🍥｜anime

[wet elm]

ty

[nova remnant]

[nova remnant]

@green temple

she doesnt want to study

cx

[keen hollow]

I'd get food from Dudette MCgee's...

[nova remnant]

ugly

[steady grail]

wow, these results are cute

[green temple]

😍🥺

[nova remnant]

@green temple

Here it comes

smol

now is time for biggies

👍

[queen void]

Yes and no :D

[gloomy scroll]

Yesn't

[queen void]

Indeed

[nova remnant]

Grabby

[gloomy scroll]

The moment person poking realized their fate

[lyric zealot]

still went for it

[gloomy scroll]

Worth

[gloomy scroll]

Tugs your shirt for more cheek rubs

[gloomy scroll]

Nom

[lyric zealot]

[green temple]

best lora i found so far

[gloomy scroll]

Your destiny has been decided

[green temple]

remind me of that zafina line from tekken 7 "your fate has already decided"

[green temple]

🔥

[green temple]

2 moons noooo

[nova remnant]

Not even 3

[gloomy scroll]

I may know the reason

[nova remnant]

I'll eat

[green temple]

i feel it`s cursed

dunno why

[unkempt cape]

crunch crunch

[bitter latch]

ok enough of that

[unkempt cape]

Do you need to know how to draw to controlnet a manga panel?

i saw an ai manga thats like easily the best i've ever seen. style looks like animagine but the panels actually look like panels i was going "To controlnet this stuff you must like know how to draw rite???"

[keen hollow]

evenin'

[vital raptor]

you can just google for images to use as references, or you can draw stick figures and use those

kind of the point of controlnet is that you can use just about anything to control the output

[bitter latch]

wd

[bitter latch]

I LOVE THIS STYLE

ooooh

Miku has entered TRON

[bitter latch]

sure is quiet....

hmm I should make these into a manga 😄

[supple raptor]

damn..... what base resolution did you use for this? or did you outpaint to get the extreme aspect ratio?

[near spruce]

the spin quantum number of the viewer is 1/20

[acoustic locust]

that must've taken 20 years to gen lmao

[acoustic locust]

25000x512

[nova remnant]

all that resolution and no butts to see

[green temple]

cool

[nova remnant]

I will

Grabby on all 3

[zinc flax]

https://tenor.com/view/aris-alice-blue-blue-archive-archive-gif-12082873086428133110

[nova remnant]

What does this mean

Cx

[zinc flax]

no grab

[nova remnant]

[visual meteor]

depressed vocaloid collection

[merry tendon]

ohh shit

you guys inspired me to do AI stuff actively btw, not sd but AI in general

thanks for that

[magic wave]

I think this doesn't count as anime, but Mikooo

Trying out a realistic model that seems fun

[queen void]

🫡

[magic wave]

Hai Nerf

[nova remnant]

grabby

[gloomy scroll]

tugs

[pale current]

Hello! The images you drew are very nice. Can you tell me your drawings and prompt words? Thank you!😅

[gloomy lodge]

just made a really thin image then did sd upscale works really nice

[green temple]

this is a lora

[green temple]

i used 2 loras

https://civitai.com/models/28742/light-and-silhouette lora 1

[green temple]

https://civitai.com/models/26200/ruins-and-light lora 2

[green temple]

what model ?

[nova remnant]

vendo pony realism

something like that

skeleton ish

@green temple smol

[magic wave]

Yeee that one

I made myself a cool wallpaper

Though the background of that pic was done by hand

[keen hollow]

you did the skull guy thingy?

[magic wave]

What skull guy thingy? :o

If you mean the video thats like just a video, I did my wallpaper :o

[keen hollow]

ah ok ok I had to click in 😄

[magic wave]

Yeah xD

[nova remnant]

You'll attract spirits to your house with that background cx

o:

music video

im gonna listen to it

[magic wave]

Is neat :o

[nova remnant]

its that genre noooo

[magic wave]

I mean I don't listen to it too often, but I like this one, saw it in a TheRussianBadger short :o

[nova remnant]

I mean

[nova remnant]

they are just like too much of the same thing I think

[magic wave]

They are, I don't really like em either, I just like some here and there :o

[nova remnant]

o:

[magic wave]

Yee :o

[visual meteor]

why are swords so hard 😖

[ornate echo]

Because if they were soft, they wouldn't cut stuff.

[visual meteor]

[bitter latch]

being chased by jarjar binks.. no....

