🍥｜anime

[desert arch]

Oh god not rebecca

[opal dome]

how did you get this style?

[terse jacinth]

custom models recipe and prompt formulas

[gritty patrol]

accidentally made a vtuber

[stable crane]

I've gotta try out multidiffusion today

[stable crane]

hm how do you use multidiffusion

is it an img2img tool or txt2img?

[lyric zealot]

bloody

[primal wharf]

Fashion

[gritty patrol]

incredible!

not sure what exactly it tried to do, but it's cool

[stable crane]

trying same image w/ multidiffusion

I think it's slightly better, will keep experimenting

if anything the tiled vae to save vram is a game changer

[gritty patrol]

yes

[stable crane]

I'm looking through the github comments to see how ppl are using it

my checkpoint might be too low detail

as it sounds like pruned really suffers in this

[fiery sparrow]

||dddddddddds||

[desert arch]

Multidiff hits hard

[stable crane]

render takes forever

[primal wharf]

Where did you learn multidiffusion, got tutorial videos?

[stable crane]

right now I'm not sure I have learned it, I'm looking at github comments lol

[desert arch]

Yeah but at least it looks good

[stable crane]

someone suggested a CFG20 highres, masterpiece, best quality, ultra-detailed unity 8k wallpaper, extremely clear, very clear, ultra-clear img2img pass, but this has been shit for me

[primal wharf]

I'm too smol brain to understand written tutorials, guess I'll have to wait since there aren't any good video tutorials on it.

[stable crane]

for example

[gritty patrol]

this is anime channel, if you haven't noticed

[stable crane]

lol

[desert arch]

Everything is anime

[stable crane]

don't worry I'm changing checkpoints now, it's loading

[ocean reef]

wohoo I got rly nice eyes here 😄

[stable crane]

so

I think that Tiled VAE is just the thing to use in txt2img

[stable crane]

this is all 1 txt2img

1536x2112 without --medvram

[gritty patrol]

multidiffusion?

if so, HOW THE FUCK DO YOU NOT GET DUPLICATES

[gritty patrol]

zamn

[stable crane]

base is still 512x764, then I turn on tiled vae, and I set upscale by 3x or 4x

[clear jewel]

Use Unprompt to design creative images

[astral aspen]

no sharingan eyes?^^

but nice image

[static flax]

guys for colors why it look like its gray

[stable crane]

missing VAE?

[static flax]

yeah

dont know best settings

[stable crane]

search this channel for "VAE"

last night we had some links and descriptions on how to enable it

[static flax]

i do have it

like the tile one

i traid those

[stable crane]

yeah but are you using a vae at all?

[acoustic agate]

you need the kl-f8-anime2 vae in your stable diffusion VAE folder

[gritty patrol]

is your pastel look because of your vae or your model?

[static flax]

i disable it cause of random results thats why am asking

[ocean reef]

model

[stable crane]

use one, or suffer lol

[gritty patrol]

which model?

[static flax]

:#

[gritty patrol]

anythingv3 vae?

[acoustic agate]

Why the v3 vae

[stable crane]

it's way better

[stable crane]

kl-f8-anime2 has better colors but often blurry edges

[acoustic agate]

[ocean reef]

disillusionmix

[acoustic agate]

I'll take better colors

[stable crane]

tradeoffs

[static flax]

this ?

[ocean reef]

yup, its good for anime

[zinc flax]

https://tenor.com/view/preference-glarses-mx-brown-cherry-mx-brown-keyboard-gif-21044520

Try which one suits you best

[static flax]

ok but i cant see SD VAE droplist you are talking about

[zinc flax]

It should be on the settings

[stable crane]

can use offset lora to reinject more color

[static flax]

found it thx

[stable crane]

search for "Quick settings" in this channel and you can add it to top of your UI

[gritty patrol]

no pruned/fp16 😭

[stable crane]

kl-f8-anime2 is blurry

[static flax]

thx

[trim current]

This from your mech lora?

[stable crane]

iirc it might be

[trim current]

Looks neat, got a few of yours downloaded but havent messed with yet

[static flax]

huge diff

[stable crane]

you can combine a lot of the mech wings, large mecha arm, spider girl

in different ratios for different effects

[static flax]

another question why it is not taking all my GPU power ?

[trim current]

That is awesome to know, can already think of a few mechs I'd like to ride lol

[gritty patrol]

i want a vae that doesnt oversaturate the colors or burn the image like anime2

suggestions?

i honestly dont know where to find vaes

[stable crane]

I don't think a lot are created

[tropic wraith]

FAQ: Is Stable Diffusion open sourced?

Yes! Even though we offer competitive rates for the computational cost of generating images on https://beta.dreamstudio.ai, you are also able to run the various open sourced versions of Stable Diffusion using your own hardware without any additional costs! Take a look at #1080946152318443610 to see how to get started!

