🍥｜anime

[oak spoke]

[gloomy lodge]

spoken like some who cant do a pushup

[cyan burrow]

oh shit this will tak ages.... 17gb if generated images to go through

[oak spoke]

why push up when you can fly

[cyan burrow]

I'll do it the easy way and just delete folders that look irrelevant

[gloomy lodge]

just pick like 30 they look nice at random

[nova remnant]

post them all on tumblr and clear the SSD

[oak spoke]

my art folder

[gloomy lodge]

^

[oak spoke]

🤣

[gloomy lodge]

i clear mine out every time it hits 15k

just pick 3 of all the good prompts and delet everything else

[oak spoke]

I have spare discord server to purely store prompts

[cyan burrow]

16.4GB LEFT!

[oak spoke]

tattoos

[gloomy lodge]

love the bright colors

[oak spoke]

color you say?

[gusty palm]

looks like a goreshit album cover

[oak spoke]

that's what I was going for

[gusty palm]

good taste the jingo

[cyan burrow]

trying to get my merge to work :<

the whole thing just crashed, fuq it I'm taking a shower

[gloomy lodge]

first time? lol

[gusty palm]

i kinda like this model

revanimated

[cyan burrow]

Dragonball She

play on wurds

[gloomy lodge]

I like it too

[supple raptor]

This man doesn't know he can run his own PromptInspector bot locally and store both the images and prompts in DMs

https://github.com/sALTaccount/PromptInspectorBot

[gusty palm]

i have a prompt channel in my disco server ><

[oak spoke]

I mean that server was already storing meme and ss

so why not add prompts there

[supple raptor]

Because the bot makes the prompt and generation parameters more easily accessible. Better than just using it in DMs is adding it to your server so you just click the reaction and it'll DM the prompt.

[gloomy lodge]

ooo i like that prompt what is it

[supple raptor]

I highly recommend PromptInspector. It'll run on basically nothing. I run it on a free instance through Oracle Cloud for one of my servers.

[supple raptor]

1girl, mecha musume, fortified suit, lime hair, twin braids <lora:mecha_offset:1.0> on Kakigori_v1

IIRC the LoRA is one of Lykon's.

[gusty palm]

it sounds cool idk anything about running a discord bot tho

unless it means those ones u invite like septapus

[supple raptor]

It's one you have to create yourself and add to your server. Creating the bot is easy. You do it on https://discord.com/developers/applications
Once the bot is created you just add the two intents listed in the repo You'll get the bot token from the developer portal where you made it and paste that along with the channel IDs for the channels you want it to monitor into the .env file that the readme tells you to create.

[gusty palm]

does this alien have boobs

[oak spoke]

naked alien

[supple raptor]

The one step that the README leaves out is that you'll want to create a python venv and activate that before running the pip install command.

[oak spoke]

time to purge boob alien

[gusty palm]

is it something i need to run like as a service

[oak spoke]

ded

[supple raptor]

On your local machine you can just run it normally in cmd/terminal by activating the venv and running PromptInspector.py, but you'll have to leave the CMD window open. If you run it on a remote server then you'll want to run it as a service so that it will stay running after you close your SSH connection.

[oak spoke]

[gusty palm]

i used to have access to a virtual server back in the minecraft days

[supple raptor]

Oracle Cloud has a free tier. It's a bit of a pain in the ass to sign up for, but it'll let you run up to two instances that are plenty powerful enough for that box for free.

[gusty palm]

that sounds cool and useful

[supple raptor]

Yep, it's great for little things like that which don't need much power.

https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm

[oak spoke]

run9

👀

[native halo]

is this sdxl?

[gloomy lodge]

no its power by my autisum

[native halo]

https://tenor.com/view/neco-arc-gif-25900194

[gloomy lodge]

What sdxl tho ive had like 4 people ask me if my garbage is sdxl

[gusty palm]

pretty decent steven spielberg

[native halo]

what model is it then?

[gloomy lodge]

fucking love that little emot and here a cute girl in dress a an ambient occlusion render, full body, Carne Griffiths, Thomas Moran, ((Arcimboldo))

Negative prompt: (((multiple girls))), lowres, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, out of frame, extra fingers, mutated hands, (poorly drawn hands), (poorly drawn face), (mutation), (deformed) breasts), (ugly), blurry, (bad anatomy), (bad proportions), (extra limbs), cloned face, flat color, sketch
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 4201519291, Size: 960x640, Model hash: 54d8a62d61, Model: darkSushi25D25D_v30, Version: v1.3.2 the whole prompt and if ya want i can show ya how i spit shine the images

[native halo]

🙏

[gloomy lodge]

just dm me or something if ya wanna learn how i do it its simple so good chance you already know how

[nova remnant]

outpainted this

[oak spoke]

face in the wall

[nova remnant]

lol i know but that's just graffiti

[nova remnant]

here is another version

[native halo]

that looks very good ima steal it

[gloomy lodge]

gosh gonna make me bluch

blush*

[native halo]

[nova remnant]

the color tone and texture is amazing

i wonder what vae and model you are using

[oak spoke]

she grown

now make her a granny

[gloomy lodge]

got me all gushing with pride

[nova remnant]

humm, are you using stylus by any chance?

[gloomy lodge]

my Vae and wtf is stylus

[nova remnant]

i use that vae too, but your images look unique somehow

[native halo]

i gotta say that model is very eyecandy

[gloomy lodge]

idk the Power of Autisum compells me Model: darkSushi25D25D_v30

[nova remnant]

you have special touch with prompting then, cause i use dark sushi too

[gloomy lodge]

Imma be Using Nubby's model later this week so if you got that one ill be promping for that one too later

[oak spoke]

speaking of eyes here have an empty eyes

[gusty palm]

StylePile processing complete. Here's a random tip:
During long cold winter nights, you can turn your PC into a heater by generating hundreds of images non-stop.

