🍥｜anime

[gloomy scroll]

Chibis takeover is going on well I see

[slate night]

Chibi takeover is best takeover

[gloomy scroll]

I have sinned... I did something else that chibi

[slate night]

GASP!

[gloomy scroll]

[young fjord]

what is it?

[gloomy scroll]

At least not chibi

[young fjord]

that looks chibi-ish

[cyan burrow]

the steps of improving an image and gradually upscaling final image is final form

oh and here's the before

another worse upscale with multidiffusion and adetailer controlnet inpainter

Soon you shall meet the next stage in evolution for Clara (C.L.A.R.A: Compliant. Liquidating. Analytical. Refurbished. Android)

[bitter latch]

keeps wanting to face away. grr

[bitter latch]

not sure where i'm going

hmms to himself

tries to make it more anime

hmm ok guess i'll switch to mistoonsanime model

let me in

changes seed last two to 69 and puts a - in front

hits go

hmmm...

thats differen

hehe , used the aloy lora

[bitter latch]

reinstalls horizon zero dawn

ok this is some MTV 90's style strange anime stuff...

Heavy metal

no one else awake ?

last one was interesting

[south sable]

https://tenor.com/view/what-surprised-gif-25739412

[bitter latch]

huh... not I expected but this wasn't it but it's nice......

[gusty palm]

how important is the clip skip setting, i tried a few gens set to 1 or 2 but didnt notice anything drastic

also these look useful?

[gusty palm]

settings>defaults>>view changes/apply arent working

[bitter latch]

o.O

[gusty palm]

beautiful

[bitter latch]

thanks

[gusty palm]

i finally figured out img2img

not enough to explain probably but i used it a few times and results improved

[bitter latch]

ok that got strange but cute

Ok I should be in bed.. nite.. THUD

[gusty palm]

goonight eepy

[supple raptor]

Clip skip 2 can really improve coherency on some models. It's a good setting to have in your quicksettings menu so you can easily change it on the fly. The option in your quicksettings list to add it is called CLIP_stop_at_last_layers.

[gusty palm]

ok thanks. i put vae up in quicksettings too

[supple raptor]

Yep that's another helpful one.

[gusty palm]

how about 'upscaler for img2img' does setting it to none disable upscaling , i would prefer it to use latent

ive been trying 'tiled diffusion' but i dont really know what it does but i figured it might be useful like tiled vae

[supple raptor]

Yeah none will mean no upscaling is done. That can result in resizing artifacts if you use img2img and set the dimensions to a larger size than the input image.

I'm on mobile right now, but I can help you out with tiled diffusion when I'm at my PC tomorrow

[gusty palm]

ok figured as much

guessing nearest or lanscoz is the least memory intensive

ive had blocky artifacts using nearest with hires fix

'just resize (latent upscale)" maybe

[supple raptor]

If you use latent upscaling you'll want to increase denoising a little. I use 0.51 for GAN upscalers and 0.61-0.65 for latent.

[gusty palm]

ok

[supple raptor]

And yeah, for img2img the "just resize (latent upscale)" option is the only way to use latent upscaling. Not sure why he didn't give us the other latent options that Hires fix gets.

[gusty palm]

ESRGAN_4x vs R-ESRGAN 4x+ do they both basically just attempt a clean sharp upscale

[supple raptor]

The end result is similar but R-ESRGAN upscalers use more resources and are quite a bit slower.

[gusty palm]

ah ok

[gusty palm]

wish i could share what im doing lol

ill try to make a sfw thing

[worthy delta]

god bless ai

🙏

[keen hollow]

Seems a little different, new model?

Colors are interesting

[merry tendon]

hey guys, can someone help me

please someone make regex for this: get: " > but if: " " > exists, choose: " " >

[rough oxide]

couple more sdxl outputs, slowly getting a feel for how to prompt more anime-like stuff on it

[rough oxide]

on some models it's more useful than others, for GyozaMix i swap between skip1/skip2 constantly because the outputs are drastically different, for most anime focused models built off of NAI/WD you want clip skip 2 and then just don't need to mess with it

[steady grail]

wow, very cute

[brittle jay]

let me try

though like every time it feels like I have to relearn it

[rough oxide]

couple more after modifying prompt, the lines can be soooo clean sometimes

lol

and i love the variety of what it can do

[merry tendon]

👍

[dapper ore]

What are the most interesting anime models right now?

[nova remnant]

Hello nice anime pictures .

[rough oxide]

just came up with a new one in sdxl not too long ago

[brittle jay]

.*(" " >)|("(?!") >).*

my friend helped me

[merry tendon]

thx man❤️

[merry tendon]

i hate regex man, but gotta use it

[brittle jay]

yeah yeah fr

he just explained to me what he did and why it worked

[merry tendon]

but it was completely worth it

@brittle jay

[brittle jay]

what did it do?

just format the thoughts differently?