[magic wave]

It uses your GPU's VRAM rather than raw GPU power :o

[static flax]

this is what i mean

i have 12GB of VRAM

[magic wave]

Training a LoRA atm, tis how it looks, its barely being used, but VRAM is full

[magic wave]

Well, I don't think everything that you do is as intensive

[static flax]

i mean the speed

[stable crane]

I think the "3D" graph there will always look low

[magic wave]

I think just some things will take big amounts of vram

[stable crane]

you can play games and still see 5% on 3D graph

[static flax]

whay not like this

[magic wave]

Well, depends on the res, also you got xformers on?

Or pytorch 2.0

One of those two, they're good to speed up generation

[static flax]

i thi

i think xformers is off

[stable crane]

yeah high res will be much slower iterations

[magic wave]

You can edit your webui-user.bat file and add --xformers to use xformers

[static flax]

k will try it

[magic wave]

But also, if you're generating high res pics, its better to generate at a lower res, and use hires fix to upscale it, its usually more stable

For example this is 512x768, upscaled by 2 at 0.5 denoise through hires fix

[stable crane]

I'm up to 8.2s/it 😢

my fan is going nuts

[magic wave]

Ooof, why? :o

[stable crane]

mega resolution

[static flax]

okaayy

[magic wave]

Oh, well, then there is your issue xD

[stable crane]

2816x2048

[astral aspen]

zombielove:

[magic wave]

Thats pretty cool :o

[karmic swift]

did she beat her meat

[astral aspen]

i think so ^^

[karmic swift]

just for that am going to check out your gallery

[somber nova]

is anyone aware of pictory?

[gritty patrol]

i have a 4gb model that i want to make into 2gb, how do i do this?

[somber nova]

do you perhaps know something similar to it but free?

[astral aspen]

oh my gallery is not updated at the moment^^

but thx anyways 🙂

[gritty patrol]

basically i want to prune(?) it

[stable crane]

this extension is godly, finally can just generate any resolution. I actually need to edit the webui to let me do higher resolutions at this point

[karmic swift]

hey i got a question

am using v5anything what is the best sampling method?

[ocean reef]

it depends. I always try at least 5 to check the one I like best

[stable nimbus]

even if you set the same seed the noises will be completely different so kinda hard to compare

[acoustic agate]

I only use the bh2 unipc variant now. It is as fast as eular A and gives much better / more-consistent results. Only downside is I need more steps, around 50.

'Better' being a hugely subjective word here.

[stable crane]

a bit scrunched

[stable crane]

interesting, never heard of this one, sounds like I gotta git pull

[acoustic agate]

Nope

Do a new install, I've seen git pull break current installs a number of times

[stable crane]

oh damn

[acoustic agate]

Don't want you potentially destroying your functioning instance ❤️

[stable crane]

shame that it got so unreliable

[acoustic agate]

It was because of one specific change

Idr what it was, more a question for @supple raptor

[stable crane]

I kinda wanna make a UI for stable diffusion to really specify the latent coupling

it's painful to use the numeric notation

I'm pretty outdated too in my current install, it wants me to update some libs on startup

[gritty patrol]

i just use DPM++ 2M Karras

always

[acoustic agate]

Unipc is twice as fast, and very similar or almost identical to one of the dpm/karras variants. Worth a shot imo.

In the sd webui settings you can toggle between unipc variants bh1 and bh2. I haven't done testing between the two.

[gritty patrol]

is unipc non existant in img2img?

[acoustic agate]

Correct

[gritty patrol]

then in img2img i shall use dpm++ 2m karras?

[acoustic agate]

I don't see why not, though I usually just use eulerA in img-img.

[gritty patrol]

why euler a tho?

i've seen many people recommending it

[acoustic agate]

Speed

[gritty patrol]

but in exchange you need more steps right?

[acoustic agate]

25 is good on euler

[worthy coral]

Hello, new here starting to learn about Stable Difussion. How can I config a default VAE? I downloaded some but they're on ckpt extension 😓

[gritty patrol]

put on models/VAE folder

[worthy coral]

Ah heh 🙂

[worthy coral]

Thank you!

[gritty patrol]

unless they're >500MB

then there's something wrong with your vaes

[acoustic agate]

[gritty patrol]

🤏

smol

[zinc flax]

Smol

[terse jacinth]

DPM ++ SDE / Euler A / Euler

[rich tartan]

where adaptive

it is so underrated

[ocean reef]

It's time for boys!

[gritty patrol]

i forgot the name but thanks google

it's karin right?

[zinc flax]

LMAO

[worthy coral]

I'm not sure why they have that face

[zinc flax]

It's Karin

[gritty patrol]

i imagined

[zinc flax]

Can finally make her small variant with lora
AI too powerful

[terse jacinth]

[gritty patrol]

i can't make my lora, i am simply unable to can

[gritty patrol]

STOP MAKING MASCULINE MEN!

[ocean reef]

NO!

[rich tartan]

lmao

[ocean reef]

muscles are life!

[acoustic agate]

Your next task is muscular smol

[ocean reef]

I prefer big tho 😏

[acoustic agate]

that's just slim

[ocean reef]

pretty guy with babies, good? 😄

[acoustic agate]

Gives off Magi vibes.