[gloomy lodge]

most of the time i make grinds and test between 5k to 15k prompts on models and pick whatever i think looks cute so winter will come with warm nights

[nova remnant]

uhh no meta data... can't see the prompt

[gloomy lodge]

a cute girl in dress a an ambient occlusion render in a slum, full body, Carne Griffiths, Thomas Moran, ((Arcimboldo))

Negative prompt: (((multiple girls))), lowres, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, out of frame, extra fingers, mutated hands, (poorly drawn hands), (poorly drawn face), (mutation), (deformed) breasts), (ugly), blurry, (bad anatomy), (bad proportions), (extra limbs), cloned face, flat color, sketch
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 4291548939, Size: 960x640, Model hash: 54d8a62d61, Model: darkSushi25D25D_v30, Version: v1.3.2

the exact prompt and seed used

[nova remnant]

cool thanks, im gonna try that out

[gusty palm]

i just got the most bbw looking mario ive ever seen

[nova remnant]

out of curiosity i used a different model instead of dark sushi with minor changes in the settings, and this is what i got .. i like them

[nova remnant]

and btw this is pro art grade

ADetailer is such a neat extension, before I had it every full body render had some weird gltiches on the faces, now they all look aesthetic

[native halo]

steven seagal the gamer

🙏

[nova remnant]

awesome concept

[gusty palm]

well thats horrifying

eepy time

[oak spoke]

death stranding

[nova remnant]

there is a visible facial difference between using cute girl and mature girl

i guess sweet would a generic descriptor

[cyan burrow]

hah with minimal autotune (like shifting sharp/flat to full and moving a key or two that I stray too far on) I still sound human and since I vocal fry and guttural etc. I could actually sing on a dark electro or metalcore song or something. Hmmmm, I might get lessons first though

[summer shuttle]

dead chat

[bitter latch]

https://giphy.com/gifs/cat-cute-roomba-3iBcRAErFhFwoTVbN5 i'm awake.. I think

[dawn barn]

Just put it at the end, it barely affects image

The start of the prompt is the thing given most attention to

best quality, masterpiece, gives it a slight bokeh effect

And affects composition SLIGHTLY

[nova remnant]

ahh ok, i wrote a simple generic prompt that seems to work fairly good for the positive prompt section
(highly detailed:1.4), (beautiful, aesthetic:1.2), vibrant colors, ultra quality, HDR, cinematic lighting, soft focus, dynamic angle

[bitter latch]

a broken effect, yeah that tracks 😄

[dawn barn]

No

Bokeh

As in

[bitter latch]

broken

[dawn barn]

Centre sharp outside blurry

[bitter latch]

got it

[dawn barn]

Focus

[nova remnant]

bokeh = blurr

[bitter latch]

so.. broken

[dawn barn]

It's a focus thing

It can look really good

[bitter latch]

so broken focus

😄

like it's got adhd

[nova remnant]

it foucses on front and back to give it a nice effect

[dawn barn]

I'd put highly detailed down to 1.2, and change ultra quality to high quality, and think twice about HDR, cinematic lightning, and dynamic angle

[bitter latch]

https://giphy.com/gifs/SWR-Kindernetz-kindernetz-wir-freundschaft-grenzenlos-LOQGPaOXl5qgc0fp62

[dawn barn]

Soft focus is the same as a slight bokeh effect

[nova remnant]

ok, i will try that suggestion, but about thinking twice? why?

[dawn barn]

HDR will result in a certain type of picture

Same for cinematic lightning

Same for dynamic angle

Cinematic lightning typically means the light comes from the centre

Top down

Or back to front

HDR means that the picture will always have dark and bright parts to show contrast

And I think it's not that common anyway with anime stuff

Since it's a photo setting

Dynamic angle will result in more diagonal positions

[nova remnant]

okay thanks, going to try those suggestions too

[dawn barn]

You don't always want them

So think twice

[summer shuttle]

what happened in here

[nova remnant]

so how about ....
(highly detailed:1.2), (beautiful, aesthetic:1.2), vibrant colors, high quality, natural lighting, soft focus

[bitter latch]

<---- is in silly mood.... i'll stop now

[nova remnant]

time to put that to test and see first hand

[bitter latch]

I'm on Paid time off today .. so yeah 😄

[nova remnant]

paid time off ... best kinda day

[dawn barn]

Why do you want to stop

[bitter latch]

yep plus haven't had two days off in a row in a while

[dawn barn]

Natural lightning will cause more sun shining pictures

Like sunlight

[nova remnant]

yes i want that bright feel

[dawn barn]

Step count might be low, or model / LoRA too strong

Dunno

[bitter latch]

need to focus on my character in SL

[dawn barn]

What is SL

[bitter latch]

Second life

[dawn barn]

And nature pictures

Uhm... as for the bright feel... How about soft lightning?

Or cheery lightning

[nova remnant]

soft light, hmm im not so sure, wouldn't that wash out some of the tones and textures?

[dawn barn]

Surprisingly just adding emotions to random stuff makes more sense than saying the wrong stuff

I mean you can just try everything

If your GPU is fast enough

Gimme one of your full prompts and I'll give it a go after dinner + walk with dog

[nova remnant]

yeah im gonna try them all out

its ok im good on rtx 3060 but thanks

want to test them out myself lol

so dramatic lighting and cinematic lighting are pretty much the same?

im looking at some of the prompts at prompthero and those are very extensive

[supple raptor]

Is she mining for cookies?

[frozen olive]

yeees

[supple raptor]

i love it

[steady grail]

a true fairy withh correct wings. what's that, a lora? mine were always a mess

[dawn barn]

They fucking suck or just wanna look good, or they know what kinda style they are chasing and are fine with adding more stuff to make their pic appear more professional

[frozen olive]

it's niji journey

:x

[dawn barn]

Just to show some of the slightly more experimental prompts

The first and last I was chasing the feel from the 2nd pic

The 4th one is singular prompt

2nd and 3rd pic are prompted using colours only

This here was with an SDXL prompt applied to a 1.5 model

Well, derived from an SDXL prompt

[steady grail]

sdxl is very detailed

[dawn barn]

None of this here is SDXL

[steady grail]

the 5 before weren't sdxl?

