🍥｜anime

[queen void]

Hey :)

[magic wave]

How you been

[magic wave]

SDXL loading may noy be fully functional tho

[supple raptor]

That would be very Automatic111 like considering it was working like a week ago in the rc branch XD

[magic wave]

Fair

Full error

Apparently failed to allocate 54 MB

[supple raptor]

funni symbols make my brain hurt, but it looks like it just ran out of vrams

[magic wave]

Seems like it

Maybe I could try again

[supple raptor]

time to download more vrams

[magic wave]

Exactly, where is those guys that send us the links for the best cheats now

[supple raptor]

updating to 1.5 didn't break stuff as far as I can tell. time to try updating to the new xformers with flash attention 2 to see if I can break something.

[magic wave]

I tried freeing up as much VRAM as possible

And my PC blue screened

XD

[lyric zealot]

[magic wave]

So there you have it lads, SDXL loading doesn't work xD

[supple raptor]

sdxl loading is working for me. It's just trying to use the vae in fp16 which doesn't work on sdxl so i get black image

[magic wave]

Oooof

I mean

I just

Restarted

So maybe now it works

Though I'm also using the bf16 version, not the full model

Dunno if that makes any difference

[supple raptor]

i'll try turning off --disable-nan-check to see if that will make the automatic fp32 vae switching thing work.

[supple raptor]

oh, yeah that isn't going to work. The model size is bigger than your vram.

[magic wave]

Wouldn't using --no-half-vae work?

[magic wave]

Yeah xD

[magic wave]

Well, isn't the full model like 12 GB?

I mean, my VRAM is 12 GB, but

XD

[supple raptor]

probably, but I hate using that because I don't need full precision vae for sd1.5 and it slows down gens and increases vram usage.

[magic wave]

Fair then :o

[magic wave]

Also bf16 is better than fp16 ye? :o

[supple raptor]

yesh

[magic wave]

O wao, das a big boi

[magic wave]

Fair then :o

I could try again

[supple raptor]

also, sdxl vae works in bf16 which is an added bonus

[magic wave]

Maybe just the update being a lil wonky

[magic wave]

Ayy das nice :D

Time to put on Free Bird and go again, my PC either detonates again, or it works

XD

Took like 5 minutes, but it loaded :O

[supple raptor]

mine was slow af too. turning off --disable-nan-check did the trick for fixing the vae problem, but it takes like an extra minute to switch it to fp32 on the first generation.

[magic wave]

Fair, I always got --no-half-vae on

UNIPC IS NOT SUPPORTED FOR SDXL?

https://tenor.com/view/vegeta-gif-19731005

[supple raptor]

might be worth making a copy of webui-user.bat to run it with --no-half-vae --medvram tbh

[magic wave]

Also, SDXL really doesn't like generating at 512/768, well, 0.9 at least

[supple raptor]

sadly, yes

[magic wave]

[supple raptor]

it doesn't. it needs resolutions that would be bucketed into 1024x1024 training at minimum. anything lower looks terrible. The architecture is relatively faster at higher resolutions than sd1.x/2.x though.

[magic wave]

I mean it is faster, but apparently very VRAM inefficient on Auto1111

Can't generate basically, while on Comfy I could generate just fine xD

[supple raptor]

Yeah Comfy is a much better experience for SDXL currently. I hope A1111 gets better because I really don't like using nodes.

[magic wave]

Yehhh, and same

Nodes are wonky xD

[supple raptor]

emphasis is also handled differently in comfy which causes lots of my prompts to look bad in comfy unless I tweak them.

[magic wave]

Oh thats weird, I didn't know abotu that :o

[supple raptor]

a1111 normalizes emphasis so the numbers try to be closer to 1.0 and comfy doesn't. that's why sometimes you can get away with really high emphasis in a1111

[magic wave]

Oh damn, thats fair then :O

[magic wave]

Oh damn I had no clue it worked like that, that makes sense :O

[supple raptor]

yeah, it's more noob friendly than just applying the weights as they're typed, but I can see why comfy would want them to be handled literally.

[magic wave]

Yeee, thats fair :o

[wet elm]

what are the parameters?

[zinc flax]

1girl, floating, sunset, reaching out, looking away, :o, brown hair, floating hair, white dress, sleeveless, collarbone, full body, feet out of frame, cloud
Negative prompt: EasyNegative, feet

[wet elm]

ty, whats the model?

[bitter latch]

haha

and no that wasn't your prompt @zinc flax

😄 yet

[zinc flax]

Personal merge, unreleased

[zinc flax]

Interesting

[wet elm]

ty

[bitter latch]

hhe changed it slightly

Sophia Lillis from dnd movie

clicky clacky noises as he changes more

[bitter latch]

ta da!

[zinc flax]

cow

[magic wave]

cat

[supple raptor]

pat

[bitter latch]

uhhhh

dats weird

[zinc flax]

doog

[magic wave]

[brittle rampart]

Asuka is no longer the same....

[bitter latch]

hmm it seems to like that background. and can you name the actress?

without cheating

[keen hollow]

Mel Gibson

[supple raptor]

https://media.discordapp.net/stickers/1133451870048895127.png?size=160&name=gib_prompt

[zinc flax]

huh i thought it's visible

[supple raptor]

If you copy/pasted the image that probably just removed it

[zinc flax]

PROMPT
1girl, dog ears, dog tails, close-up, :3, :d, paw pose, (chibi:1.2), white t-shirt, long sleeves, floppy ears, fur, one side up

NEGATIVE PROMPT
EasyNegative, (logo:1.2), (fingers:1.3), hand```

[brittle rampart]

Moses and the Great Flood!

[gloomy scroll]

[supple raptor]

miku staying in shape

[gloomy scroll]

https://tenor.com/view/winnie-the-pooh-pooh-bear-belly-tummy-gif-16232018

[magic wave]

[gloomy scroll]

[magic wave]

[magic wave]

Gob gob

[wet elm]

could i have the parameters for that?