[merry tendon]

wait

this is what the ai sees:

sometimes the ai did this ""> so it catches it and makes it into this ">

@brittle jay you there?

[brittle jay]

oh sorry

just went for a sec

[brittle jay]

so it just formats the thoughts?

[merry tendon]

the thing you made fixes it if the ai fuckes up

[brittle jay]

ye

[slate night]

I like meinapastel and the toon/been/seek/feelyou models

[brittle jay]

finally I have nitro back

[rough oxide]

i think SDXL has some of the best potential for anime right now if you learn how to use it. weird pitch i know. but it has like the widest range for expressions & freedom and doesn't get stuck on sameface like many of the other models do. backgrounds will often also be more detailed and there's just soooo much you can do with it.

i still need to do a lot more testing but I already don't wanna go back to 1.5-2.1 based models

oops

i was trying to mention someone

[rough oxide]

idk why the reply didn't work but ^

[rough oxide]

anime-based models can get some expressions down, i had a good anger one in gyoza before, but sdxl seems to be able to exaggerate more when you want really exaggerated stuff

this one in gyoza was pretty good

something about this feels nicer to me than what i can pull off in anime models though

• better clarity on eyes and smaller details without needing to use adetailer

[merry tendon]

what model?

[pale monolith]

nubby's kakigori

[dapper ore]

I don't know how to install SDXL

[steady grail]

sdxl is just a model, place it in model\stabble diffusion

the problem is that they say it needs at least 8gb vram 😦

[slate night]

Not sure what happened here but reloading the UI fixed it

[pale monolith]

looks like latent upscaler?

[slate night]

OH you're totally right, I had latent instead of anime6B

Oopsieee

[merry tendon]

i actually like latent upscaler

[rough oxide]

if your hardware is up to par i can walk you through it the entire way in dms

[slate night]

I like it too, it doesn't work for some of my models though

[pale monolith]

hugepaint8x for me

[jagged wasp]

I've been using 5x latent upscales and I think it helps a ton with details

[rough oxide]

that's nice, i always had problems getting the "looking down, from below" style thing

[jagged wasp]

i do base image --> (1/6 latent upscale --> resample) x 5

[supple raptor]

https://cdn.discordapp.com/emojis/1025170639495102474.gif?size=64&name=uwudance&quality=lossless

[rough oxide]

the way the shoulders are drawn look a bit too generic tho

[merry tendon]

ayo that look nice

[rough oxide]

the thing is, face is great right, and i like way hands r animated there

[merry tendon]

what model btw?

[rough oxide]

but everything below the head

i can tell is AIgen

the face is absolutely amazing there though

put that face on a diff body and it'd be perf

[pale monolith]

hmmm

[supple raptor]

it isn't quite that simple. It uses a different pipeline which means the WebUI you're using needs to support SDXL in order to use it. Auto1111 doesn't support it yet on the main fork. ComfyUI has the best SDXL support, and Vlad's fork of WebUI had some level of support last time I checked.

[rough oxide]

right; which is why i'm more than willing to help someone through the installation process.

[slate night]

It's my own mix, has some Neon Diffusion and some of djz's neon checkpoints in there too

[rough oxide]

i even have workflows for anime-style models with clip skip too to keep everything in one place

[merry tendon]

ok holllyyy still same model?

[rough oxide]

iz gud but i can still see NAI influence, here's one of the ones i did with gyoza where i broke out of the traditional style

[steady grail]

found my new desktop

[pale monolith]

nah i'm going through my best gens at this point xD this one i think is kenshi

[rough oxide]

love the heart eyes though

whenever i look back on my best gens i see a lot of sameface T_T

it's kind of hard to break that NAI influence

[steady grail]

what a weird monstrosity!

[near elbow]

she's angry that you looked at her fingers 🙂

[merry tendon]

can't boot up sd rn, but ill also post some old gen's

[near elbow]

THIS is weird... "an anime style image of a weird monstrosity, beach background"

[rough oxide]

since i already posted a majority of them here i'll do an optional link, what I like about SDXL is how different each gen, even within the same prompt is https://www.pixiv.net/en/artworks/110016204

[rough oxide]

like each image there has a distinctively different quality to it from the other ones

[near elbow]

ok, less weird more cute

[rough oxide]

i love the faces on all your gens, it's always the body that gets me but i can't do much better with the models i use for my own taste xD

[near elbow]

demon cats

[near elbow]

meeeeow?

[rough oxide]

it's almost there, til it gets to the hand. but overall that is nice

[near elbow]

double mecha head

[rough oxide]

i may be too picky for my own good xD

[pale monolith]

im not done yet! >:D

[rough oxide]

lol

[near elbow]

mecha! gundam!