[nova remnant]

dpm++ sde karras with 30 steps might be generating too much texture

20 steps look more pleasing

[supple raptor]

transparent wings

https://cdn.discordapp.com/emojis/1028509125044359259.webp?size=64&name=MomijiWide&quality=lossless

[bitter latch]

test

good i'm back, stupid inet

[dawn barn]

No

Metadata should be in there

First 5 A41, after that the 4 ComfyUI not my workflow I was using someone's else's SDXL workflow then just put an SD1.5 model in tehre

[dawn barn]

AAAAAAAAAAAAAAAAAAAA

HOW

[supple raptor]

I didn't do anything special. My model with this noise offset lora just does it apparently.

[dawn barn]

Do you have metadata included? Automatic or Comfy?

Comfy

Nice

Thanks

[supple raptor]

Yeah I always post with metadata unless it's an edit that doesn't have it. They're from automatic.

[dawn barn]

Huh they are from automatic?

Then ComfyUI added automatic metadata import

What is the LoRA?

[supple raptor]

Oh, that's neat. That sounds like it must have been a pain in the ass to code with the number of extensions that are out there that add stuff to the metadata.

[dawn barn]

I don't think they did all the extensions

Just the base stuff

The Lycoris e.g. didn't get loaded as a LoRA

[supple raptor]

It's just a locon extraction from the original noise offset model. I got it from someone on here, but I can't find the link at the moment. I can upload it if you want.

[dawn barn]

thanks, I'm gonna try it without it first

What about the negatives?

negative_hand, (KHFB, AuroraNegative)

[dawn barn]

How do you extract it?

[supple raptor]

with the script at .\tools\extract_lyco.py found here:
https://github.com/KohakuBlueleaf/LyCORIS/tree/main

[dawn barn]

This here roughly sums up my outputs

It's like ok but not that comparable

[quiet bramble]

Any anime models trained on SDXL?

[gloomy lodge]

just get good at normal sd same diffrance

[dawn barn]

I just downloaded CounterfeitXL but it seems like it'll take a bit more time

[quiet bramble]

If there isn't a big difference, then I'll just stick to 1.5.

[dawn barn]

I mean

[dawn barn]

You are not going to notice the difference immediately

Since there's a lot of effort put into 1.5

And SDXL is new so ppl still need to figure it out

[runic vault]

it doesn't help either that sdxl needs a crap ton more power versus 1.5 to even use in the first place.

[nova remnant]

https://civitai.com/models/119163/shikianimexl
Anyone tried this model?

[stable nimbus]

Anyone know dataset for AOM3A1 (the 2.5d stuff)

[bitter latch]

hmm doing momo-miku daughter seems to help. not sure if it's the - between the names or the word daughter thats giving me slightly diff faces

goes random

level 1 random engages( goes from 1 to 10, 10 being insane level )

I don't have ducks and they aren't in a row.. I have squarrils.. and they are everywhere!

[bitter latch]

SQUIRRLE!

and her daughter

[bitter latch]

ok now she's crazy! hehe

and last one for abit...

[supple raptor]

AOM models are all merges, not finetunes (hence the name "Abyss Orange Mix") so there is no single dataset.

[gloomy lodge]

Yo nubby the girl in the fish bubble is your model btw looks real cute gotta do some more testing but so far sooo cute

[supple raptor]

Oh that one came out really cool! Cute is definitely what my model does best.

[limber moss]

can confirm, uwu factor very high

[supple raptor]

https://cdn.discordapp.com/emojis/1097118673682038904.gif?size=64&name=AyameHeadbang&quality=lossless

[gloomy lodge]

model goes hard

[oak spoke]

sdxl also incompatible to a lot of lora

google chrome be like ^

[sonic echo]

which model is this? that's an amazing style

[gloomy lodge]

kakigori_V2 by @nubby UwU

[sonic echo]

thanks!

[gloomy lodge]

also the fish pick is the same model

[oak spoke]

bright light

[stable crane]

outline 😂

[weary roost]

how to prompt character to looking away?

not straight to camera?

i tried looking away and not looking at viewer, still got a result looking to camera

[bitter latch]

facing forward

oh looking away ummm.. hmmm

facing away?

or facing (direction) maybe?

[weary roost]

oh! thankyou! will try that one

[bitter latch]

or a POV away

POV away

[weary roost]

oh, and how to get some sparkly things..kind of light petal, i dunno what to type actually also i got no reference lol

[runic vault]

light particles?

[weary roost]

yeah kinda light that particles .. the middle one

[weary roost]

its hard to remove thigh, negative prompt already use skin

[runic vault]

have you tried pants?

(white background:1.5), traditional media, ((knight)), ((fantasy)), (spear), shield, full body, red eyes, black hair, long hair, ((gauntlets, armored boots)), pants, 1girl,

[bitter latch]

borrows the prompt

tries some different settings

[weary roost]

it works

oh but didnt tried your prompt

what does the () function?

[weary roost]

where i can learn that kind of prompt?

[runic vault]

it adds more strength to the prompt

[weary roost]

oh!

and the 1.5 ?

[runic vault]

just another way

it would equal to (((((1girl)))))

much cleaner

[weary roost]

1.5 equal to 1girl?

oh

[runic vault]

no this (((((prompt)))))

[weary roost]

sorry, missintrepet

ok got it

thankyou !

[weary roost]

where you do your prompt?

[supple raptor]

Here you go @weary roost https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#attentionemphasis

[weary roost]

whoaaa, thanks a lot!