Blue hair guy is Ugo

https://tenor.com/view/paimon-ugo-magi-labyrinth-ofmagic-blush-gif-19355183

[gritty patrol]

make someone with a 16 pack and i'll let you be

[acoustic agate]

Only 16? 32 is more realistic.

[ocean reef]

🤣

[ocean reef]

this one is better 😄

[acoustic agate]

anyone got a prompt I can try out in my new merge

[gloomy scroll]

1girl, (🌊🌴:1.3)

[ocean reef]

portrait of a young very beautiful cute tribal woman, in a post apocalyptic city overgrown with lush vegetation, dark, gritty, intricate, volumetric lighting, volumetric atmosphere, concept art, cover illustration, octane render, 8k, dynamic pose

[acoustic agate]

[gloomy scroll]

Oh no this doesn't sound good

[ocean reef]

I cant get a baby with my currently loaded model -.-

[gloomy scroll]

So your model likes ?

[acoustic agate]

to clarify, nothing was nude, just suggestive and a bit too much skin

[worthy coral]

It's getting worse VAE's were supposed to fix these faces, any idea why they come out so bad? Not sure if it's a prompt issue or a VAE Issue.

It happens with every models, so I don't think the models are failing.
The current vae is: kl-f8-anime2.ckpt

Alredy changed the VAE option in Settings > Stable Difussion.

No idea how to fix it...

[acoustic agate]

you don't have restore faces enabled, do you?

[worthy coral]

No, I don't

[acoustic agate]

what model

Here's nerfgun's good face embedding. Put it in your positive prompt and it should help a ton.

[gloomy scroll]

Why I didn't know such thing existed

[worthy coral]

Everyone; neverendingDreamNED, RealdosMix, PerfectWorld_v2Baked... have been tried in each one of them.

[acoustic agate]

what's your negative prompt

[ocean reef]

😏

[acoustic agate]

For those who need them, here's all of the negative embeddings I've collected.

[worthy coral]

(painting by bad-artist-anime:0.9), (painting by bad-artist:0.9), watermark, text, error, blurry, jpeg artifacts, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, artist name, (worst quality, low quality:1.4), bad anatomy

[acoustic agate]

do you have the bad artist embedding?

[worthy coral]

Not sure, honestly, I don't know what I'm doing... : s

[acoustic agate]

I just sent a .rar with my negative embeddings. I recommend you unpack that and put the embeddings into your embeddings folder. The folder is in your root stable-diffusion folder.

Once the embeds are in there, they will automatically work when prompted. No need for a UI restart.

[worthy coral]

I'm getting some prompts from Civitai and try learning while I read reddit, forums, git and that kind of things

Can I write you privately?

[astral aspen]

add a few freckles, skin is to clear

[acoustic agate]

sure

[ocean reef]

more boys!

[gritty patrol]

looks cool

[astral aspen]

make a pirate crew in this artstyle like this:

[compact warren]

Colurfol

[ocean reef]

There you are!

[astral aspen]

not bad

[ocean reef]

I got solo captain too 😄

[lucid rapids]

anyone know how to enable Loras on my webgui?

I dont even see the folder for it.

[acoustic agate]

red button under the big orange generate button

[acoustic agate]

[acoustic agate]

are you on the automatic1111 webugui

[lucid rapids]

where to check

[stone onyx]

It's the button under the save icon @lucid rapids

I think you're on an old version

[lucid rapids]

oh ok

[ocean reef]

under save button is placement of saved styles

[acoustic agate]

^

[ocean reef]

isn't it last one?

dunno how old UI looked

[acoustic agate]

old UI didn't have a button for it

[stone onyx]

hmmm weird, I'm pretty sure this icon pulled the lora settings for me

[lucid rapids]

weird i thought my webgui should be updated for me

[acoustic agate]

did you install this webui https://github.com/AUTOMATIC1111/stable-diffusion-webui

[lucid rapids]

idk i got all this setup months ago

[ocean reef]

so it's old one for sure 😄

[acoustic agate]

[ocean reef]

back to my handsome guys!

[lucid rapids]

yea i have that one,i just need to update

[acoustic agate]

My honest suggestion is to delete you current install and reinstall it from scratch with the above link. Lotta updates since then, high probability your git pull breaks the install. Worth a shot I guess :V

[lucid rapids]

yea something happened

[ocean reef]

https://youtu.be/3cvP7yJotUM i recommend this one

with auto updates 😄

[stable crane]

lol

[karmic swift]

love it

[worthy coral]

I downloaded a LORA file in Civitai, how do I use it? ><

[karmic swift]

how i love you brother, my gen are getting way better now

[acoustic agate]

Noice

[karmic swift]

girls behold

[stone onyx]

put it here stable-diffusion-webui\models\Lora

[acoustic agate]

And surely you mean the AI's gens are getting way better. Its the one doing all the work, you're just asking it for a picture after all

[karmic swift]

am doing some work guiding it \

[acoustic agate]

Tell that to the legal system

[karmic swift]

i just got legalized oh may gaa

[karmic swift]

boner ???? huuuuh

[acoustic agate]

https://tenor.com/view/osaka-azumanga-daioh-angry-oh-my-god-oh-my-gah-gif-26425247

[karmic swift]

cursed

[astral aspen]

cemetery girl series:

[ocean reef]

motorbike is weird tho xD

[somber nova]

does anyone here know a good text to voice AI that's free?

