant]

And then some rgb picture

[vestal crypt]

you can try putting the vae into the Stable-diffusion folder
and rename it into something like XXX.vae.pt

[gusty palm]

i admit im using an prebuilt. but ive replaced every component except for the motherboard

i wish i had just bought components knowing what i know now

[nova remnant]

My PSU is the only survivor

450w Corsair something

[gusty palm]

psu was like the first thing i replaced

cuz it had no gpu power thingy

[nova remnant]

Some old ones won't have the connection for new gpus

[gusty palm]

yeah the prebuilt came with a gt720 that didnt need one

[nova remnant]

My other PC doesn't have support for anything xd, super old psu

[grizzled oyster]

wth

now its generating blank images

[wispy canopy]

Very often prebuilts have some proprietary connectors making them pita to upgrade

Looking at you dell & hp

[gusty palm]

i remember the hp we had

the psu was positioned in a way that you could not put a gpu in because it was blocking the pci port