[rough oxide]

another example from my pixiv

that one is from gyozamixv5 again iirc

[pale monolith]

alright I'm bringin my pixiv out too then

[rough oxide]

always liked the outputs gyoza coule pull off

aiiiiight

i love that

that is really good

[merry tendon]

i remember when you first posted that in here

[rough oxide]

that is perfection <3

lol

[pale monolith]

im almost done with my next one actually

[merry tendon]

can you show us?

[rough oxide]

love the concept of that

[rough oxide]

oooo

[rough oxide]

could use some work on the eyes but that looks great as a concept

[merry tendon]

those yandere ones were some testing images i did for my trained lora(so didnt knew what settings were best for it)

[near elbow]

close but no cigar 🙂

[rough oxide]

using some kinda cyberpunk lora/embedding?

[rough oxide]

the fade to dark on one side there is so good

[near elbow]

no, just prompting

[rough oxide]

oh, was asking cuz the style of word was pretty good

despite not being able to fully get there xD

[merry tendon]

this is just fukin scary

[rough oxide]

i feel like this one is peak

[merry tendon]

yep, i am posting from new to old

this one is kinda bad, but also has a nice vibe to it

[near elbow]

bring in the madness

[rough oxide]

reminds me of the style i'd get in wd1.3 but more detail on hair

[merry tendon]

these 2 are my fav's

[pale monolith]

close this was ouroboros my 1.3 dreambooth model xD

[rough oxide]

nvm 1.3 was a bit worse than i remember

o

lol

close-ish then

xD

[rough oxide]

1.3 could do the most poggers bubbles

i still dunno if i can recreate that kind of bubble effect again lol

[merry tendon]

love alex chow style

[near elbow]

last one for now - cu

[rough oxide]

then back in like WD1.4 days i'd try to focus on everything but the face xD

...or was that still 1.3

[nova remnant]

hi, im just starting to have fun with SD, could someone please advise me how to get result like this? idk which checkpoint will be the best

[rough oxide]

hmm

xD

1girl, red theme, horror theme, black hair, red eyes

comes to mind

[rough oxide]

as an initial skeleton

[nova remnant]

hmm, and whichever anime model from civitai?

[rough oxide]

bunnyspoon or kakigori might be able to get close-ish to that style

tried to run a similar prompt thru xl just now lol

[nova remnant]

oook thank you ill check

[rough oxide]

might need to describe background

and use some kind of aura too

[pale monolith]

aurora is flexible and prefers anime style but is more semi-realistic, amnesia is highly lora compatable but its base style is more realistic

[rough oxide]

some more xl tests lol

for that kind of style

hmm

[merry tendon]

this one is just

[pale monolith]

mmm amnesia :3 also spooky bloody skull

[rough oxide]

o

i pushed it

really hard into that style

lol

maybe too much? xD

[gritty spruce]

sdxl 0.9 via comfyUI and then feeding those to deforum

[rough oxide]

@nova remnant well i could tell how you to pull this off in SDXL xD

[rough oxide]

i may have pushed it too far tho

xD

[merry tendon]

ok ill stop spamming you guys now xd

[rough oxide]

pulled back a little on the theme

[rough oxide]

i don't think anyone minds lol

but uhh

it might be better if you post multiple in one thing like that ^

instead of like 10 individuals

it'll group em up together

[merry tendon]

shit, forgot about that

sry

[rough oxide]

u can do up to 10 at once, and it groups them up nicely

[merry tendon]

ik ik i just forgot

[rough oxide]

@nova remnant

from my sdxl workflow xD

it'll be totally different on an anime model

[pale monolith]

[nova remnant]

wow that's epic

thx @pale monolith I will definitely check amnesia

[pale monolith]

❤️

[rough oxide]

was the best output so far i think. also uhhhh, kakigori has aurora in the mix iirc and does better styling (imo)

than aurora by itself anyways

[rough oxide]

loveee this one

[pale monolith]

amnesia showcase :3

[merry tendon]

my pfp

[pale monolith]

"1boy, handsome, male focus:1.2, portrait, from side, young man, hyper-realistic, painting (medium), <lora:[LoConLoRA] pseudo-daylight偽日光 Concept_style:1> lora:DiscoElysiumStyle_v1:0.57 lora:Aetherion_style:0.33 lora:cutesexyrobuttsStyle_cutesexyrobuttsAnything:0.45 lora:Bloodborne-style:0.35"
it likes loras

[merry tendon]

last pics for now, gotta go eat something

[rough oxide]

2,7,8 all so good.

that's wut i mean tho, in like 2 and 7 the hands/bodies/etc actually look stylized

for that specific kind of aesthetic

it doesn't look generic, those look really good

[rough oxide]

lol and then in that one i like everything but the face xD

[pale monolith]

xD

[rough oxide]

but really in the batch u posted, i loooove 2 7 and 8

[pale monolith]

i really like the massakama lora too

[rough oxide]

that's good

very good :3

left eye could use fixing but the concept is good

any plans to try and make an amnesia-type model off sdxl once that's out? ;D

lol

[rough oxide]

well

gotta wait for the main anime models to come out first i suppose before anyone can do work like that xD

hopefully some big anime model project is done on sdxl after 1.0

[pale monolith]

doubt, amnesia is a mix using my older mix, and i don't have any of the mixing values plus its like 20 odd models mashed together

[rough oxide]

ah

i may have to try it sometime then regardless

[rough oxide]

love that style

almost looks like it could be a frame of like a bowsette cgi film xD

[pale monolith]

xD

[rough oxide]

oooooo slime girl is really good

how was that done

is that still amnesia model?