[ocean reef]

https://micmonster.com/#get_app_now I like this one

[somber nova]

i'll try that one

is it free?

[ocean reef]

yup

[somber nova]

thanks 🙂

[opal dome]

what im trying to do is to generate the character and machinery separately, then photoshop them together

[ocean reef]

That would work for sure, but I'm too lazy to do that 🤣

[worthy coral]

Have been trying to replicate this pic using the provided generation data, but for some reason mine is so much different, the faces also looks different.

The expected result: https://civitai.com/images/135522?modelId=5477&postId=93755&id=5477&slug=lucy-cyberpunk-edgerunners-lora

[ocean reef]

Do you have Lora Lucy installed?

[worthy coral]

Yep

[ocean reef]

what about model?

[worthy coral]

The same, I thiink : https://huggingface.co/lysdowie/dalcefoV3Painting/blob/main/dalcefoV3Painting_dalcefoV3Painting.safetensors

[ocean reef]

Then it might be hires differnce

btw I recommend to use lower resolution and use hires instead

It does magic

[opal dome]

if you want to replicate, everything has to be the same: seed, prompts, embeddings, model, sampling method, resolution, vae, CFG, clip skip, eta, etc., even then, the author might had done inpainting, which we dont the data on

[worthy coral]

Yeah... maybe it's the VAE

[ocean reef]

oh yeah, vae, forgot about it 😄

[worthy coral]

I should find some VAEs to use.

Only got kl-f8-anime vaes

Any site to search VAEs?

[ocean reef]

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main most used one

[opal dome]

google, they are usually on huggingface

[worthy coral]

Many thanks ❤️

[worthy coral]

Where can I find Clip skip: 2, Mask blur, Model hash and Denoising strength in Text2Img generation?

[ocean reef]

Denoising strength shows up when you select hires. fix

[worthy coral]

Didn't see hires yet ><

Where are they?

[ocean reef]

under sampling model, 3rd one

as for clip skip, go to settings -> User Interface tab and add "CLIP_stop_at_last_layers" in the Quicksettings list

then apply settings and reload UI

you will have it on the top of UI then

[worthy coral]

😓 Can't find... heh : S

[worthy coral]

AHHH

❤️

Thank you so much 🙂

[astral aspen]

what does clip skip 2 excactly?

see it sometimes as recommendation but dont know why

[opal dome]

generalizes prompt, 1girl => female, something like that

[astral aspen]

sorry im not a native english speaker so some stuff is hard to get for me

[ocean reef]

https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/5674

[astral aspen]

ahh ok i understand the concept, thanks nice to know

how can i test that?

[ocean reef]

I dot get it full either xD

The best way to test... is just change parameter on all same setting (seed especially)

[astral aspen]

so maybe its faster generation with that cause it will not care about 6 deepness level? at clip2? (12/2)

[opal dome]

just level 12, i think

[opal dome]

it gives better generations because if it follows the prompt too closely, results are "bad"

[astral aspen]

yeah but if 1.5 has 12 standard deep levels and i set to clip 2 means that 2 depth instead of one so 12 overall divided with two 12/2 are 6 are am i wrong? so deep is 50% with clip 2

but it also means that the prompts has to be smalelr or?

cause of 6 level instead of 12

[opal dome]

for anime stuff, mostly use common danbooru tags as prompts

[astral aspen]

yes i only will understand, i have after 6 prompt items a lower deep so its less accurate but stays in the prompt even its over 6 but its any "cow" instead of the especially descirbed cow) (as example a cow XD)

very interesting the link, ive learned much

and clip 2

this information fits perfect in my mind cause of the rpg-4 modell which works with several predifened layer, its quite fun and very accurate-so now i know the background

the most modells dont work this way but in the background theres is a limited logic for that, i diddnt even know the excact numbers, now i know

[supple raptor]

The values of clip skip represent the layer that the clip model will stop at but in reverse. There are 12 total layers. 0 means all 12 layers will be used, 1 means 11 of the layers will be used, 2 means 10 of the layers will be used.

[gloomy scroll]

At least she has more than enough fingers to grab things

[ocean reef]

[astral aspen]

cementry girl series v2

[opal dome]

kiara?

[white zenith]

ye

[white zenith]

tinkering with the easynegative embedding, interesting as a style embedding

[white zenith]

melty cfg scale

[valid sun]

Cat

#1080946152318443610

Cat #1072220168534642768

[final vault]

everytime I came back on discord, I see anime channel on a new location

[primal wharf]

Beautiful eyes

[elfin geode]

see you tomorrow when we turn anime into an announcement channel

[cyan burrow]

when you turning off topic chat into anime/pol/ ?

we need to know the truth @elfin geode and spread it through memes and pointless 24/7 bickering that brings zero solutions to nonexisting problems

[primal wharf]

Balancing 2 burgers on 1 leg

[opal dome]

japanese goblin?