[bitter latch]

waits and waits some more. maybe I used to much.. hmmm

well thats interesting

and oh boy what a prompt: clean shining silver paladin armor, baroque painting, heroic, ultra realistic. cinematic, dynamic. (magic:1.14) the gathering style. epic fantasy, insanely detailed, 4k, symmetrical face, rpg character reference. gorgeous. dnd digital art by artgerm and greg rutkowski and alphonse mucha
Negative prompt: epic still of kim kardashian trapped in a transparent alien liquid, wet flowing hair, gooey skin, illustration, unreal engine 5, 8 k, made by h. r. giger. trending on artstation, digital art, rendered in (octane:1.04), epic lighting, cryengine, digital painting, concept art, smooth, sharp focus, illustration, artstation trending, full body, cinematic, videogame cover art, feminine, very detailed eyes, artgerm, jeremy lipkin and michael garmash and rob rey
Steps: 20, Sampler: UniPC, CFG scale: 7, Seed: 2664660720, Size: 512x512, Model hash: 184e85d891, Model: AnimeToonPlus, Denoising strength: 0.5, Clip skip: 2, ADetailer model: person_yolov8n-seg.pt, ADetailer confidence: 0.3, ADetailer dilate/erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.7.9, Hires prompt: "(Fantasy background:1.5), traditional media, ((knight)), ((fantasy)), (spear), shield, full body, red eyes, black hair, long hair, ((gauntlets, armored boots)), pants, 1girl", Hires resize: 1024x1024, Hires steps: 10, Hires upscaler: SwinIR 4x, TI hashes: "an6: 54c90e776f2c, kkw-Extreme-Neg: 43eba3ed8ad5", Color Enhance: 0.5, Pad conds: True, Version: v1.5.1

[weary roost]

what a negative prompt you type there haha

[bitter latch]

used an extension Dynamic prompts. magic prompt and feeling lucky settings

[weary roost]

steps means for better result right? or am i wrong?

[weary roost]

these emphasis only works for stable diffusion right?

[runic vault]

well you in a stable diffusion discord

[weary roost]

ok i know your pic mean

so, after reading about () and [].. i can use prompt [car] to generate less car, instead negative prompt: car.. right?

[stable crane]

I don't think that's how [] works

[weary roost]

can () increase its attention to negative prompt?

[oak spoke]

Yes

[weary roost]

thankyou

[gusty palm]

queercore lego

how do u get the touhou characters

oh i have a cirno meme

[gusty palm]

hulk smash

[supple raptor]

Where's the meme tho?

[gusty palm]

I guess not as much of a meme as I nemembeted

[nova remnant]

orange mix vae and anything vae they both look so washed out compared to kl-f8-anime2.vae

i had to basically delete those other vae generated images

[native halo]

u should try clearvae or graupel vae

[nova remnant]

never heard of those, do you have links for them?

[native halo]

https://civitai.com/models/22354/clearvae

https://huggingface.co/p1atdev/graupel-v1/tree/main

[supple raptor]

This is the only VAE I use for anime. https://huggingface.co/nubby/kl-f8-anime2-blessed/blob/main/WD1-4-kl-f8-anime2-bless09.ckpt

It's kl-f8-anime2 but with the contrast dialed back just a bit which makes it look a lot better on most models.

[nova remnant]

thanks, checking them out now

[gusty palm]

the most recent clearvae ias a 404

[native halo]

😔

[gusty palm]

cleanvae looks like a updated one

[native halo]

yea i get less nans with those

[nova remnant]

three years from now and i wonder how far they can push SD towards perfection

given politicians don't screw things up as they always do

[gusty palm]

realtime ai powered video game engines

[nova remnant]

yes and video productions too

the world is going to wake up

and then ... and then... creation of new life

[native halo]

but first the old ways will try to stop it,using only thing they know

[nova remnant]

yes they will try till they realize if they try to stop it they will only collapse with the old ways

the reason for the changes are not a fancy desire, its a necessity cause the current systems are unsustainable in many ways, but people have to catch up with the reality

the fancy desire just seems to coincide with the changes to come

[gusty palm]

hiding models behind copyright and guidelines and closed source

im loving these pop culture wildcards

[nova remnant]

copyright doesn't make sense with the concept of singularity

[native halo]

not that,there will probably be a war between religious groups when ai gains more traction maybe in the next decade

[gusty palm]

i thought u meant restricting the capabilities to an oligarchy

[nova remnant]

religions are mostly a violent practice for those who follow it to the tee

[gusty palm]

true

[native halo]

yea i guess we will have to wait and see

[nova remnant]

the reason for violence is not necessarily always religion tho, its fundamentally because of scarcity

[gusty palm]

humanity has pretty much achieved post scarcity nowdays

just bad distribution of resources

[nova remnant]

yes the technology is there but not yet implemented

[nova remnant]

outdated beliefs and practices are in the way

[gusty palm]

i dont wanna get banned for heresy

[nova remnant]

what do you mean

[gusty palm]

i live in florida

[nova remnant]

poltiics religion money are all running out of use, its not about heresy

[gusty palm]

its practically illegal to be critical of outdated beliefs

[native halo]

dont worry soon our new AI religion will start,will grow and grow and take over

[nova remnant]

that's controlled ai not the true definition of Ai independent of human biases

and a lot of the human biases come from the three fundamental systrems that are now fast becoming obsolete and unsustainable (money, politics, religion)

[native halo]

🙇

[nova remnant]

those are sensitive topic for discussion but least we can do is acknowledge the reality

cause culture, society, nations are built upon those decaying systems

[gusty palm]

tahts becoming rare

acknowldging reality

[nova remnant]

well i say that for our own sake

if we fail to acknowledge the reality the reality will check in with us eventually

i think there is a reflection of that already in our social and ecological spectrum

honestly tho, i think we are in a cross road of utopia and dystopia ... just gotta make the right collective choice

[gusty palm]

knowing we could be like star trek but end up like the ferengie

this is space dandy ive never heard of it

[nova remnant]

star trek or ferengi ... well its pretty obvious we need technological abundance that can supersede human control to break free from this rutt

and i say that from a constructive point of view that can create abundance in the production and distribution chain

but to argue about human baises in their most fundamental state, its all about love desire and resources

we can try and alleviate a lot of those conflicts by taking care of resources, love idk

[gusty palm]

i mean like how we already produce enough food and housing for everyone

[nova remnant]

yeah

[nova remnant]

a lot of those go to waste in order to prioritize business profit through buy/sell

those are arificial scarcity and a structural violence in our system

[gusty palm]

ye

[nova remnant]

but honestly i think those will become manageable through technology that are coming online

[gusty palm]

i imagine if we invented a repicator theyd still find some excuse to withhold resources

[nova remnant]

because of fear of scracity

[gusty palm]

ive become cynical

[nova remnant]

me too and my bigger concern is human lust

in this system we live now love is exploited

humas are used for it, not loved

and social conditioning teach us you can't love everyone

and if you do that's considered cheating and immoral

[nova remnant]

on top of that there is the obvious reasons for economic scarcity

and exploitation, so distrust is there built into this economy

[gusty palm]

read soul of man by oscar wilde

not totally relevant but u might like it

ive been meaning to do a video about it

[nova remnant]

i would love to watch your video but im not into reading books

[gusty palm]

i only read the first bit (¬‿¬)

[stable nimbus]

yeah but there must be some dataset that contributed to the 3dish feel right

[nova remnant]

i mean i was never into reading books to even begin with

i spend a lot of time on youtube and audio files

[gusty palm]

i prefer videos too

[weary roost]

guys, need help..

youre in car, and its raining..and you seeing the car in front you, the red light from the car in front of you seems like have a "blurry" effect..

what you call it?