[pale monolith]

uhm no i think its aurora, and a really good slimegirl lora xD

[rough oxide]

o

ic ic

yes, i concur

very good slimegirl lora

that is peak

[pale monolith]

here is copied info pasted onto amnesia tho

[rough oxide]

used to play vrchat with a friend who always used a slimegirl model

reminds me of that xD

hmmm. probs justr needs more gens on random seeds

cuz it's like almost there

[pale monolith]

me too megu me too

so sleepy

too many call backs this weekend

[rough oxide]

was still playing with the prompt from earlier, got lots of nice results imo

[gloomy scroll]

[gloomy scroll]

I tried doing something different today but here's chibi

Making prompts is hard

[cyan burrow]

have him stand in front of a full on eclipse with a horizon of the suns corona

[gritty sundial]

Yes, I am going to do

[cyan burrow]

at the same time do this "ring light on eyes" thing and "dim lighting in foreground" with a lora or locon, lycoris with offset noise

[gritty sundial]

like this?

[cyan burrow]

mmm no like a big ass ecplise just behind him. You know how optical illusions can in some places make the sun or the moon over the horizon appear bigger than mountains?

@gritty sundial

here's how it looks in some places when you're in the right location during a total eclipse

a very rare thing to have it be 100%

[gritty sundial]

yes, I understand but I can't do it, it's coming out like this

the prompt I put was this "A boy, with the universe in his hands, with his eyes fixed on the universe, his eyes full of tears, ((total eclipse in the background)) with a solar corona horizon, ring of light in his eyes, dim lighting in the foreground" see if you can change and fix it

[cyan burrow]

A total eclipse, large in centered and symmetrical composition, directly behind A boy, ash white hair color, medium short hair, cobalt blue eyes, gaze focused on a spheroid in his hands, containing the galaxies of the universe, 2.5d Digital Illustration, Digital Art, Anime Art Style, Dim Lighting, Dark in foreground, back lighting, centered angle shot, HDR, 8k, Octane Render

try that @gritty sundial

or something like that

I dunno what model you use so I can't really know what language it speaks

[gritty sundial]

awpainting_v11

[cyan burrow]

I will be shameless here and link my model https://civitai.com/models/6209/dark-gemini-v3

[cyan burrow]

on the side menu there's a pruned version

and a fp16

[cyan burrow]

that became a moon eclipse

A total solar eclipse, large in centered and symmetrical composition, directly behind A boy, ash white hair color, medium short hair, cobalt blue eyes, gaze focused on a spheroid in his hands, containing the galaxies of the universe, 2.5d Digital Illustration, Digital Art, Anime Art Style, Dim Lighting, Dark in foreground, back lighting, centered angle shot, HDR, 8k, Octane Render

that should be it then

[gritty sundial]

I will increase the cfg

[cyan burrow]

no keep cfg at 7 in automatic

try using my model, it understands natural language better and is very versatile

https://civitai.com/api/download/models/49207?type=Model&format=SafeTensor&size=full&fp=fp16 for the smallest size fp16 version

https://civitai.com/models/15879/loconlora-offset-noise and use this lora @gritty sundial

(not mine)

it uses offset noise to create more realistic shadows and darkness and stuff instead of having characters be lit in pitch black paces with no sources of light

[pale monolith]

hmmm

[gritty sundial]

it's downloading, when I download it I'll do it again

[cyan burrow]

that's certainly unique

[cyan burrow]

purdy

arf

I fucked up so now I have to reinstall the entire venv

do not upgrade pip to the latest version

[gritty sundial]

@cyan burrow In your model it was like this, I liked it, but did it come out the way you wanted?

[cyan burrow]

mm no but in my model it needs more natural language so hold on let me write a prompt

[gritty sundial]

ok

[cyan burrow]

(a boy holding a spheroid with a universe full of galaxies:1.15), his gaze rests on the spheroid as (he stands with his back to a majestic total solar eclipse:1.149) BREAK the horizon of the eclipse shows a bright display of solar flares BREAK 2.5d Digital Illustration, (Digital Art, Sci-Fi Aesthetic, Anime Art Style:1.5) BREAK (Ultra-Wide Angle, Double-Exposure, High Contrast, Sunlight, Hard Lighting:1.12), Ray Tracing Global Illumination, Screen Space Reflections I have no idea if this will work

gonna try it myself, don't forget the lora, I'll see what negatives I put in

testing it myself now

yeah ok too much focus on the spheroid

damn it, he wont not face the eclipse

@gritty sundial I'll try and get that eclipse working

I mean I got these but theyre not exactly....an eclipse

[gritty sundial]

I give up man

the business does not do what we want

[steady grail]

what's wrong with plms sampling? i had to put 150 steps to not get images with noice!