[primal wharf]

https://tenor.com/view/fubuki-confused-hololive-gif-18015941

[elfin geode]

im sold, sounds like a plan

[primal wharf]

Finally, been trying to get burger and fries on the same tray, but separated as individual items.

[compact warren]

[eager wharf]

This one gives me serious APB vibes
(Images are spoilered because its not ai)

[keen hollow]

I like this one

[smoky viper]

Hjbh

[sharp bluff]

theres a fork of the image browser why did no one tell me this

[south sable]

Try some hyper networks, loras, or embeds you like to hone the anime look. It can take some tweaking to get it just right.

[brittle jay]

trying to install oogabooga through the one click installer but I get this, any help?

[karmic swift]

how do you prioritize the background more than the character @stable crane

[stable crane]

counterfeit v2 really likes backgrounds

there's only a few background words that vaguely influence the theme

fence is one of them lol

[karmic swift]

there really is a lot to learn goodness gracious

[stable crane]

same model, but no background tokens, still you see it gave it a background of sorts

[karmic swift]

upscaled ?

[stable crane]

tiled vae

it's really good lol

[brittle jay]

I sometimes use atmospheric perspective or landscape or full body or whatever

just use danbooru tags

[karmic swift]

perfect scale 10/10

[gritty patrol]

how.

[flat moth]

🤷‍♂️ used a prompt I whipped up earlier today and was tyring out tiled vae/multidiffusion

[gritty patrol]

congratulations, you have just made one of the best ai artworks out there

[karmic swift]

facts bro

[karmic swift]

and am still struggling with portraits

[tepid swift]

#1072229332577169498 full profile

[white zenith]

kinda like the lower sat look

[flat moth]

that no vae stare, I agree though it can look nice sometimes

[white zenith]

it is actually using anything v3 vae, not quite as flat as no vae in the example

probably a bit too flat though, kind of just want to get a bit away from the glossy/over polished "ai" look

[flat moth]

if you're using aom2/3 try based64/65 mix

[sage mountain]

any Kaiji fans? (trained lora)

[cyan burrow]

only saw one of the real action movies

thinking of watching the anime

[wind girder]

Hello sunshines 🙂 how is the Anime team today ?

[cyan burrow]

Horrible

you?

[wind girder]

quite ok, even if up at like 6AM for no reasons

[sharp bluff]

publishing new lora soon yay (blood)

[desert arch]

Good afternoon

[terse jacinth]

WAAAAA

[terse jacinth]

3AM and ready for a new day of waifus and colors

[wind girder]

I just managed to plug a local install of a GPT model into comfyUI 😄

this is going to be a good day

[terse jacinth]

noice

local is best

[mild tree]

Morning

[desert arch]

Morning

[mild tree]

Fucking insane

[desert arch]

That one's another random prompt

[mild tree]

Ye

But like

Its insanely good

[desert arch]

It's the model not me 💀

[mild tree]

💀

[mild tree]

We are getting more and more useless

[desert arch]

They still need to fix the hands though

[mild tree]

Yep

But with hand models in the near future it won't be hard

Like it's just about making the ai understand what hands are and what they do

[mild tree]

Regret, Can you try a cyberpunk city wallpaper type gen pls

Since your model slightly reminds me of ink punk model

[terse jacinth]

waifu ?
waifu

[desert arch]

Tried the cyberpunk but not the city yet, will do

[mild tree]

Thanks

[desert arch]

Just the city?

Or a person as well

[mild tree]

City

[desert arch]

Something like this? not upscaled yet

[terse jacinth]

J I S O O

[astral aspen]

cementry girl plays a violin

ok enough purple haired cementry girls for today XD

[mild tree]

Oh ma gawd

I died

[desert arch]

@mild tree Upscaled a bit, resolution's not really for wallpaper i think

[mild tree]

Tysm

Can i use this in a video

As cover art

[desert arch]

Sure, i don't mind

[mild tree]

yay

Lets go

[azure oracle]

rose

[terse jacinth]

wifu

[grand geyser]

@terse jacinthYES

[final vault]

🫂

[terse jacinth]

my batches, lol

[grand geyser]

dam looks so good so colorful

[terse jacinth]

[grand geyser]

om all max

almost

omg

[terse jacinth]

its more efficient

batch size 4 divide my render speed by 2

batch size 9 divides it again by 2

so it cost me the price of 4 to produce 9

[astral aspen]

i need a better gpu but for that i need a better cpu for that i need a new pc for that the 7950x3d has to be avaiable but sadly he sint buyable in my prefered shop 😦 amd dont deliver at the moment and no one knows when it gets better

then i will pair it with an 4080

[grand geyser]

looks insnae i ll try that

[wintry marten]

Anim

[grand geyser]

how much time took you to gen floyd

[desert arch]

My gpu isn't even a real gpu 💀

[terse jacinth]

around 18s per batch size 9

[grand geyser]

@astral aspenget outside of oyur city :[

you have no choice

[astral aspen]

i need a new one, its faster thats save me time or spend me time to make more images XD

[grand geyser]

what gpu floyd?