[nova remnant]

curious to know since you feel cynical about the ways things are how do you stay motivated for changes

[gusty palm]

glare or or lens flare I think

[gusty palm]

I don't :c I'm super depressed and barely function

[nova remnant]

i mean do you look upto any aspect of life for the changes you hope for?

[gusty palm]

I'm gonna go to the free clinic next week and also hope my brother gives me a job at this fire extinguisher company

[nova remnant]

i don't mean to sound insensitive but human labor for survival is something that i find very cruel

[gusty palm]

im mostly hoping stuff doesnst get worse and ill be happy if it improves im worried about going homeless

and same

[nova remnant]

i know a few people in my cricle who live on social wellfare funds, you dont haev to go homeless if situation is that extreme

also i think UBI could be a common thing part of the changes that are inevitable

[weary roost]

thankyouuu

[gusty palm]

i applied for a disability appointment but the interview isnt til end of september

[gusty palm]

theres a few different types too i havent had good luck getting stuff like that in diffusion so far though but ive seen it

[nova remnant]

a month left, hope things go well

[weary roost]

yeah, i want to set that kind of effect to my current prompt..but i dont know what to type lol

[gusty palm]

johnny bravo

[nova remnant]

can you share the prompt you are using for that effect?

blurry headlight?

i closed my SD console

[gusty palm]

maybe stuff like lens flare,glare,long exposure,bokeh,

[nova remnant]

yes

those with the mention of headlight maybe

[gusty palm]

i saw one that had like a long exposure with swirling type effect i wish i could remember the prompt

blender has a glare node that would p[robaly have hepful terms, light streaks

[nova remnant]

ahh blender, i wanted to try it, but never got around to using it

[gusty palm]

u could try adding the effect in post

i can almost picture the effect youre trying for

[gusty palm]

how about u btw

[nova remnant]

i've given up on most expectations but i think there are massive changes possible that can turn the course of life for everyone and profoundly better

and part of that motivation comes from technological progress and girls in general

i'm also glad Ai is commonly focused on now, that could potentially revolutionize everything about society

[gusty palm]

girls are amazing

[nova remnant]

i think if you define girls who are not suffering from social constructs, and scarcity they can be the prime source of life's joy

[gusty palm]

i used to think alien confirmation would unite humanity

[nova remnant]

how do you depict aliens tho?

[gusty palm]

well more the idea of knowing humans could be part of like a friendship with another species

[nova remnant]

ok so in movies they are depicted as strangely figured characters, do you think of them that way?

[gusty palm]

ooh

idk they could be anything

i like the greys tho

[nova remnant]

i find it hard to resonate with that concept that are portrayed in movies

[gusty palm]

star trek has so many species i just assume theyre not gonna look like anything i could imagine

[nova remnant]

well honestly im more fascinated by girls than strangely figured unseen aliens

and when i talk about girls i just don't mean their physical apperances, thats one yes, but also their inner femininity that you can't just describe in physical terms

so yeah, if there is one thing i wanted to cherish in this entire universe, its the girls

[gusty palm]

alien girls

[nova remnant]

girls can be so gorgeous they may seem like aliens to some you know?

gorgeous is just one attribute, but also sweet, cute, kind, adorable, etc

i just can't wait for the day when technology has outpaced human limitations to remove all sorts of physical and intellectual burdens of socity

when that happens you will get to see true beauty of humanity

[gusty palm]

same

alien girlm

[nova remnant]

aww naaa, organic girls are the most beautiful

btw you might be interested in youtube channel that discusses automation, technology, Ai .. look for the channel "Peter H. Diamandis"

the video i'm watching there now talks about putting humanoids in workplaces and homes that can make human labor a choice and not obligatory

[gusty palm]

ok cool

pholisophy tube is a good one

[nova remnant]

philosophy tube? i think i see it

if this is the right one

[gusty palm]

yuss

stuff like that has been too serious for me lately i love science channels though like pbs spacetime

im not sure if she has a video about scarcity but it sounds like something she would im trying to think of any yters that talk about post scarcity

[nova remnant]

interesting you commonly mention the term post scarcity, are you familiar with venus project?

[gusty palm]

https://www.youtube.com/watch?v=imNtSPM3-r4&t=1s&pp=ygUHa29zZW1lbg%3D%3D
this is a great listen, its fiction a speculation about the future of humanity scifi

that soudns vaguely familiar

[nova remnant]

will check in a bit

thanks 🙂

[gusty palm]

yw i subbed to peter h too

[weary roost]

havent found it..i just tried bokeh haha

[gusty palm]

castlevania

[oak spoke]

so we become furry?