[magic wave]

Prompt:

(1girl, one girl), (solo:1.3), (brown_hair:1.3), (inugami_korono:1.3), (twin_braids:1.2), (brown_eyes:1.3), (red_collar:1.1), (thick_lips:1.3), (light_smile:1.1), (happy:1.3), (half-closed_eyes:1.3), (empty_eyes:1.3), BREAK

(standing:1.3), thick_thighs, medium_breasts, (dog_ears:1.1), (steam, steaming_body:1.3), BREAK

(black latex bodysuit:1.3), (hairclip:1.1), BREAK

(black_background, simple_background:1.3), BREAK

best quality, masterpiece, detailed background, highly detailed, intricate, detailed face, long eyelashes, detailed eyes, lora:GoodHands-vanilla:1 lora:Mossa:0.5 lora:boldline2:1
Negative:
negative_hand-neg, (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (NEG-badartistnegOG:0.6), NEG-bad-image-v2-39000, (loli:1.3), monochrome, greyscale, (muscular:1.3), (jacket, coat, cape, cloak:2)
Adetailer Prompt:
(1girl, one girl), (solo:1.3), (brown_hair:1.3), (inugami_korono:1.3), (twin_braids:1.2), (brown_eyes:1.3), (red collar:1.1), (thick_lips:1.3), (light_smile:1.1), (happy:1.3), (half-closed_eyes:1.3), (empty_eyes:1.3), (hairclip:1.1),

best quality, masterpiece, detailed background, highly detailed, intricate, detailed face, long eyelashes, detailed eyes, lora:GoodHands-vanilla:1 lora:Mossa:0.5 lora:boldline2:1

[brittle rampart]

why you use same prompt in ADetailer?

[magic wave]

Its different

Reduced

https://tenor.com/view/thanos-avengers-endgame-gone-reduced-to-atoms-gif-26539274

[brittle rampart]

fumu.... oki-doki and which model in ADetailer?

[gritty sundial]

how?

[gloomy scroll]

[elfin bolt]

animatediff

https://github.com/guoyww/animatediff/

[magic wave]

The default one :o

[elfin bolt]

cli version

https://github.com/neggles/animatediff-cli/tree/main

[magic wave]

Oh damn I just saw that, that's great :o

[gritty sundial]

which is better this https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimizations or this https://www.reddit.com/r/StableDiffusion/comments/158zob5/automatic1111_updated_to_150_version/

[elfin bolt]

quite unstable, instruction is super vague

and the result is only 16 frames as of now

[lyric zealot]

what do you mean?

[elfin bolt]

but this is big for free local installation

[gritty sundial]

these two are stable diffusion install links aren't they?

[elfin bolt]

not that it will beat pikalabs or runway 2 any soon 😦

[gritty sundial]

which one should i install

[lyric zealot]

ah, if you are talking about SD webui installation, then the most popular one will be auto1111

bet like 90% of the users use auto1111

[magic wave]

Fair, still nice tho

[gritty sundial]

No man, I think you misunderstood me. what is that link you sent for?

[keen hollow]

@elfin bolt - you got the gif animatino thing to work?

[elfin bolt]

it also works on nsfw

[magic wave]

Ooof

[elfin bolt]

but prepare for the body horror

[gritty sundial]

is it a new stable diffusion more up to date?

[magic wave]

Fair XD

[lyric zealot]

that link was just a reddit post saying that auto1111 got updated to v1.5

[elfin bolt]

yeah, its pain in the ass to install

[gritty sundial]

oh lol

[lyric zealot]

yup yup, so it's the same thing

[keen hollow]

I didn't have success. You running native linux?

I grabbed the new auto1111--any notable changes?

[gritty sundial]

Sorry I didn't understand, I use a translator because I'm from another country, and the translator translates everything wrong

[elfin bolt]

try this

[lyric zealot]

no problem, but if you to use a good translator, this is pretty good at understanding context https://www.deepl.com/translator

[elfin bolt]

there is auto111 extension for animatediff too

[sonic bolt]

Generates beautiful faces

[elfin bolt]

https://github.com/continue-revolution/sd-webui-animatediff

[sonic bolt]

but when it comes to body, it's trash

[keen hollow]

I did but couldn't figure out how to execute on Windows. Not running native linux.

[elfin bolt]

maybe this is the easiest one to install

[keen hollow]

oh?

[keen hollow]

oh nice

[elfin bolt]

3 days ago it was bugged because the dev did a typo on the coding on motion module

but it was fixed

now

[keen hollow]

ooff.. uses big scary pickle files

[elfin bolt]

yeah proceed at your own risk

the main motion module is .ckpt

[supple raptor]

If you need help with Neggles' AnimateDiff on windows, I can help. That's what I use.

[keen hollow]

I'd be very interested in how you set this up.

[gritty sundial]

Does anyone know how to make images faster?

[gloomy scroll]

I tried installing it and gave up after I realised neggles used different venv than I usually do after '[.dev]' wasn't recognized

[gritty sundial]

mine takes a long time, like 1 minute

[keen hollow]

Someone needs to convert to a safe version. 😦

[sonic bolt]

long ahh neck

[gusty palm]

1 minute is about average for me for a normal quality image

[keen hollow]

Seems fine to me. 😄

[keen hollow]

What resolution?

[sonic bolt]

i know what she could do with that neck

[elfin bolt]

seems like it has different architecture than standard SD checkpoint? its weird that nobody hasnt turned it into .safetensors

[gritty sundial]

default setting

[supple raptor]

you can use regular python -m venv venv (I did) and it should recognize it. If you use conda venv then it's probably different. Gotta run in powershell I think though.

[keen hollow]

Default setting of what?

[elfin bolt]

2k stars on github tho

[gritty sundial]

512x 512

[supple raptor]

Which optimization are you using?

[keen hollow]

I use a different type of sandboxing for all this, so I typically don't use venvs or conda. 🙂

[gritty sundial]

@echo off

set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:51
set PYTHON=
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--allow-code --xformers --skip-torch-cuda-test --api --update-check --no-half

[gloomy scroll]

I think I'm probably using some pip package

[keen hollow]

I got the repo cloned but couldn't even find the dang animatediff script at all. 😐 I'm not familiar with the "-e '.[dev]'" pip command, and feel like that borked for me.

[keen hollow]

what specs?

[supple raptor]

👍 which GPU do you have and which version of xformers shows at the bottom of your webui?

[gritty sundial]

from my gpu or from my settings?

[gritty sundial]

gtx 1660 Super, xformers: 0.0.20]

[gloomy scroll]

Actually I tried that also last time but it didn't create venv folder for some reason

[keen hollow]

mmm, not too familiar with how a 1660 performs...

[gusty palm]

im a 1060 here might be similar

[supple raptor]

You might want to try adding --opt-channels-last to your commandline args and compare speeds before and after. I've heard it can help on 16xx gpus but it slows it down a bit on my 3080 12gb.

[gritty sundial]

error

[supple raptor]

my bad, it's --opt-channelslast

[gloomy scroll]

@supple raptor

[supple raptor]

you used single quote not backtick around the .[dev] right?

[sonic bolt]

fixed long neck problem

[gloomy scroll]

I jsut copied from read me

[keen hollow]

Would you use this toilet?

well I guess that killed the chat

[supple raptor]

are you running in powershell or cmd? I just tried again with the --upgrade flag added because I needed to update mine anyways and it's working for me in powershell.