[grand geyser]

insnae

im generating with rtx 2060

[astral aspen]

have an old 1070 laptop @ 8gb here, but "she" does her job pretty well , she is an old girl but supports me well Xd

[terse jacinth]

you can probably do batch size 2 or 4

[grand geyser]

yeah

[desert arch]

Laptop with an igpu here

[grand geyser]

i saw someone with rtx 4090 doing max sampling max batches 40it

atleast we can generate :d

[desert arch]

Yep

[terse jacinth]

oh i can do larger batches

just not efficient

batches squared are better

[grand geyser]

yeah takes alot of time

i generated 100 imges 1 batch on 4 3it

[terse jacinth]

1 image goes at ~6 it/s
4 images at ~3 it/s
9 images at ~1.5it/s

you have ot think you are using power equations

[grand geyser]

but on sampling method euler

xd

[terse jacinth]

512x768
512x768(2) = 1024x1536
512x768(3) = 1536x2304

so if you render a batch 4 or 9 it will fall on those resos

and you don't waste any power on black pixels

just basic linear algebra maths

512x512 (this used to be over 15it/s in november, not sure what happenned)

[grand geyser]

wow

insane

kawaiii

[terse jacinth]

so yeah doing batches of 9 is effectively saving me time and electricity

best would be 4x4 or 9x9 imo

[grand geyser]

the thing is it takes alot of time atleast for me

[terse jacinth]

do batch size 4

and batch count 1

for your gpu this should work well enough

[grand geyser]

average 1.28

[terse jacinth]

that's good

[terse jacinth]

20 series aren't that fast to be honest

[grand geyser]

yeah

[terse jacinth]

they use older RT cores

[grand geyser]

thought to get 7800xt

but i heard amd not suppport

[terse jacinth]

30xx and 40xx have new tensor and RT cores that are much better at those fancy AI maths

[mild tree]

[grand geyser]

i wish i could order in us and ship it

its less xd

[acoustic agate]

🥚

[astral aspen]

nice one

[terse jacinth]

https://tenor.com/view/hello-gif-25128806

[grand geyser]

tried to generatte rias gremory

[open ibex]

I'm new using Stable Diffusion. I've been testing Mistoon_Anime Civitai Model.

[desert arch]

Bit of blood

[ocean reef]

Happy Easter for those who celebrate it

[astral aspen]

happy eastern @ all

[grand geyser]

omg looks so nice

[terse jacinth]

clyde stop annoying me

[acoustic agate]

idk if underwear counts as sfw

[terse jacinth]

latex-ish armor doesn't pass but bra does

clyde is biased as usual

[grand geyser]

noo i cant send explicit content

its not sfw

oh bra rip

[terse jacinth]

boat-chan

[astral aspen]

pls be aware that we have eastern so its bunny ear time instead of cat ears

XD

[terse jacinth]

bunny girl ?

got it

[desert arch]

Bnunuy

[astral aspen]

haaah much better, so makes eastern sense 2023

[terse jacinth]

bunny girl

[desert arch]

Eat the borger

[astral aspen]

bunny ears,girls,food - perfect picture!

[wind girder]

prompting has come a long way

[desert arch]

Indeed

[terse jacinth]

you wrote it wrong

[wind girder]

the secret is hidden here

[terse jacinth]

best 1girl

[wind girder]

let's try best 1girl

[desert arch]

Just put 1boy, thomas shelby, instant rizz

[wind girder]

WOW

let's not

[terse jacinth]

[wind girder]

it didn't exactly get it

I got oil paintings of girls, and not a single cloth

[terse jacinth]

sounds about right, if you have nsfw in your models

[wind girder]

it made this prompt

Best, masterpiece, hyperrealistic, Painting, close up, girl, radiant, beauty, perfect skin, natural light, oil painting, gouache, classic, warm colors, contrast, Kodachrome, Scott Campbell, Alex Ross, Frank Frazetta, John Romita Jr.,

the GPT model I use don't get the 1girl token

[terse jacinth]

ah, yeah maybe not then

[wind girder]

best 1girl waifu

fun thing is, since it generates a new prompt each time, it's completly different styles each seed

[terse jacinth]

do bigger batches size

if it works like dynamic prompt

[desert arch]

Not bad

[wind girder]

nope, I run the model once only per seed. it's long to generate the prompts itself through GPT.
I can add dynamic prompts node though

[wind girder]

yep but I'm not in automatic. I need to simulate dynamic prompt through 20 nodes, but that can work

preparing a big prompt machine :p

[terse jacinth]

i see, keep me informed

prompting is something i'm not that good at

[wind girder]

well, I think I'm going to lower it on batch size 1 in fact, and more focus on highres fix, and dynamic post processing