[gusty palm]

its like a story about humanity splinters into multiple different planets and they eacho evolve into different things

i think furry is one of them tho uwu

[oak spoke]

ahh

[nova remnant]

im kinda enjoying the story ;D

fictional and entertaining

lol

ok i need to stop laughing a bit

i find it cute that you shared it

[oak spoke]

@gusty palm @nova remnant you should try watching this
https://www.youtube.com/watch?v=WfCRJMsEubg

really cool concept

[nova remnant]

ok but give me few more mins im still watching that video katie shared

[oak spoke]

alr

[nova remnant]

omigosh lol

@gusty palm its so funny lol the story is pretty long too

[oak spoke]

I got a feeling that at that point human wouldn't remember this old version of us

to them they were always like that

[nova remnant]

lol behold the hedonists species ...
the hedonists were created by the QU to be the pampered pets, living carefree lives of pleasure, their world was a tropical paradise of succulent fruits and lakes of sweet bacterial juice, they were the only animals on the planet, and had no predators or competition to deal with, under normal conditions this might lead to overpopulation, but the Qu designed the hedonists to only get pregnant after mating an enormous number of times, so the population was stable, and the hedonists filled their lives with eating, sleeping, and mind blowing s**, thier minds were blissfully empty, cause who needed to think when they are having such a good time

[oak spoke]

for all we know we are new human already

[nova remnant]

im not too fond of the idea of passing genes generations after generations, i think death is a big crisis and only accepted by the collective cause we are incapable of doign anything about it as of now

altho reincarnation and after life are also something to think about

[oak spoke]

lol fat chance

would be nice tho

[nova remnant]

we dont know

[oak spoke]

just get tortured endlessly in hell

sound like fun

[nova remnant]

about hell that's just man made construct

inflicting fear for control

[oak spoke]

the whole concept is manmade construct

out of fear of death

[nova remnant]

that's a valid point too

but energy is said to live on

[oak spoke]

not lived more like recycle into smt else like

water

you drink water eventually you piss it out

and eventually someone else drink it

cycle go on and on

[nova remnant]

i mean the energy that's not a physical element about life continues to live maybe in another dimension that we can't reach from here

i can share an interesting short video on dimension and how we perceive up to 4D and nothing beyond through our innate senses

https://youtu.be/C6kn6nXMWF0

watching the cosmic apocalyspe

the thing about the present day narratives with religion and it's conditioning on society are absurd

i dunno where to even talk about it, just doesn't make sense to have a religious discussion, i mean if we are willing to change our lackings, crisis, inadequacies we should talk about how we built our society

we have cognitive limitations if you watch that video about dimension, its interesting we at least understand it logically even tho we lack access

[nova remnant]

i'd rather choose to appreciate the beauty of girls and nature and life instead of worshipping and living in fear

[bitter latch]

weird outcome

tired

goes back to bed

[bitter latch]

goes back to sleep thud

[gusty palm]

he's such a nice guy too, the author

[oak spoke]

you said it almost like he died lmao

[gusty palm]

ahha

hes done a ted talk apparently i didnt even know about

ive been posting to instagram

my page looks very cool now

[bitter latch]

huh.. loks like she's in a jeffies tube in star trek

[bitter latch]

hmm wonder how she got on a the Melinium falcon

[cyan burrow]

how do you give away a philosophy degree?

[bitter latch]

"this is a private club miss"

[gusty palm]

head empty (⁠◕⁠ᴥ⁠◕⁠)

[oak spoke]

what do yall think?

[supple raptor]

You could say that the datasets of all of the models that were merged together to make it contributed to the look, but even if you had those exact datasets and combined them and then trained on them it wouldn't mean the resulting model would look even remotely similar.

[stable nimbus]

Wait so by mixing and merging the models can generalize to something of which a dataset does not exist?

[supple raptor]

Merging is just averaging the weights together, so yes. When you merge models together you will end up with something that falls somewhere between the two. When you add MBW merging it becomes more complicated as different blocks get merged with different multipliers. In short, yes.

[stable nimbus]

are these blocks responsible for different stuff

is it possible to do very specific merging? like only for hands or somethin

[zinc flax]

Yes

[supple raptor]

Sort of, hands will normally be primarily located in layers IN00-IN02.

[stable nimbus]

i guess training loras on models like those is gonna be pretty hard then lol

[supple raptor]

Not really. This is how Aurora was made, and that model is quite good for training LoRAs. AOM2 was also used for training a ton of LoRAs and they're mostly fine, despite all of the problems that model had.

[stable nimbus]

i mean aom3a1

very 3dish

i tried trianing a lora on that for eiki

it didnt retain the 3d stuff

because the dataset was 2d

[zinc flax]

Might wanna stick on training with nai though since most models use that as a base

[stable nimbus]

i did try that

didnt really help, output was still 2d with aom3a1

[supple raptor]

That just means your hyperparameters and dataset caused over fitting on the style. Basically, you accidentally trained a style LoRA.

[stable nimbus]

but i think my lora mightve been badly trained in general lol

My dataset was like 30 images before, but i have one with 1700 now, might try again

I tried on sdxl turned out shit

@supple raptor any suggested tutorials

[supple raptor]

Not really. There are some that can get you started with the basics. But I've never seen one that can actually help figure out what needs adjusting to get your desired result.

[stable nimbus]

wat do then

My outputs look like this 1% of the time

and then this 99% of the time

[oak spoke]

that look surprisingly interesting lmao

[stable nimbus]

That one looks pretty nice and funny

but it gets old quick

[oak spoke]

are you using default 1.5 model or what?

[stable nimbus]

sdxl base model

[oak spoke]

ahh

guess it got the new model syndrome lmao

[stable nimbus]

this is my first lora i think my hyperparameters are just bad

[oak spoke]

here's my sdxl based model result

[oak spoke]

maybe problem with some source images?

or not enough of them?

[stable nimbus]

1700

i think theyre all pretty good quality

[elfin bolt]

are you training it on base sdxl? try training it on lykon's Animediffusion or CounterfeitXL

[stable nimbus]

im just waiting for wdxl

gonna try on that

but even the base sdxl model is very good at anime so shouldnt really be a problem i feel

[elfin bolt]

Maybe the tagging? did you use BLIP2 or Danbooru tags?