[gloomy scroll]

cmd

Guess I'll try again with ps

[gusty palm]

does batch count not use variation or is it the sampler im using, it seems to generate images with different seeds

[gritty sundial]

How do I create a shared link?

[supple raptor]

not sure how it would work with other sandboxing options, but the way this works is it uses the setup.py to install the dependances. As I understand it, it's just an alternative to using a requirements.txt file.

[gloomy scroll]

Okay started installing with powershell

[supple raptor]

seed variation is only used to generate a slightly different image with a fixed seed, both batch count and batch size will iterate the seed number by 1 for each additional image genned. I believe the seed variation settings will be applied to the new seeds as well if you use it, but will be less useful in that case.

[keen hollow]

hmm, never ran the setup script, i suppose I could try that. May wait for the models to be created in safe versions.

[gloomy scroll]

@supple raptor what xformer version do you use with animatediff?

[supple raptor]

0.0.20. Just use pip install -U xformers

[gritty sundial]

share=True

[supple raptor]

If you're using a python venv in powershell, you shouldn't need to directly run the setup.py script. That might work for other methods though.

[gloomy scroll]

Next time to run into OOM error

[supple raptor]

oh, you want to know how to activate this? just add --share --gradio-auth username:password (replacing the words "username" and "password" with the actual username and password that you want to set for the WebUI). Since the shared link is public, you absolutely want a username and password.

[supple raptor]

did you download sd1.5 in diffusers format as well as the motion modules already?

[keen hollow]

wasn't the share stuff considered extremely insecure?

[gloomy scroll]

Is it this?

[supple raptor]

That was because people were using it without a username and password so all someone had to do was guess/bruteforce your URL to gain access to the WebUI.

[gusty palm]

ive shared my server with like my closest bffs but i trust them w my life basically

also gpu temp protection helps lol

[keen hollow]

I thought the auth implementation was determined to be pretty bad.

[supple raptor]

That's one of the motion modules. I would recommend getting the sd1.4 version as well as it seems to actually be better. https://huggingface.co/guoyww/animatediff/blob/main/mm_sd_v14.ckpt

You'll also need the sd1.5 model in diffusers format which is the stuff here other than what I've crossed out in the screenshot https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main

[supple raptor]

I'm not aware of that, but it isn't something I've paid a ton of attention to since I don't use it.

[gloomy scroll]

||All this trouble for OOM||

[wet elm]

why do you separate some parts of your prompt with BREAK and blank space? does it help with something?

[gloomy scroll]

Blank space makes it easier to read

[gloomy scroll]

BREAK: #🍥｜anime message

[wet elm]

oh so it does not affect the generation?

[gloomy scroll]

Yeah blank spaces are ignored

[supple raptor]

BREAK forces the CLIP to start a new block of 75 tokens which can help with tokens bleeding into other tokens, and applying the strength that you expect to them.
Blank spaces are ignored by CLIP and are there for human readability only.

[wet elm]

oh thats nice to know, ty

[magic wave]

I use em to separate like the 4 main parts of my prompt

[snow flume]

prompt:
Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background

[keen hollow]

nacho time

[keen hollow]

ok i'll bite

[wet elm]

damn that is useful, so like if i want to describe a character could i do something like

1girl, black hair, long hair, red eyes BREAK

schoolgirl uniform, skirt, sailor uniform BREAK

hairclip, hair ornament, headband BREAK

quality tags here

[keen hollow]

ok I"ll bite on this one too 😄

eh, less interesting...

[magic wave]

Yeah, I usually break em into Face / Body / Clothes / Background / Quality

I also just realized thats 5 parts not 4

xD

[magic wave]

But, seems to b gud

[wet elm]

lol i just thought of the most generic thing

[wet elm]

yeah that seems nice, sometimes my images will miss something that i put towards the end

[snow flume]

Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background_

[wet elm]

thats mystical af

[magic wave]

Yeee, though what I like most about it is the less bleed on prompts, specially colors

[snow flume]

prompt:
Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background

[supple raptor]

Yep, just keep in mind that tokens nearest to the beginning of a block of 75 tokens are more inherently strong than ones that come later, so you may have to adjust the order of your tags for this prompting style.

[wet elm]

oh so like the thing where you say for example red hair, blue skirt and it mixes the two

[supple raptor]

yep

[magic wave]

Yeee

[wet elm]

tysm this will help a lot :)

[gloomy scroll]

I think I need few more extensions since this block seems too smol

[magic wave]

[snow flume]

prompt:
Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background

[snow flume]

prompt:
Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background

[snow flume]

prompt:
Ethiopian AI
Make more like black Girl half Robot and use Ethiopian Flag colors Green Yellow Red and blue
Add some Amharic Alphabet in Background

[supple raptor]

When writing your prompts, try to write it as a description of the image rather than as if you are telling someone what to draw. Also, you can generate your own images for free in #1100170312106127410

[gloomy scroll]

1girl, big boobs, thick thighs, awesome background
Did I do it right?

[gloomy scroll]

Maybe I did something wrong

[supple raptor]

1huge, girl boobs, thiccccc thighs, really really awesome background
Is what you want.

[gloomy scroll]

[supple raptor]

if the thighs can be described with less than 5 c's are they even worth mentioning?

https://cdn.discordapp.com/emojis/819258048593264671.gif?size=64&name=herobonk&quality=lossless

[wispy canopy]

can't wait for sdxl1.0

[wispy canopy]

don't make me go back to the horny models to summon these succubi XD

[gloomy scroll]

#🍥｜anime message

[wispy canopy]

sigh, unzips

[supple raptor]

it used 0 booru tags so it should be @wispy canopy approved

🤣

[wispy canopy]

lemme load up horni waifu generator in th emeantime

[wet elm]

what is RP?

[gloomy scroll]

You trying to make me read that block?

[gloomy scroll]

Also RegionalPrompter

[wet elm]

i just looked it up and didnt understand a single thing lol

[wet elm]

yes 😈

[wispy canopy]

of course... it doesn't respond well to 1girl... guess what is left xD

udder horror lol

[gloomy scroll]

https://tenor.com/view/bouncing-tits-birds-gif-12159497

[wispy canopy]

... what was i thinking

[gloomy scroll]

RegionalPromter can specify where parts of the prompt have effect

[wet elm]

ayo since when could we do animations? also it prob requires like 10gb ram so i wont even bother lol

[gloomy scroll]

Animatediff can make animation from still image

[supple raptor]

512x768 uses 8.5GB. It's using AnimateDiff

[wet elm]

wait... so i could divide an image in two and describe different characters?