[terse jacinth]

you know it's POTG when you see this ^

[terse jacinth]

https://tenor.com/view/miia-monster-musume-anime-blush-sweating-gif-16442752

[supple raptor]

2Bunny

[astral aspen]

nice one

[grand geyser]

fire

[astral aspen]

neo sci fi ship fight scenes

[zinc flax]

Really like how this accidental model merge turned out lol

[pseudo pendant]

Blue sky

[terse jacinth]

this is trippy

[wispy canopy]

uhoh... i started trying to improve my model... not sure if that's a good or bad thing. it's kinda fun tho, trying to fix the things it's bad at, whilst not trying to lose the things it's good at

[wispy canopy]

i got my waifu schematics back tho ^^ that wasn't possible in the original fluke

[terse jacinth]

i need to get in on that stuff soon

checkpoint or lora ?

[wispy canopy]

it's a whole checkpoint

i had created a fluke hentai model which is actually kinda fucking good

but i lost a bit too much of my art style things, so now i'm trying to re-introduce the non-anime concepts, but i also want it capable of doing hentai yet 🙂

[acoustic agate]

@wispy canopy i used that model you told me about in my TrexMixv3 merge

[wispy canopy]

oh, my model has already been abused for another merge? 😄

[acoustic agate]

Aye

[wispy canopy]

nice ^^

my model got 500 downloads in civitai, it's nothing much, but still 500 downloads and even 10 followers?! 😮

[wispy canopy]

can't do it fulltime now tho -> am cooking for tomorrow

@acoustic agate i'm honored my model is good enough to be mixed into yours ❤️ where can i find yours, wanna test it now that i know it exists!

[acoustic agate]

The three versions haven't been uploaded yet. I'm working on editing the collab video.

[wispy canopy]

that's more improtant 🙂

[acoustic agate]

But right now, sleeeeep

[wispy canopy]

looking forward to that one too 🙂

... dude

you didn't stay up all this time?

please say no

[acoustic agate]

I esited fr 9 hrs strait all night

[wispy canopy]

esisted fr?

[acoustic agate]

#collab chat to see progress

Edited

[wispy canopy]

aight, going to knead meatballs, can't type anymore xD

[acoustic agate]

Knead

Kneed

Knee

Kneecap

[wispy canopy]

uhoh.. .where is your thought-train going? XD

[acoustic agate]

North

[wispy canopy]

and sleep deprivation are some pretty powerfull drugs

you'd best take a rest trexdel ❤️

[acoustic agate]

But the camels

They need milk and shade

[acoustic agate]

The sun is too harsh

[cyan burrow]

soombuck

[acoustic agate]

Soom

[acoustic agate]

01146

[nova remnant]

hi, is there a free trial for stable difusion?

[wispy canopy]

you mean: "there is free stable diffusion"

stable diffusion is free open source software

you just have to have a GPU stron genough to run it

[cyan burrow]

i think I need to get a big 2.5" storage

[nova remnant]

ok, and are there some servers like midjourney to get acces too?

[wispy canopy]

that. i wouldn't know.

pleased with these results, i think i might be on the right track 🙂

[steel trail]

these are nice

[wispy canopy]

the quality of the nsfw images did degrade a bit tho

a gouache by Vhils , anime (lineart:0.4), solar themed, epic splash art , gothic art, gigachad This is such a nice prompt to test stuff with

@ocean reef i'm reworkign my model. it seems the husbando's are getthing thirstier!

[cyan burrow]

apparently my alucard/hellsing lora decided that it wanted to turn Alucard into a woman and give Seras his clothes

[gritty patrol]

oh i thought you guys were making bunny girls for easter

[wispy canopy]

perhaps later. right now i need to improve my model XD

[desert arch]

Good evening

Did make some bunny girls earlier

[wispy canopy]

Not sure if losing all hentai capabilities is worth it over losing this ❤️

[wispy canopy]

cursed.

or blessed

depending on perspective

[steel trail]

Mario has become a bunny girl princess

O.o

[ocean reef]

looks good but muscles are kinda too defined

[wispy canopy]

i'm still finetuning 🙂

[ocean reef]

It goes in a right direction 😄

[wispy canopy]

I noticed the old model lost the capabilities to do this kind of shit. so i'm re-introducing these concepts (oh and bonus husbando and waifu) (all same base model, no loras or embeddings used)

[desert arch]

Massive

[nova remnant]

One of my latest AI artworks with some manual fixing.

Thanks for positive reactions ❤️

[wispy canopy]

I don't fix shit. I just dump random stuff in my prompts and enjoy pictures 😄

oh wait. this is supposedly my new and improved hentai model

[nova remnant]

I am a bit of a perfectionist... So I always try to fix and improve

[karmic swift]

buildings etchi

[nova remnant]

Those are very good looking

[wispy canopy]

nah, this one can do full hardcore xD

[wispy canopy]

but i want it to be a generic model, with full hentai/hardcore capabilities, after my first model was accidental pretty good, but lost some stuff

[karmic swift]

can u show some husbands

[nova remnant]

I also would like to start training my own models one day. But I need to get myself to learn how to do it. Should not be too hard

[wispy canopy]

i just made a remix XD

[nova remnant]

Oh that also works

[wispy canopy]

i was drunk and dissapointed by a generic model's capabilities for genitals

[nova remnant]

Merging models is very popular

[wispy canopy]

and i did some dumb shit, made an awesome hentai model

now i want to improve it xD

[nova remnant]

That's the model you are talking about?