[supple raptor]

plms is one of, if not the first sampler. It's very, very inefficient.

[steady grail]

add some more very and we will agree lol

ty

[gritty sundial]

Background

[cyan burrow]

wtf just happened here?

[south sable]

a toast to the end of the world?

[gusty palm]

settings this lora weight to negative numbers makes the highlighter thingy change to normal colours, does that mean its not working or is it just like a recognition bug

[south sable]

SDXL 0.9 + 1.5

[gusty palm]

ive been doing the thing where u put the result into img2img and upscale by like .2

2 or 3 times, im not sure if thats the same as hires fix but i like what it does

[jagged wasp]

kinda has a poloroid/instax mix with pastel feel

I'm doing the same but in comfyUI I upscale the latent by 1/6 then run another 30 steps and it adds details each time

[gusty palm]

that sounds like what im doing i think

[jagged wasp]

https://up.reselltek.com/u/kwRzGd.png

[gusty palm]

im gonna make a gif with the progress probably tonight

woah

is that comfyui

the nodes

[jagged wasp]

the final refinement i do a 4x using uniscalev2 so my final output is 6624x4960

https://up.reselltek.com/u/uqB6HG.png

yup spaghetti central whenever i try to make a workflow in it

[gusty palm]

lol

is that an addon for auto111 or a alternative ui

[jagged wasp]

different kind of UI called comfyUI

https://up.reselltek.com/u/EV499C.png it's node based but I think it does a better job and faster than auto for this incremental hi-res i got going

[gusty palm]

looks like blender

[jagged wasp]

ya the node tree layout for sure

[gusty palm]

i might give it a try

[jagged wasp]

can't hurt you can point it to the auto1111 folders so you don't have to re-download any of the models/loras

[gusty palm]

ah that was my main concern

is the install similar to auto11

[jagged wasp]

on windows you just download a zip and run the main.py, there's a config file you can point to the automatic folders and that's it

[gusty palm]

neat

does lower denoising stregnth preserve less of the original image

[jagged wasp]

the extra detail steps I think make a big difference from anything i was getting before

lower denoise means preserve more of the original noise so the higher you set it the more the image will change

[gusty palm]

ahh

and i agree i love how it just sort of subtly enhances the details

i tried playing with the lora:add_details:-0.5 thing in the img2img prompts but im not sure if im convinced its working with negative numbers to reduce details

[jagged wasp]

i think it's 0-2.0

so 0.2 would be less details

i have it set to 1.5 on these

[gusty palm]

so less than 1 but above 0 for reduce?

[jagged wasp]

it's what i see but each one of my refinement steps adds more of the effect

but anytime i do < 1 it is definately softer

[gusty palm]

ok great ill try some more

so far 1200x1200 is my limit since it takes like 3 minutes

not feeling like im even close to oom though

[jagged wasp]

what gpu

[gusty palm]

1060 6gb

[jagged wasp]

not terrible times for that size

[gusty palm]

im down to about 40 seconds for the initial images so im happy with it

[supple raptor]

0 is no effect, if you want the opposite of what the LoRA is for, you'd use negative values. Usually lora:example:-1.0 is about as far as you would want to go, but you can always try going lower.

[gusty palm]

ok, it appears its just a visual thing in the prompt keyword window with a negative value but i think its still working

[merry tendon]

👍

[gusty palm]

SD1.5-WD1.3-LD-Trinart-Merged gave me a bedroom with no character the other day randomly so i wanted to try to make some more

[gusty palm]

and now i need more

[worthy delta]

Dayum they full on look like sketches

Also shirtless dudes allowed here or nah

[supple raptor]

Yeah shirtless dudes are fine, just be mindful of how low their waistbands are.

[worthy delta]

thank you!

[merry tendon]

whyyy https://www.youtube.com/watch?v=-Bdb2KOb_zI

[steady grail]

pretty useless thing to do. as people can crack app today, they will be able to crack metadata tomorrow..

[merry tendon]

it is still annoying

[steady grail]

everything they will do in future will be pretty annoying. they intend to run windows i cloud, for example

[merry tendon]

i dont belive that last argument

lemme see some proof

[steady grail]

free to look for yourself 🙂

[worthy delta]

https://tenor.com/view/metal-gear-rising-metal-gear-rising-revengeance-raiden-senator-armstrong-source-gif-25038929

[steady grail]

because it's easy to find them and i'm lazy lol

[worthy delta]

is skimpy clothing allowed on women? @supple raptor

[nova remnant]

Hello very cute anime.