[gloomy scroll]

1070 8GB

[gloomy scroll]

ye

[wet elm]

thats cool im gonna download it lol

because i cant get two different girls in same picture without inpainting or smth

[wet elm]

ty

[supple raptor]

This is the repo that I use if you want to try it out. https://github.com/neggles/animatediff-cli

[stable crane]

I just realized that CFG 15 can look really good sometimes

[gloomy scroll]

Using latent couple extension which is somewhat different than RP

Since I don't have example with RP

[wet elm]

hmmm interesting

[gusty palm]

i almost alwayus use 15 cfg

[wet elm]

wide af damn

[stable crane]

I've been on 7 CFG for a long time

[gusty palm]

nubby inspired me

[stable crane]

but the colors are really popping at 15

[gloomy scroll]

I was wondering why that image looked weird but after seeing that second one...

[supple raptor]

i gen at 7.5 cfg

https://cdn.discordapp.com/emojis/1088512173908709466.gif?size=64&name=mikupeek&quality=lossless

[gloomy scroll]

7

[unreal ridge]

8

[stable crane]

still won't listen to my instructions at 15

[supple raptor]

I used to use 11.5

[gloomy scroll]

You could try dynamic thresholding if you want even higher cfg

[stable crane]

15

will compare to 7

[stable crane]

a plugin?

[gloomy scroll]

Extension

[stable crane]

I can't post the 2nd image

[gloomy scroll]

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding/wiki/Usage-Tips

[stable crane]

but trust that it does not pop as much

[supple raptor]

in which way?

[stable crane]

both colors and other features don't pop as much

usually I would experience image corruption higher than 12 though

[gloomy scroll]

7 vs 15

Didn't bother doing highres fix

[stable crane]

maybe depends on the prompt or model too

sometimes it can get so oversaturated like you see

[elfin bolt]

anyone tried this?

https://github.com/dvruette/sd-webui-fabric

Poor man's personalized RLHF

It uses like and dislike system just like midjourney and SDXL

[stable crane]

hmm

it's actually compiling some kind of model alteration?

[gritty sundial]

does anyone have an anime model, well, that is not generic

[elfin bolt]

it's uses your input images as positive and negative feedback on the autoencoders

it messes with the decoder process of generating image

[stable crane]

wym by generic?

[stable crane]

oh cool

[elfin bolt]

generate batch images, give them like and dislike, Fabric put those liked image as positive feedback and disliked one as negative feedback. In theory, you could get the image you wanted after several iterations

[gritty sundial]

Generic Vs No Generic

[unreal ridge]

oh, I see what you mean

[stable crane]

you could always grab some style loras

right looks like pika's model a bit

[unreal ridge]

or just browse civitai to look for something that doesn't look too generic

[stable crane]

this is lottamix2, I feel like it's not generic

[gloomy scroll]

I this what you would call generic?

[gritty sundial]

no

[gloomy scroll]

Good since it took hours to get that prompt working with all those extensions

[supple raptor]

I-is this generic?

[stable crane]

that's cute as hell

[gloomy scroll]

"A...Am I g...generic "

[unreal ridge]

good hands 😭

[gritty sundial]

no

[supple raptor]

Kakigori_v2 will be much better than v1 was at hands 🙌

[gloomy scroll]

I forgot I was supposed to test kakigori v2 rc1

[supple raptor]

https://cdn.discordapp.com/emojis/1118589868726034432.gif?size=64&name=MomijiWag&quality=lossless

[gloomy scroll]

I was working on my magical girl project

Now I need to figure out how to get good mecha hands

[gritty sundial]

generic VS no generic

did you understand what i meant?

[gloomy scroll]

So something else than NAI or Any

[gloomy scroll]

Well Kakigori isn't generic

[gritty sundial]

this pose, this feeling for the camera, I find it generic

[stable crane]

it actually seems hard these days to gen the one on the left

[unreal ridge]

I don't think the one on the left is generic at all

[gritty sundial]

it's just an example, not specifically like that

[stable crane]

"generic" too generic of a word

[elfin bolt]

there is no such thing as "generic" in artistic viewpoint

[gloomy scroll]

Here I can't think any of these would be generic
https://huggingface.co/closertodeath/ctdmixes/tree/main

[elfin bolt]

maybe u have been saturated with such images in the past and labelling on the left as generic

[gritty sundial]

I think you didn't understand, I'll take many examples for you to see

example, it seems that it is the model that makes it generic, you know, it seems that the model makes the art generic, I'm not talking about the characters

[gloomy scroll]

Me remembering wall of AnyV3 images

[gritty sundial]

but changing the subject, don't you have a good anime model for me?

[gloomy scroll]

I feel like I have been blocked

[unreal ridge]

reminds me of some of kether's gens

[elfin bolt]

glad we diversified alot from that

Now there is a heavy incestous problem in civitai for asian faces

[supple raptor]

that left one is one of kether's gens

[gusty palm]

hey thats mine uwu

[unreal ridge]

oh

[steady grail]

what does fabric? training?

[gritty sundial]

lol

generic or no generic

[gusty palm]

the one on the right

[gloomy scroll]

@supple raptor Time for seed hunting

[gusty palm]

i loved it idk if its generic or not

[gritty sundial]

this is the perfect and no generic model I need, what's the name?

[gloomy scroll]

I'm testing Kakigori v2 but v1 is available in CivitAi

https://civitai.com/models/100505

[gritty sundial]

I think you don't understand what I mean by generic

[gusty palm]

i think like sometimes things arent very unique which isnt always a bad thing

that star fleet picture was a result of stylepile randomly giving me a cool art movement and artist combo

and then upscaling revealed faces hidden in the paint splatters

[gloomy scroll]

I think generic for Roluda means images made with NAI style

[gusty palm]

its super subjective

[elfin bolt]

Kind of but not quite, the extension allow you to hone in to images that you liked by giving feedback

[gusty palm]

wats nai

[unreal ridge]

Novel AI

[gusty palm]

ah alr

[elfin bolt]

it's similar process to https://discord.com/channels/1002292111942635562/1108646783090958336 this channel

[steady grail]

where do you apply such feed backs? on training? or on prompt?

[elfin bolt]

in the extension panel on auto111

[steady grail]

yeah, but how this affect your generations?

[elfin bolt]

lets say for example, you wanted to generate "1girl, inside bottle" and u generate 3 images

1 image shows a girl inside bottle while the others show girls holding a bottle

you give a like to that one image and dislike to the rest

then start generating with the same prompt

as you iterate more, the extensions will force the models to generate an image that you like to see (which is a girl inside the bottle) and only that

[steady grail]

cool

refine the image from gen to gen

[gusty palm]

there was some early ai projects that trained models like that using community input

vaguely similar to that i think

wait so were u saying u liked my paint splattery pic or is it bad

[unreal ridge]

wide

[gloomy scroll]

[gritty sundial]

where do i put lycoris model

@gloomy scroll

[gloomy scroll]

Lora folder

[gritty sundial]

I don't have to download extension no?