[stable nimbus]

danbooru

[elfin bolt]

ah thats a problem, SDXL is trained using natural language

so your usual danbooru tag will be messy on SDXL

[stable nimbus]

This is the output from sdxl with a tag based prompt

base model

[elfin bolt]

your prompt is sparsely being understood by u-net

its better to try prompting naturally than using danbooru tag using base sdxl

[stable nimbus]

if i retag my dataset with blip2 do you think itll be better

[elfin bolt]

also limits your dataset, for characters, 60 is already more than enough

try to make it with more or less consistent artstyle

[stable nimbus]

well im willing to train for longer for higher quality

[elfin bolt]

Well try limiting your dataset, use blip2 and add persistent keyword to every caption "anime style"

if you want to use danbooru tagger, try WDXL 0.9, counterfeitXL, or lykon's animediff

[cyan burrow]

1.5 Clara(img1+2) vs XL Clara (img3+4)

[gloomy lodge]

never giving her a pencil again she ruined my work

[dawn barn]

SDXL is trained on both

[elfin bolt]

source?

[dawn barn]

Search for "CLIP and OpenCLIP" in this discord

Also might be on the model page

They specifically did it because nobody really got used to the way to prompt 2.1

Besides all the other stuff that happened

So they just use both clips to process the thingy

[elfin bolt]

I might be missing something, but double clip doesnt necessarily meant "danbooru tag"

[dawn barn]

So, we used CLIP with 1.5

Which is the CLIP from openAI

This can handle danbooru tags, and generally likes a more tag-style prompt

But we also use OpenCLIP, which is an open source CLIP model

It was trained using natural language only

Which is why it can't really handle danbooru tags

That was used for 2.0 / 2.1

[elfin bolt]

I mean yeah, but these double clip model is likely still being trained using natural language captioning rather than danbooru tags

thats my assumption

thats why we need a large finetuning effort like WDXL to make it able to recognize the danbooru tags

[dawn barn]

From my experience on the base model, it handled the base tags just fine

Don't literally copy the danbooru tags with the underscores

I don't think they are in the dataset? (Which is accessible btw)

But write it tag style

Just with like spacebar in between them and not the danbooru specific underscores

If possible

I mean tbh I didn't try it on the danbooru tags

But from what it sounds like you had a bad experience with these

[elfin bolt]

its not exactly bad experience, but more it doesnt respond to the prompt as opposed to the first NAI leak back then

[dawn barn]

I personally think this is because SDXL is not really an anime model

I'm about to try things out on anime models

[wet elm]

is there a way to merge models with loras?

[brittle jay]

is this ai?

[runic vault]

yes?

[brittle jay]

how

model?

[runic vault]

based66

[brittle jay]

I C

thanks

also why use NAI-based vae?

[runic vault]

it works

[brittle jay]

I mean wd1.4 vae also works

[runic vault]

i find it tends to be too saturated to use

[brittle jay]

then there is the fix thingy

oh wait it doesn't fix saturation

only brightness and contrast

[gusty palm]

i had multiple scammers DM me on instagram today trying to get me to sign up for opensea (which i did) but then they tried multiple ways to get me to send them eth to "initiate a transaction"

[brittle jay]

bruh

[gusty palm]

on the plus side i have officially "minted nfts" whatever that means and have a wallet and avoided scams

but also missed out on like 15,000$ sales on my artwork

[brittle jay]

everyone on instagram is a scammer

[gusty palm]

tru

;[

[brittle jay]

or a groomer

[gusty palm]

or f*mboy meme account

[brittle jay]

that I wouldn't know about

[gusty palm]

ya mea neather haha

[keen hollow]

What's new folks? More people moving to Comfy/SDXL still?

[gusty palm]

iconic

[gloomy lodge]

Idk SDXL is over rated

[slate night]

I can't even run sdxl so lol

[gusty palm]

thats so good

[bitter latch]

oops on hte size

oh I like the new uniform

sings: a most pacular thing

there thats better...

[bitter latch]

now thats minimalist drawing

to thicc?

😄

[merry tendon]

I haven't posted in here so long

what are you guys up to currently? something new? (i am mostly active in the AI ERP community lol)

i only saw that sdlx came out

[gloomy lodge]

stuff like this

[merry tendon]

what model?

[merry tendon]

ah, i remember that one

[supple raptor]

I'm like 15 hours into BG3 since launch yesterday.

[gloomy lodge]

what is BG#

BG3*

[sharp bluff]

when yall go to update your models and loras do you guys delete the old versions manually?

or is there some magical tool im missing

cuz going thru these and deleting is a painnnn

[gloomy lodge]

Just done use lora i dont

[wise pilot]

hmm how i fix this ?

[gloomy lodge]

Vae

[gloomy lodge]

Vae

[wise pilot]

ihave this

[gloomy lodge]

try a diffrent vae might do the trick

[wise pilot]

i'm using symlink before, it tell me can't load vae but on webui it still showing vae selected

use 3a3 still have purple

[gloomy lodge]

i think are are loading it in the wrong spot maybe one sec

[wise pilot]

i have this

[gloomy lodge]

ah so you dont got one

[wise pilot]

did i missed something

8months not using SD

[gloomy lodge]

ye i think 1 of 2 thing happened you got your vae in the wrong folder or you dont have one yet

[wise pilot]

but why ur settings have upscaling/face restoration

mine

[gloomy lodge]

Addons or your model is not updated

your Vae should pop hp here tho id ya got one

[terse ore]

Anyone use the Stable Diffusion plug-in for PS? I'm trying to generate outpaint to extend an image of a boxer so his arm can be seen because it's cut off from the image. But when I try, it gives me a completely different image in the outpaint selection area

[wise pilot]

i'm trying to inpaint but it always lead to awful result, it's not even bend with image

[native halo]

because you are using too high denoising strenght

[gloomy lodge]

what are your setting on it

[terse ore]

I tried a lot of different settings and none worked out.

[gloomy lodge]

just send me a screen shot of what ya got so i know what you are doing

[terse ore]

OK let me try

[terse ore]

I'm trying to expand the image to shoy more of the scene like the glove etc

[gloomy lodge]

your gonna want the steps to me a lot higher like between 60 - 90 for this kind of outpaint

also denoise try something lower like .35- .55 that range

[terse ore]

Selection and Masked I put to "original" and it doesn't expand the image in the white areas for outpaint, it just changes it like as though its inpainting

[gloomy lodge]

do fill

might be what ya need

if that does not work set steps to 80 ish denoise to .8 and fill might work then

i dont have the time one as you so not 100% on how to use this version

[terse ore]

Hm I tried all those things and also changed to "fill" on both Masked and Selection but still nothing. I think I'll change each setting one by one and eventually find out what works perhaps

[gloomy lodge]

I got no idea whats wrong with it on your end then cus the other versions of this seems to work

[terse ore]

I'll go through tutorials or something idk. Kinda late here so I'll have to try tomorrow. Ty for trying to help out though

[wise pilot]

hmm my inpainting result have exactly same as original image use to inpaint, is that weird?