[nova remnant]

Hi, very nice anime picture.

[steady grail]

how did you do this?

[gloomy scroll]

This one has pnginfo

[steady grail]

🥰 🥰 🥰

[steady grail]

wtf is this , //, :D in the prompt? lol

[tiny rivet]

Meet Vamie

[lilac hazel]

https://github.com/guoyww/AnimateDiff

[elfin bolt]

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

[cyan burrow]

really the clinical term for peen is blocked to say here?

oh well, on clipdrop I drew a johnson and it ended up on a wanted poster, ironic

yes I'm still 12, even after aging into my 30s

[south sable]

[south sable]

dang, not working with my lowly RTX4000

[proud hare]

Drims

[cyan burrow]

rtx4000-what?

[south sable]

Quadro RTX-4000

[real nexus]

Wow!

That's the first time I see a perfect pencil sketch style

[supple raptor]

I am loving this lora https://civitai.com/models/110244?modelVersionId=118849

[real nexus]

It is really good

[south sable]

https://tenor.com/view/steal-anime-k-on-ritsu-tainaka-mio-akiyama-gif-24287493

[supple raptor]

the mans just the lora I found

[supple raptor]

apparently this was supposed to be ainz ooal gown. I think the ai may need to check the source material again.

[drowsy dove]

Starting the Llama2 prompting

[cyan burrow]

I shall now use my powers for evil, to spread chaos across the world with my prompts

I will become the John Kramer of AI Imaging

[cyan burrow]

Cyberpunk, the most beautiful places on earth unless you're not rich

[queen void]

I love the artstyle

[supple raptor]

Thank you! I've been really enjoying this LoRA

[slate narwhal]

And so WDXL 0.9 releases... I wish I could be at home to test it

[worthy delta]

ayo wut its out??

[slate narwhal]

Haru just released Waifu Diffusion trained on SDXL 0.9 (as a test), SAI gave him permission

[worthy delta]

ayyyeee thats good to hear

[slate narwhal]

https://huggingface.co/hakurei/waifu-diffusion-xl

[drowsy dove]

llama2, sdxl, I'm hitting the data cap this month

[cyan burrow]

You have a datacap?

[drowsy dove]

100 GB on a 4G modem

[cyan burrow]

oh

I don't use my 4g for much, mostly at home anyway

[gusty palm]

imagine in ten years where rendering happens in realtime and u just flow through settings and seeds like a video game

[cyan burrow]

You can do that already with a really high vram and regular ram, nvme ssd etc and a good plugin in photoshop. I think Gyre AI can do things almost in realtime

if you use an API to a server

or local powerhouse machines

[gusty palm]

oh true

[cyan burrow]

Gyre AI has some sick features

[gusty palm]

that would require some serious hardware , i was thinking about how far raytracing has come in a half decade

[cyan burrow]

an AI healing brush, AI paint brush, outpaint and inpaint boxes

[gusty palm]

is it similar for krita or glimpse

[cyan burrow]

you can use gyre ai together with runpod, that I think would be extremely fast

https://gyre.ai/

It has a 3d mode for it so you can move around ai generation layers in 3d

and set up nice scenes

[gusty palm]

intense

i just finished the last season of good doctor ><

[steady grail]

cool

[bitter latch]

wll.. that didn't work out . almost did

wtf is with the eyes?

[cyan burrow]

probably poor merging on the lower lanes

[worthy delta]

I got a question is sdxl 0.9 going to become the new standard instead of 1.5?

[bitter latch]

is random

ok i need idea's

[gusty palm]

love that style

[bitter latch]

just changed to model mistoonanime

ok maybe not all but it helps

o...k then

[gusty palm]

any good models for mimes or like the white face paint type look

[bitter latch]

hmm not sure

[bitter latch]

there's a lora called bonbon

now let's try it with the bonbon lora

[gusty palm]

ooooo

https://civitai.com/models/23540?modelVersionId=28116

this will be nice ty

[bitter latch]

welcome

now... lets make it chibi!

[digital harness]

Still VERY new to this all but some things I generated I am decently happy with, still learning to do it correctly and how to prompt things correctly

[bitter latch]

looks good

[digital harness]

[supple raptor]

0.9 definitely wont. Once sdxl 1.0 is officially released it probably (hopefully) will overtime, but I wouldn't expect the transition to be quick. We've got ~9 months worth of finetuning our current models and other tools on SD1.5 based models so it's going to take some work from the community to get SDXL up to par in my opinion.

[cyan burrow]

and that's assuming they can optimize it for a large demographic of users

if the userbase is small, it's hard to justify time investment

[supple raptor]

Plenty of people have 8+GB of vram. Progress, might be slower, but there will be more than enough people able to run it for it to be worthwhile to work on it.