[gloomy scroll]

But I don't know if webui supports lycoris by default or does it still need extension

[gloomy scroll]

All Lora/Locon/Lycoris/Loha go to lora folder

[gloomy scroll]

I have these but don't know if these are still needed

[gritty sundial]

AssertionError: extension access disabled because of command line flags

[steady grail]

if they go in lora, why i have a lycoris directory?

[gloomy scroll]

Works either way

[steady grail]

oh, good to know

i expected a different result

[gloomy scroll]

Have you edited your webui-user.bat?

[stable crane]

really nice

[gloomy scroll]

Here's without and with third eye

[stable crane]

@supple raptor can you @ me when v2 releases? 😄

[stable crane]

third what?

[gloomy scroll]

fixed

It's 3am

[stable crane]

aha

[gloomy scroll]

New try

[supple raptor]

Check your DMs

[gloomy scroll]

I should rly make openpose stickman smaller but editing mask was easier

@supple raptor lanscape looks rly beautiful

[stable crane]

really nice composition

the girl is also mad cute

[supple raptor]

yeah awooo7 and vanilla both enhanced the landscapes on it I think

[stable crane]

OSError: [Errno 28] No space left on device

sad times

[supple raptor]

[stable crane]

idk how this keeps happening

can you set your SD model folder to another folder than A1111 install?

cause at some point I gotta just move them all out of there, do a new install

[gloomy scroll]

I use mklink

[stable crane]

smart

linux / wsl?

[gloomy scroll]

win10

[stable crane]

oh win supports it too?

[gloomy scroll]

mklink /D "\webui\models\Stable-diffusion" "\path\to\models"

[stable crane]

ty

[stable crane]

so crisp

is that pure txt2img?

I see you had adetailer

@supple raptor it's cute

[unreal ridge]

I feel like I'm using too many loRAs. The walls get all cursed depending on the prompt

[stable crane]

my gens don't look like Vil's though 😢

[unreal ridge]

ayo 👀

[supple raptor]

Which VAE are you using?

[stable crane]

anything v3

[supple raptor]

That one might not have high enough contrast for it. I use my 90% contrast version of the wd vae (which is also baked into the model for convenience). Blessed VAE might work better if you want to stick to a NAI based VAE.

[stable crane]

oh I'll try the baked one now

[unreal ridge]

Bob cut >

[stable crane]

@gloomy scroll 's girl

[supple raptor]

One of the models in the mix has quite low contrast on it's own, so a low contrast VAE might struggle on it.

[lyric zealot]

debatable, because i still loved Ponytail

[unreal ridge]

I lied

Hime cut is my favorite

[supple raptor]

short hair with long locks >

[unreal ridge]

BALD

[gloomy scroll]

Did you check my images exif?

[stable crane]

just saw a baldifier lora

[stable crane]

I did, that was your prompt above

[unreal ridge]

oh no

[gloomy scroll]

The prompt is not special part but compination of Controlnet and RegionalPrompter

[gusty palm]

is adetailer the same as lora:add_details

[gloomy scroll]

Adetailer is extension

[unreal ridge]

nope

[stable crane]

lmao

[unreal ridge]

LMFAO

Fuck, I spilled my drink

[stable crane]

ahahaha

[gloomy scroll]

Adetailer is automatic inpaint for face/hands/body and with default settings it pretty much increases face quality every gen imo

[gusty palm]

ooo

[supple raptor]

imma be honest... I expected it to be much less cute XD

[slow needle]

our dorito of virtue

[keen hollow]

Re-sharing this from yesterday just because I like it so much. 😄

[gusty palm]

is there a resource to download VAEs or

or upscaler models

[bitter latch]

civitai is the most popular

not sure about upscaler

[gusty palm]

o they have a vae filter neat

[supple raptor]

upscaler models come from here: https://upscale.wiki/wiki/Model_Database

[gusty palm]

sweet

[bitter latch]

whoa

[bitter latch]

a1111 has some big updates last few days

[gusty palm]

how do update

[bitter latch]

I put in my web-user.bat file the line git pull

git pull

hmm seems they added color correction

or is that one of my extensions

?

[bitter latch]

oh thats adetailer

hmm not sure I like that yet

makes th ings weird

[lyric zealot]

you meant the adetailer?

[bitter latch]

yeah that

thats without

[lyric zealot]

depending on what model you are using, i think it will fix faces most of the time

example of how adetailed fixed eyes

[bitter latch]

hmm ok so it helps

got it

tries a few diff things

[lyric zealot]

it's really good at far shot, when your subject is kinda far away

[keen hollow]

what updates? I didn't really notice anything

[lyric zealot]

new auto1111 update supports sdxl

you do know that the first one is straight up nsfw right?

and spoiler-ing it isn't gonna be okay

[gusty palm]

is there a way to take 2 images and like, seamlessly merge them side by side

[lyric zealot]

the way i can think of, is to merge them using some external program like photoshop, then import them to SD with inpainting, inpaint the middle part and give it your prompt

[gusty palm]

ookie

[lyric zealot]

just notifying you again, you need to remove the first pic because it's straight up nsfw

[indigo wagon]

I'm sorry it was just a question I didn't have another picture

[lyric zealot]

i mean, i understand, just don't want you to get banned

[lyric zealot]

and so, you can probably get similar lora from civitai

[indigo wagon]

Oki thanks so much for letting me know! where can I learn to use it? I'm new to AI and I don't know what this is

[keen hollow]

crazy, no controlnet or anything

[lyric zealot]

well, you are in SD discord, so people would probably recommend using SD. There are many ways to use it, a popular choice is a webui called auto1111. https://github.com/AUTOMATIC1111/stable-diffusion-webui

[keen hollow]

I love getting just absurdly large canon guns 😄

[bitter latch]

thats what she said

[keen hollow]

😐

[bitter latch]

hmmm (turns adetailer/ed? back on)

[gusty palm]

what kind of prompt do u think would be useful for that

im not sure which images i wanna join exactly yet but

[lyric zealot]

depends on context, like you will probably wanna prompt something that can be the link between the two pic

or you can even just not prompt anything (will probably not work tho)

[stable crane]

@supple raptor i like it more than v1

[supple raptor]

https://cdn.discordapp.com/emojis/1093687331023765546.gif?size=64&name=owoSpinFaster&quality=lossless

I think I liked the best images from v1 a little better, but it's way easier to get good images on v2