[gloomy lodge]

pics?

[gloomy lodge]

crop and resize first thing first

[native halo]

set seed to random and try to change sampler to eular a

[gloomy lodge]

^

[wise pilot]

still same result

...

euler a

crop and resize

random seed -1

[gloomy lodge]

fuck it want to send me the pic ill fix it

only other think i can think of

[wise pilot]

yep it work now, i just forgot masked that area

[gloomy lodge]

lol

[bitter latch]

hmm I would post what I got with that prompt but well , not sure it's sfw

or this channel haha

[bitter latch]

cowboy bebop?

[jagged wasp]

feels like it, bot straight up made an anime I'd actually watch

[bitter latch]

hmm interesting

[jagged wasp]

this one's kinda crazy thing is scooping up whole planets

[gloomy lodge]

the magical creatures that make the art got lazy only did half

[native halo]

what model is that?

[jagged wasp]

animeArtDiffusionXL base refined with anyLoraAnimeMix

[native halo]

🙏

[bitter latch]

turns the crazy rando meter up a notch or three

[ebon plume]

A girl on bycycle

[gloomy lodge]

INDIAN detected

[bitter latch]

takes two alleive and goes to bed.. thud.. ow

[stable crane]

anemone?

[jagged wasp]

anewhat?

[stable crane]

haha looks like a character anemone

[jagged wasp]

ah

[native halo]

greetings fellow eureka seven fan

[stable crane]

man I don't know if I finished that show

I was very close to the end

[native halo]

hand drawn eyegasm

😍

[stable crane]

rahxephon was a fun one too

[native halo]

that one too

Gasaraki too if you like military

[stable crane]

90s rise up

[terse ore]

What are some general consensus things on SDXL so far? Idk too much about it

[gloomy lodge]

Nothing really worth swaping over for imo

[astral drift]

Gonna need better finetuned anime models before it's worth switching over completely

[gloomy lodge]

^

[astral drift]

It's better for complex Photorealistic concepts though

[gloomy lodge]

ew real woman yucky

[astral drift]

It also performs better with weapons like swords and guns, they actually hold the weapons correctly without using a Lora 😂

[supple raptor]

If you have less than 10GB VRAM it's viability is questionable at best even once good anime models come out. Less than 8, I wouldn't bother at all.

[native halo]

maybe someone someday will optimize it to run on 6gb? like they did with 1.5 to run on 2gb

[cyan burrow]

haai

[cyan burrow]

for training or just using sdxl?

are you saying we can't even use it with less than 10gb? viably anyway

cause I kinda notice that 8gb is really slow on generations

[supple raptor]

The speed is the issue on 10gb. I'm talking inference. not training.

[supple raptor]

You can use it on 6gb with ComfyUI, maybe even 4gb. It's just incredibly slow because it has to load one component at a time.

[native halo]

[cyan burrow]

I notice SDXL on 3070 ti is really slow

O

I've been considering adding things like 4070ti or waiting for this Q4/Q1 24 for series 5000

[native halo]

no one has tested the new 4060 ti 16gb?

[cyan burrow]

The reviews on it has been a universal very loud "THIS BLOWS"

it's not worth the cost

[native halo]

or maybe a quadro p5000 16gb i tried to find SD benchmarks/reviews but looks like no one has it or no one has uploaded benches of it

[cyan burrow]

I was gonna buy a 3090ti 2nd hand until I realized it's outmatched by 4070ti in performance (not vram) and almost outmatched by re gular 4070, same deal

and if I buy a regular 4070 new Ican get one for like 600 euro, alongside my 3070ti and with newer extensions etc able to make better use of multiple gpus

I could count it as 20gb in a limited way

They can handle different processes in the same generation now I think

like simultaneously

[native halo]

yea and i guess 5000 series is gonna be even better,but if ai goes even more mainstream then the prices are gonna go even higher just like during cripto thing

🙏

[gloomy lodge]

just go to cali and steal from 3 stores and you good

[cyan burrow]

I don't think prices will go even higher cause a lot of the AI market are prosumers trying to get in

they will lose an entire huge chunk of demograpics

something they are learning this generation

I think they will introduce many levels of them

maybe even upgrade blocks if theyre smart

imagine buying a 5070 and it's 16gb but you need 24gb, instead of an entirely new card you buy an 8gb vram module

and done

[steady grail]

on confyui, without anything extra, works even on 4 g

but using it without any extra net is worth it?

[meager sun]

ahoy
I've heard that it's preferred to set the image size as 512x512 or 768x768 since most of the models were trained on those, but I'm wondering if there is a specific list of resolutions that are preferred to set? I also tend to just scale to image size from the 256, for example 768x768 or 768x1024 critical to apply those type of rules while choosing the image size?

[steady grail]

depend from the training of the model. read their description, to know the best settings

using controlnet, usually you can encrease resolution, without mutations

[karmic heath]

Oh boy, things have changed since I last checked in on SD. Anyway, here's a test with Latent couple + ControlNet

also WD is actually good now??

[keen hollow]

Here's today's Tifa

[keen hollow]

wow this grid is a gig in size lol... i gotta cut back on my batches 😛

[near elbow]

remember!

and yet another look

[wise pilot]

so reason i messed inpaint recently is by misunderstood about denoising, less = less changes, more = more changes not less = more changes, more = less changes kewk

[native halo]

asuna animation
https://www.youtube.com/watch?v=fEBPhU3z6DE

[steady grail]

wow, cool, what model is this? sd xl?

@gusty palm

[gusty palm]

projectUnrealEngine5_v30,

i just found it