[cyan burrow]

yes but thats what I meant

8gb

that's what most people will have now

followed by 12

or 16?

[worthy delta]

the most used gpu on steam i think is the 3060

which is normally 12

or 8

itll take some time for people to get 16gb in their gpus fir a while

[digital harness]

Running 16 myself now but shafted by the use of an AMD GPU

[worthy delta]

if ai art is alr this good just imagine it in a good year or two

[cyan burrow]

yeah but the 3060 is slower even with 12 vs the 3070 or 3070 ti 8gb, wellesp 3070 ti

[digital harness]

I guess it is a question of datasize you can cram and a bit less for speed

Speed being important aswell

[cyan burrow]

it's possible we'll plateau soon, or reach a legal threshold where only groups in certain countries or hiding their location can continue

[worthy delta]

i dont think itll be restricted to certain groups tbh

if it did it would get leaked eventually

plateau is more likely

[digital harness]

Most likely due to reaching the limit of consumer hardware

[cyan burrow]

not restricted to them, I mean only they can develop

cause if someone is openly developing a popular AI that's illegal in their country

well, I don't need to explain

[worthy delta]

someone would find a way to develop it outside of the certain few

its tricky to actually control this type of technology

it is already very wide spread

[gusty palm]

i hate the corporate "user guidelines" on ai thats why i got into this

safety sucks all the fun and usefullness out of it

[worthy delta]

there should be some restrictions on ai

but the corpo's just want the technology to themselves so they can screw over artists

[worthy delta]

art like this is free

and ai art shouldnt be locked away from the people

[south sable]

https://tenor.com/view/crowd-applause-applaudissement-foule-claps-gif-18331500

[gusty palm]

not jsut artists imo

[worthy delta]

yeah thats why there a writers strike

[gusty palm]

love them for that

anyone use the unspeakable horrors inversions ~ wondering if theyre useful

it adds like 100 tokens

[wet elm]

can anyone help with this error? i lost my commanline args and cant start the webui

fatal: bad config line 1 in file C:/Users/User/.gitconfig
fatal: bad config line 1 in file C:/Users/User/.gitconfig
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: ## 1.4.1
Commit hash: <none>
Traceback (most recent call last):

[bitter latch]

Star FRREK!

mistoonanime model :

[gusty palm]

watching strange new worlds rn

[gusty palm]

what kindof tags are u using for the star trek stuff

[bitter latch]

star trek

[bitter latch]

hmm ok that came out interestig

[gusty palm]

thats gooood

i like the rabbit bibrator

or maybe its just a weird thumb

[magic wave]

Two LoRAs trained, may retrain both, already retraining the 2nd one

[verbal frigate]

Rev 1.2.2 is rlly horny

Idk why

I asked for a cat with a collar (sus warning) original was even more nsfw but i fixed it a bit, STILL NOT A CAT

[digital harness]

Trying to generate a cat challenge (impossible)

[magic wave]

Cat™️

[gusty palm]

close nuf

do any chat models work for auto

or is chat a whole different beast

[magic wave]

You gotta use Oobabooga for chat AI

https://github.com/oobabooga/text-generation-webui

[livid shuttle]

Hello everyone, I'd like to make the Pharaoh Atem (Yami-yugi) from Yu-Gi-Oh in different poses and places in ancient Egypt (or Yami in Japan), I can't find the right prompt for a nice result, could someone help me? Thank you very much! 🙏🏻

[elfin bolt]

train a lora

or use this https://civitai.com/models/93564

[livid shuttle]

Thank you 🙏🏻wow they are amazing, beautiful, I love 💞. Can we use a Lora on the discord? Or only with Stable diffusion? I don't know how to use it.

[elfin bolt]

Hey, download the lora and put it on your local sd install on models/Lora

if you're using automatic111, you can enter the lora in prompts and its activation prompts (Atem, 1boy, egyptian clothes, eye of horus, jewelry, dark-skinned male, multicolored hair, spiked hair, purple eyes)

[livid shuttle]

Thank you for your help you are so sweet.

[haughty mica]

Some of my favourite anime background creations 🙂

[worthy delta]

Are loras easy to train?

[gloomy scroll]

Reminder why it's called STABLE diffusion

[nova remnant]

chibi invasion

if that's what ww3 will look like, no need to draft me I'm enlisting

[verbal frigate]

Ikr

But why is stable diffusion so sus

Like was it trained on an nsfw database?

[verbal frigate]

Nice third leg

My fav generation so far

[steady grail]

with such a cute armada, i would chose to support war .

[steady grail]

from now on we should call this server, chibi diffusion lol

[bitter latch]

hehe

hmm she has no face...

[steady grail]

i'm about to birth a weird result

but looks fascinating

[bitter latch]

ok...

waits

[steady grail]

just 5%

[cyan burrow]

There's something about that art style that kind of weirds me out. Even though it looks nice and all

[brittle rampart]

Maybe. but I really like this model.