[stable crane]

I love going back to old gens with new model

interesting to see how each renders

[finite herald]

meemku

[gusty palm]

i tried using the wd-sd-trinart-ld model i got from the sygil ui, and the results are so weird

it does like interior rooms really well but anime and stuff is so weird looking

[stable crane]

I always got jumbled stuff out of trinart

@supple raptor I love it

first lotta, turned out amazing

[gusty palm]

is lottamix urs

[stable crane]

yeah this is me https://civitai.com/models/94401/lottamix

tf it's a 404 page

[gusty palm]

:3 i downaaaloded it and followed u

[stable crane]

oh nvm it works now 😄

[stable crane]

nice, hope you like it 🙂

[gusty palm]

lets see what clip skip 6 does

[gusty palm]

thanks

[gusty palm]

is (token) the same as (token:1) and token

[polar anchor]

Hello 🙂 I was told to come in here for some possible help with prompts/models/loras. I’m really trying to create landscapes and scenes. I had ONE image that I like and now I can’t replicate it. I like the old style cartoon anime style. Cel shading, flat color, etc. I’m struggling to get that. And I’m struggling with my prompts. It seems like it’s only picking up half of what I’m actually wanting it to make. Any suggestions would be appreciated. I’m attaching my butterfly garden which is the style I’m trying to keep (yes it’s mine, I’m not copying anyone’s pic) and then another I generated today that just nothing like it.

[lyric zealot]

no, (token) = (token:1.1)
yes, (token:1) = token

[polar anchor]

For reference the prompt for my second image was “a studio ghibli illustration of a mystical bazaar at night, moon shining, potion and enchanted shops, spooky theme” I have like no detail in it, it didn’t really get what I wanted it to make. I’ve tried different models and I’m getting the same “look”

[supple raptor]

Kakigori_v2 is now publicly available https://civitai.com/models/100505?modelVersionId=126070

[umbral depot]

OMG

NUBBY!

[supple raptor]

How have you been? Haven't seen you around much lately.

[umbral depot]

I uhhh took a break

Been good though how have you been

[polar anchor]

Was this to me? Lol

[stable crane]

everyone download this

[stable crane]

new model?

[zinc flax]

I guess you could say that
A pretty recent mix I did to test if I can fix hands with mbw lol

Looks pretty consistent fortunately

[supple raptor]

I was just letting everyone here know since there's been a fair amount of interest in it as I've been posting my progress pics. Have a look at the preview pics and see if it looks like something you would like. It does happen to have more of the flat colors that you mentioned you were looking for.

[polar anchor]

I see that! I’m downloading it now! 🙂 thank you so much

[stable crane]

it's very versatile

[supple raptor]

My pleasure! Sharing the stuff I make is half the fun. I hope you enjoy it

[polar anchor]

I think I need to work on my prompts lol. I feel like it’s not the models fault. I need to get out of the MJ mindset lol

[supple raptor]

Prompting is definitely a bit different on these models. Click the button I've circled in red on any of my preview images and you can view my generation parameters and prompts for some examples.

[polar anchor]

Thank you! I’m seeing what happens lol trying something now if I get good stuff I’ll share it on civit to help out!

[supple raptor]

I appreciate it. Let me know if you have any questions. Might be a little slow to respond for the next hour but I'll be around for a while after that.

[polar anchor]

Oh that’s ok. It’s 1AM here. I need to sleep but AI Art is life lol.

I was trying to make something cool for my page for tomorrow but I’m too annoyed lol

[magic wave]

I just saw this, but whats your negatives?

And also VAE

[polar anchor]

You know. I was today years old when I realized I didn’t have a Vae set. Ugh. Any recommendations?

And negatives right now are “CGI, 3D, realistic, 2.5d, blurry, bad quality, bad anatomy, ugly, low resolution, logo, signature, people”

I’m well aware my negatives probably suck lol so feel free to correct the shit out of me

[polar anchor]

🥺🥺🥺 thank you. Seriously. Lol now I have something to keep tweaking but this is perfect so far

[magic wave]

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt Use this one

And those negatives seem fine

Though I personally use

negative_hand-neg, (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (NEG-badartistnegOG:0.6), NEG-bad-image-v2-39000, (loli:1.3), monochrome, greyscale

[supple raptor]

^ For my model you don't necessarily need that since a customized version is baked into the model ^

[magic wave]

(the embeddings i use)

[magic wave]

Oh fair then :o

[polar anchor]

@supple raptor I’m so excited! Ok I can sleep now lol. I at least feel good I found a model lol

[supple raptor]

Glad you're liking it. Your gen looks better already. Get some sleep and I'll look forward to seeing what you come up with tomorrow!

https://cdn.discordapp.com/emojis/999416119045656596.gif?size=64&name=YuukaWave&quality=lossless

[gusty palm]

awesome downloading nubby

[brittle rampart]

Have a good day Sensei! There is a question. What is your opinion on the usability of the Lobe Theme for AUTOMATIC1111 WebUi?

[autumn violet]

Looks really good, might be fun to mess around with merging too.

[supple raptor]

Thanks! I think this version should be more compatible for merges too. Adding awoo7 seemed to help a lot with that.

[autumn violet]

I still don't know completely what I am doing, but it is fun to mess around with.

[gusty palm]

I installed lobe but it said it was incompatible with cozy

[bitter latch]

didn't come out the way I wanted.. odd

worked fine in non anime but put in anime model and bam

suppose to be day of the dead voodoo queen with death mask

well I guess but no

[supple raptor]

A lot of it is just guessing what will happen, but there's certain parts of it that you can get a feel for to make your guesses more accurate.

[autumn violet]

Yeah, I used the charts a bit when I tried merging superdupermagic mix or whatever it was into mine.

I don't feel like I figured out that extention though. It didn't seem like it was actually using the sliders I set, so I used the old one. I will have to try and work with it more and try lora merging too some time.

[bitter latch]

ok don't mess with her...

ok to bed. nite

thud

[supple raptor]

Oh, they probably weren't then. You have to make sure the MBW checkbox is checked and click the "Set to Alpha" button to send the slider values to the Weights for Alpha..." field. You can confirm you did it right by looking at the merge recipe shown above the preview image or by looking at the recipe in the console.

[autumn violet]

I checked the box, but I don't think I did the second thing. Thanks for the tip! I will mess around with it tomorrow probably.

[supple raptor]

Yeah, I missed that the first few times I tried too. It isn't an intuitive interface, but once you figure it out it works well.