[wispy canopy]

yeah, hentai model

[nova remnant]

It has a very nice style to it

👍

[karmic swift]

extra zesty no ?

[wispy canopy]

very.

my baseline requirement -> it should be able to do futa out of the box -> i know it is good enough with anatomy to do most things if that is the case

[karmic swift]

welp i was thinking manly , suit, smoking

[nova remnant]

is this chat SFW only?

[wispy canopy]

very sfw

[karmic swift]

yup

[nova remnant]

oh ok

[wispy canopy]

so talking is... perhaps the border

but posting nsfw is a 110% no-no

[nova remnant]

is Character in underwear fine?

[wispy canopy]

@karmic swift i'm throwing my old favorite prompts vs husbandos and sees how it performs

[wispy canopy]

if it's ecchi in intent, most certainly not

[nova remnant]

Oh I see

[wispy canopy]

if anything is ecchi/nsfw/lewd in intent, just don't

[nova remnant]

Is there a channel here for that or no?

[karmic swift]

4th one is best

[wispy canopy]

nah, this is a pure sfw discord server

[nova remnant]

Oh I see

that's cool

I will just post my sfw artworks then

[wispy canopy]

same here

my tesnting prompt has some serious hardcore images in between, but that's to see if it can do generic concepts and hardcore both xD too bad i can't show it off here

[nova remnant]

That was my first artwork I posted ever

Very happy with how it looks

😄

[wispy canopy]

you should be, it's amazing 🙂

[nova remnant]

Seriously?

Thanks

[desert arch]

Yeah pretty damn good for first time

[nova remnant]

I am doing my own OC characters

and writing lore for them

[stone onyx]

1st artwork? Damn that looks really good

[nova remnant]

Not really a 1st artwork...

[stone onyx]

I remember mine was beyond tragic

[nova remnant]

First one I published online

It roughly took me 2 weeks of practice to get to this stage

[karmic swift]

same

[wispy canopy]

behind tragic?

[karmic swift]

beyond*

[wispy canopy]

well, i just learned more ways to be random xD

[stone onyx]

[wispy canopy]

THESE HUSBANDOS 😮

[stone onyx]

I type like a boomer on my phone, people make fun of me all the time

[desert arch]

Exactly

[stone onyx]

😭

[nova remnant]

it looks very good

[karmic swift]

imma go try to make some rn

any good loras, eface

?

[wispy canopy]

none. i kinda never use them...

[stone onyx]

Try wlop

[karmic swift]

i use v5 anything model btw

[wispy canopy]

OH NO

[stone onyx]

I've never used loras with the anything models

[wispy canopy]

my sari armor disintegrated 😦

[nova remnant]

Also made some emotes for my discord server

[karmic swift]

ayoooooooooooooooooooooooooooo

these are cute overload bro

[stone onyx]

That's actually pretty creative

[nova remnant]

Thanks

I try to be creative with AI

I have always wanted to be an artist

but I lacked skills to actually make art

AI art allows me all of creative freedom I ever wanted

It has been a wonderful journey for me so far

[karmic swift]

villain arc story ?

[nova remnant]

😄

[desert arch]

Real art is like, really hard

[stone onyx]

That was pretty adorable 🥹

[nova remnant]

AI art is also not that easy

[desert arch]

Takes years to master

[nova remnant]

I think it requires a good understanding of some stuff behind real arts

[somber nova]

anyone know a txt to voice AI thats free?

[nova remnant]

I usually fine tune my artwork for 1-2 hours

to get all details to be perfect

[wispy canopy]

i tried drawing some 20 odd years ago or smth

[nova remnant]

Also thank you all for positive reactions 😭

I was afraid to post anything here for a few days

[stone onyx]

Had the same, but people are super nice

[nova remnant]

True...

[wispy canopy]

one of my first SD outputs back on sd1.3?

[nova remnant]

It's still very cool

[desert arch]

Used novel before SD

[wispy canopy]

well, i suppose my models and prompting did improve

[karmic swift]

the glow up

[wispy canopy]

lemme redo that first image

i sppose... elf, forest, green hair, red robes

aight, lets cook something up!

[karmic swift]

go ahead

let him cook

[terse jacinth]

https://tenor.com/view/cook-breaking-bad-hoodie-gif-13536614

[wispy canopy]

well, having a decent model is half the work, i suppose

[desert arch]

I guess

[wispy canopy]

nah, prompting skill do help too 🙂

[desert arch]

Nowadays just put 1girl and you get a good art

[nova remnant]

it has awesome chibi style now