[steady grail]

tey looks like on drugs lol

but true, everything else is very cool

[brittle rampart]

they just look at the world like a child with wide eyes.

[steady grail]

wow

legs are weird, but the complex is attractive.
maybe i will fix her legs later

[brittle rampart]

btw... for exmple: little fly agaric.. But i liked this is model!

[steady grail]

🤣

i don't see why, but i delete it if makes you feel unfonfortable

[bitter latch]

just trying to be helpful and help you NOT get in trouble

[steady grail]

yeah i know, maybe my tastes are just a little off, or naive

[bitter latch]

I mean it was interesting but what she was holding .. ummm

[steady grail]

a balloon?

the prompt was hearth

[bitter latch]

that was no balloon

[steady grail]

metadata are still inside

[bitter latch]

hmm I put park and get street half the time. what?

[steady grail]

same metadata of this https://cdn.discordapp.com/attachments/1091193032273043516/1130830454061273179/20230126135562-2392200248.png

too bad i don't get the same organge tonality 😦

guys, someone knows how do i get this organge light? i love it so much

[regal pivot]

maybe it's a VAE thing..?? ,, there are others more or less saturated ...@steady grail

[steady grail]

maybe...

it's just an impression, or euler a is more performing when doing chibi?

[regal pivot]

@steady grail Which one do you use ?

[steady grail]

now i set it on auto, before was on blessed2

[slate night]

orange lighting in prompt

or like

Actually yeah I'd be worried anything else would probably just give you the fruit lol

[steady grail]

lol

[regal pivot]

@steady grail@slate night
maybe I'm wrong but I have the impression that the metadata corresponds to some kind of Img2Img, upscaling process... but the image was generated with a slightly different prompt or with a Lora .. I don't have space to change models and try everything the same...

[steady grail]

yeah, had the same impression.
the prompt i'm pretty sure is accurate, but the bg is the result of an edit

[slate night]

Oooh gotcha gotcha

[steady grail]

for this model, this is an heart

[slate night]

Heart... kidney bean... same difference to AI lol

[steady grail]

ehehehe

[regal pivot]

I can't even get a cute chibi

[steady grail]

my mission became finding the right formula to farm chibi 😎

i'll became the chibi master!

[slate night]

Hindellsle

[steady grail]

is a real publication?

[regal pivot]

@steady grail me?

[steady grail]

yep

oh

[regal pivot]

@steady grail ... it is only RNG ....but ..add anime_cover ? (i not remember the tag).. for call it

[steady grail]

some of the prompts of this image are pretty obscure.
for example, m_akram, m_ouroboros, , //, :D

pretty cute

abissorangemix is pretty cute

oh...

interesting

[regal pivot]

@steady grail bye bye ..
I am deleting those captures above ..HF

[steady grail]

see ya

chibi is justice 😎

[supple raptor]

m_akram is the activation token from this model: https://huggingface.co/flamesbob/akaramModel/tree/main

m_ouroboros is the activation token from this model: https://huggingface.co/Eppinette/Ouroboros

*//* isn't actually a tag, I'm not sure what Vil going for with that one.

:D is an open faced smile like the emoji

The image you're trying to replicate has the metadata from an Ultimate SD upscale pass, which means the image was previously generated (likely on a different model, that contained the two models that I linked) and then upscaled using AOM2.

[steady grail]

so, i'm missing a bunch of metadata. the prompt is incomplete

ty bro

[supple raptor]

The prompt is most likely complete, but you're missing the parameters for the original image.

[gloomy scroll]

*//* is just from some japanese prompt

[steady grail]

i'm learning so much studing this image

[gloomy scroll]

Maybe I should have checked what that image even contains

[steady grail]

ehehehe

[supple raptor]

You may have a challenger for the title of Chibi master, but look at the bright side.... MOAR CHIBIS!

[steady grail]

hell yeah!

[gloomy scroll]

Also that Miku image was done with AOM2

not just upscaled

[supple raptor]

Ah, I think I know what makes it look different from typical AOM2.

[bitter latch]

runs in circles and falls over

[gloomy scroll]

This one is the actual original

[steady grail]

cool, ty

[gloomy scroll]

Sry this one

[bitter latch]

ok

[gloomy scroll]

@steady grail Other versions didn't have loras

[steady grail]

ohhhhh

[cyan burrow]

Silly Chibi, that's not how you sweep floors

[supple raptor]

I was right, was pretty sure I could see those two loras in it XD

[steady grail]

oh, an addnet lora mix

i usually just add lora to the prompt

[gloomy scroll]

That wasn't always supported

[steady grail]

oh

interesting

[gloomy scroll]

And I think there was some difference between lora in prompt and addnet

[steady grail]

oh...

[twilit copper]

i really like this style

[bitter latch]

img2img with this