Feel free to msg me if you need help

[brittle rampart]

@supple raptor Thanks for the new version. according to my modest tests, the new version works almost twice as fast! It was one of the nice things. Without scaling, the image acquisition time was reduced quite a lot! At the same time, the quality is on top as in the first part! I am very pleased with the opportunity to work with your model

[supple raptor]

I'm glad you're liking it! I can't take credit for any speedups you might have gotten, but I'm happy the quality came out as good as it did

[supple raptor]

Do you know what you used in your mix to get such good :3 mouth? Mine basically never come out as good as yours on anything I've tried.

[zinc flax]

Used 3A for it

https://huggingface.co/sleepotimer/Model_A

Here's a better example I guess lol

For that merge I used 3A only for the face + hands btw

[supple raptor]

Thank you! I just realized I have this one in my models folder and haven't even used it yet XD I'm gonna have to try it out a bit

[zinc flax]

Yes yes do try it
It's an accidental mix that turned out great lol

And with surprisingly decent hands

[supple raptor]

It really did! All of your gens are mega cute

[zinc flax]

rip feet though

[supple raptor]

Feet are the real challenge. Most people just don't notice it because they're hidden in like 95% of gens due to framing or shoes.

[zinc flax]

True that, or something like feet focus I guess

Old gen with feet

Though it's all gacha in the end

[autumn violet]

Will do, thanks for always being willing to help. I appreciate it a lot!

[brittle rampart]

the style just got cuter!

i like it!

[supple raptor]

I'm slightly annoyed that of all possible times, this is when v1 decides to bless me with 8 mostly decent hands XD

[brittle rampart]

it seems to me that there have always been BIG problems with the fingers

[autumn violet]

Hands? We don't need hands.

More fingers is better actually

[brittle rampart]

oh nooooo!!!! hands and fingers is a beutiful!

[autumn violet]

Eventually everything will evolve to crab anyways.

[brittle rampart]

everything should be in moderation and in the right amount

[autumn violet]

I'm just messing around

It is late

I should sleep

[brittle rampart]

g`night

[brittle rampart]

it was very atmospheric

[glacial bane]

What anime style is thizz

[zinc flax]

ExpMixLine v3?

[glacial bane]

Art style

[brittle rampart]

in my opinion, v1 better depicted objects taken in hand

[glacial bane]

You really like kids huh.

[glacial bane]

What kind of short are those again?

[supple raptor]

Yeah I'd agree based on the last ones

[glacial bane]

That went old to gold.

[queen void]

I'm back :)

[glacial bane]

@lyric zealot what artstyle is thiz? Been tryna replicate it

[brittle rampart]

@supple raptor It seems to me that in v2 more attention is paid to the background. but emotions are slightly distorted in relation to the request
v1 and v2 on f same setting and prompt. however, the colors are more saturated and vibrant

[queen void]

nice

[wispy canopy]

sdxl 0.9

can't wait for the 1.0 drop ^^

[brittle rampart]

how i love the style of watercolor and paints!!!

[summer shuttle]

i usually set the cfg scale to the lowest to reduce the contrast

[brittle rampart]

how much?

[summer shuttle]

1-1.5

depends on what model youre using

i use breakdomain

[brittle rampart]

hmmmm

[summer shuttle]

details

[brittle rampart]

I already watched PNG-info. thanks

[brittle rampart]

That REALY teleport!

[lyric zealot]

I believe I was using expmixlineV3, maybe with negative moredetail Lora, I can share the metadata in a bit, eating right now

[lyric zealot]

here's the metadata

[elfin geode]

hi anime!

https://discord.gg/tySdsD77?event=1130919343941758978

[gloomy scroll]

[elfin geode]

i come in peace

[gloomy scroll]

Jae you were supposed to type longer message

[elfin geode]

im in a good mood today

[lyric zealot]

first gen, already loving nubby's new model

[wet elm]

sorry for the late ping but do you think im using too many embeddings on my negative prompt?

(bad_prompt_version2, FastNegativeV2, AuroraNegative ,KHFB, negative_hand, ng_deepnegative_v1_75t, verybadimagenegative_v1.3), (painting by bad-artist-anime), (painting by bad-artist)

[gloomy scroll]

Does your images look good in your opinion?

[wet elm]

kinda, the lighting is good but very obviously AI, and i feel that even when i change models the style doesnt really change, and i wanted a more flat like anime screencap thing sometimes

[gloomy scroll]

If you are happy with your gens then doesn't that mean your negative is fine

If you want more flat. 3d in negative could help

[wet elm]

the thing is that im not that happy with my gens, i feel like im doing something wrong everytime cuz even when i try to replicate all the parameters it comes out somehow worse, but the main problem is that i cant really change the style and i think that my negatives may be the problem

maybe its an extension that im using idk

[gloomy scroll]

Try this
verybadimagenegative_v1.3, (low quality, worst quality, bad anatomy, bad hands, normal quality:1.5), artist name, (claws:1.3), (mini skirt:1.2), (leotard:1.3),

[supple raptor]

some negatives, and especially negative embeddings can significantly impact the style of gens. It would definitely be worth experimenting with some different ones.

[regal coyote]

use detail loras with -1 to -2 strength, should give a flat effect

[wet elm]

will try

[supple raptor]

what I did to figure out the negative embeddings that I like to use is run an XY grid with each of them individually to get an idea for what they're doing and then make some guesses to combinations that might work well and make another grid to compare those. Then just test and repeat until satisfied.

[wet elm]

damn thats a lot of patience lol, but i may try it, should i do just the embedding or the embedding + the rest of the negative to compare

[supple raptor]

I did embedding + the rest of the negative

[wet elm]

DAYUM

[unreal ridge]

I used to do the same

I had to tweak my neg for 2 hours to get something I like

[stable crane]

wow great work

[unreal ridge]

and it's still a lot of neg embeddings

[wet elm]

ok, what setting should i use in the XY plot? sorry im very new with XY

[wet elm]

could i see yours?

[stable crane]

ikr it's great 😄 nice one

[unreal ridge]

yos

verybadimagenegative_v1.3, AuroraNegative, KHFB, negative_hand-neg, letterboxed, borders, white border, black border, (extra fingers, deformed hands:1.2), (worst quality, low quality:1.4), (text, logo, signature, artist signature:1.2), monochrome, worst face,

[wet elm]

yoo ty, what did you do to get what you liked?

[unreal ridge]

tried adding/removing while testing the same seed of my favorite prompts, probably not the fastest method but it worked out in the end

[supple raptor]

Prompt S/R

[wet elm]

ty, also great work with that image damn

[supple raptor]

@wet elm this is the negative that I settled on:
negative_hand, (KHFB, AuroraNegative), (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)

[wet elm]

damn its waayyyy smaller then mine

OH now that i realized it may be my quality tags too

[regal coyote]

nice

[unreal ridge]

My model is struggling to not do horny

[bitter latch]

holy heck my inet is on point with steam today, just downloaded 8.9 gigs at 107.7MB/s 9faints)

[gloomy scroll]

If you want flat change that add_detail to -0.5

[stable crane]

nsfw, skimpy
😭

[stable crane]

KHFB adds a lot alone

[gloomy scroll]

[wet elm]

imma try that when i get home

[supple raptor]

They honestly don't do much. 90% of the time I forget to take them out even when I lewd prompt. Seems like they reduce the just reduce the accidental lewds.

[stable crane]

those can be the best ones

[gritty sundial]

generating image in stable diffusion wears out the video card?

[wet elm]

could i have the parameters for that damn

[stable crane]

every image you generate you lose 0.1s of video card life

jk but it probably has an impact

running a card hot for long periods of time vs. not, I feel like my card works harder on images than videogames

[regal coyote]

1girl, solo, horns, pointy ears, purple eyes, looking at viewer, short hair, breasts, bare shoulders, flower, plant, upper body, bangs, tattoo, blue hair, purple flower, parted lips, outdoors, medium breasts, hair between eyes, scenery, overgrown

(masterpiece, best quality, ultra-detailed, novel illustration, highres, absurdres, incredibly absurdres, animated, anime screencap, video, original, character request, borrowed design, borrowed character, novel illustration, official wallpaper, cover page,:1.2)

easynegative

[glacial bane]

Thank you very much as usual you're really generous

[wet elm]

Ty, and the model?

[lyric zealot]

pretty sure everyone here is willing to share their metadata

[glacial bane]

Tiktok Stable diffusion: WE AINT SHARING NONE

[supple raptor]

it does. gaming is a constant load on the core which generates more heat, genning is a relatively light load on the core with a heavier load on vram for maybe a few minutes at a time with time to cool down in between.

[glacial bane]

Still you're all kind in many ways

[stable crane]

the question is, is it worth it to even have a videocard if you're not gen'ing? ;P

[regal coyote]

counterfiet + some other model merge. i forget

[nova remnant]

I've been away playing in the LLM realm, is SDXL a good advancement for us anime people or should I sleep?

[glacial bane]

Anyways if i ahve integrated graphics should i add an external GPU for my laptop so it generates faster??

[gritty sundial]

generating image in stable diffusion wears out the video card?

[glacial bane]

Does that help though.

[glacial bane]

Vidoe card?

[neon linden]

GPU

[gritty sundial]

yes

[bitter latch]

Vid'DOH card

[glacial bane]

It wears it down? I mean sd does use GPU

My SD uses my CPU because its integrated graphics

[stable crane]

it might be bad if you had it generate batches of like 300

[neon linden]

I tried to ask that on support a couple of times but since they couldn't exactly pinpoint if there was an average difference so far I only used google colab to generate myself

[stable crane]

but click -> gen -> wait -> click should be totally fine

[glacial bane]

Just delete

@stable crane does external gpu for laptop solve the integrated graphics slow generation?

My cpu is louder then a ps4

Sounds like I'm mining Bitcoin

[steady grail]

hello guys, where do i find KHFB?

[bitter latch]

umm fail ..?

[glacial bane]

Indeed a fail

[unreal ridge]

https://huggingface.co/SweetLuna/Aurora/tree/main/AuroraEmbeddings

[bitter latch]

NO FAIL

[glacial bane]

I gyatt something in my eye

[gritty sundial]

how? google colab limited the use of stabble and you can't even use it anymore

[neon linden]

oh you can

you just have to re-start it after 1 - 5 images

it's a slow process but it's just pressing 5 buttons and I'm not in a rush

so I can slowly generate them

[bitter latch]

I make furry lewd's so .. no not using that 😄

[glacial bane]

Like a minute or so?

[bitter latch]

they disaprove

[neon linden]

oh no, the generation is fast

it is just that it can take up to 12 min to re-start

[unreal ridge]

You can. They just blocked words like
stable diffusion, sdwebui

[neon linden]

because of loading times of colab

[glacial bane]

Tames me like 40 minutes to generate one picture

With good results

[gritty sundial]

send me the link?

[glacial bane]

I have integrated graphics

[bitter latch]

laptop?

[glacial bane]

Indeed

[bitter latch]

thats better

[glacial bane]

That's why im asking if external gpu helps

[steady grail]

(nude:1.2) (nipples:1.2) (miniskirt:1.2) hrgiger-drmacabre, Style-Psycho-neg, 001glitch-core, deformityv6, image by bad-artist-anime, (Neg_Facelift768:0.6), easynegative, (bad-hands-5:1.5), (bad_prompt_v2:0.6), (KHFB, AuroraNegative) (worst quality:1.4), (low quality:1.4), (lowres:1.4), (monochrome:1.5), (greyscale:1.5), magazine, cover, comic, vignette, blur, out of focus, gaussian, ((text)), ((writing)), ((logo)), amateur drawing, blurry,

[stable crane]

yeah you need a good GPU to get better speed

[bitter latch]

I use a laptop too but i have internal 3060 Ti 4gig

[neon linden]

you can find it by just googling "automatic11111 colab"
it's the first link, by "TheLastBen"

[steady grail]

i usually use this as nega

adding the last nega

[bitter latch]

there's a laptop I want that has 4060 ti 8gig

[steady grail]

i just put all the nega embbedding i have and hope for a better result lol

[glacial bane]

External gpu works aint bouta spend that much on a laptop

[neon linden]

I might try to make a custom colab replacing words that reference stable diffusion to see if it works better

[glacial bane]

My laptop is already 400$

[steady grail]

for generating, a laptop is not the best choice. they tend to overheat

[bitter latch]

mine doesn't

got extra cooling pad under it

[glacial bane]

Indeed i make mine rest for 4 hours

[bitter latch]

with fans

[glacial bane]

Me who puts ice

[steady grail]

do you run a diagnostic program to see the pc temperature?

[unreal ridge]

Some people are changing it to microsoft excel lol

[glacial bane]

Jk i near the AC it dont heat up

[bitter latch]

yep and it barely gets upto 88c every so often

[neon linden]

lmaooo

microsoft excel

[steady grail]

don't let your card go over 100 C°

[bitter latch]

the vid card that is

[steady grail]

good

[neon linden]

if doing that removes the 1 -5 limit it will be well worth it

[glacial bane]

No but i can tell. Just by feeling its cpu

[bitter latch]

plus I keep my room cold

[glacial bane]

Will it go kaboom

[bitter latch]

not sure about cpu or other things. any good program I can use?

[steady grail]

no, just melt out

[glacial bane]

Oh

Ims sleep

[steady grail]

i use speedfan