[green temple]

cool

[green temple]

very similar to Hyakkimaru from dororo

[nova remnant]

Grabby

ears

[green temple]

[green temple]

elf

[nova remnant]

elf grabby ears

[green temple]

😢

[nova remnant]

what's wrong with elf

[green temple]

she will eat me at night

[green temple]

cool

[nova remnant]

Have you been playing dwarf fortress or something

Why would they eat you

[green temple]

it`s known that they eat people😢

[acoustic locust]

dwarf fortress mentioned

[nova remnant]

Elves in dwarf fortress are questionable

[nova remnant]

Ugly eyes but squishy

[ivory pebble]

They look nice tho wtf

[nova remnant]

If it was a horror movie yeah

[ivory pebble]

She's supposed to be scary so I guess it works

[nova remnant]

👍

I like the big art

On the walls and stuff

Very shiny

[slate narwhal]

yeah, backgrounds are nice with these, unusual with pony

[nova remnant]

Is it tpony

Doesn't look very pony yea

[slate narwhal]

Yes, TponyNAI

[nova remnant]

No big art in there

[gloomy scroll]

Massive art

[green temple]

nice

[bitter latch]

hmm little burnt

hmm interesting

[green temple]

wow

[bitter latch]

yep I love this effect 😄

[green temple]

what is it called , and do you use a lora with it ?

[bitter latch]

actress Golden filigree plasma , more details lora:Silhouette:1 Silhouette, light and shadows, Smokey overlay, ( light rays in a dark room:1) soft pastel color and shading

the acctress part is just a wildcard

[green temple]

oh same as my lora

i will try the prompt 👍

[green temple]

👍

[bitter latch]

wanna really mix it up, try adding holographic before the silouet

or holographic vibrant pastel colors

[green temple]

the view of my images are always from side, i tried to change the angle with the lora but it doesn`t work

[bitter latch]

hmm try , facing viewer, looking at viewer or such

[green temple]

tried this but didn`t work , maybe i should go for less weight for the lora

[bitter latch]

welop time to goto work. later

[green temple]

🤣

[opal cradle]

Make something for me I dare you

[green temple]

like what ?

[opal cradle]

Well, first explain to me, how generating anything via Discord works, since I joined just a while ago

[green temple]

i don`t generate via discord🌚

[opal cradle]

Well these were generated by me

[green temple]

nice

[nova remnant]

[nova remnant]

first one is the most smol

making shirts like that on ai would probably confuse people

"What language is that ?"

Lmao

[green temple]

[opal cradle]

Your real pros

[terse ore]

Made this render in blender:

Put it through img2img. Any way to stop those artifacts in the background coming in? Also the background is supposed to be one thing but it changes up the material construction

[green temple]

how`s the drawing line ?

[nova remnant]

Its perfect

no need for img2img

[green temple]

what do you think about it

[green temple]

indeed

[nova remnant]

not really, img2img will change the image to what the model think it should be

only thing I can think of is some controlnet tiling stuff

[terse ore]

Right, I rely on that so only do style prompts for img2img. The background in the original render looks like rough stone but in the output it's smooth and polished

I used canny and depth together

[nova remnant]

by drawing line do you mean the style

or maybe just the outlines

[green temple]

yeah just the outlines

[nova remnant]

looks cute

[dire bridge]

is that a box of cholate?!

lol

[green temple]

🤣

[nova remnant]

loads up staff like a shotgun

She stays strapped

She knows what's up

Also hey caaattt

[nova remnant]

You been gone

I've been playing badurs gate LOL

It had me in a chokehold, its such a good game 😩

After all I heard about it

Made me want to ignore the game completely even tho I liked playing d&d

Dawg what did you hear

Mods will bonk me if I say

You can imagine

I have so many mods for it

I'm tired of having to change my settings to what I like everytime I open forge

Wtf do I do

Ik there's a tab in settings to change it but I can't find it

[nova remnant]

Forget bag game and generate elves

👍

I love the wat your eyes come out 😭 I gotta get more pony stuff

Try making the lora

:<

I don't have the buzz for it

I still have the stuff you gave me to make it I just haven't gotten around to making a new account to get buzz

Is it more than 500?

I have 0

💀

What did you spend on

Idr

Bath time

I'll try it rn

Let me make a new account for the buzz

gl

Get back here mf, youre helping 😭

Is this is concept or style

Should be style

They have a description for each

It's a very simple Lora

read

[nova remnant]

i refuse

[nova remnant]

And get the wd tagger extension

The what

To do the thing

I'm so confused

The site will ask for a text file with the tags for each image

That extensions does it

Dawg

You're saying words but I'm not understanding

Explain to me like I forgot everything AND like im dumb cuz I forgot and I'm dumb

[nova remnant]

frieren <3

[bitter latch]

"you think you can escape me?"

[terse ore]

What prompts do you use to make flat color stuff with SDXL?

[visual meteor]

Goth lolita

[nova remnant]

weird model x3

[nova remnant]

it wasn't me

[visual meteor]

What model are you using to get hands that aren't horrific?

[rough oxide]

nai diffusion v3, but also i always cherrypick and/or fix hands if there's something wrong with them. this one had messed up hands before i inpainted for instance

the ecclesia one genned with a messed up pov hand in the top left actually

but that was an easy fix

...of just removing it and replacing it with more background xD

hands without cherrypicking r still always a mixed bag

[nova remnant]

SDXL pony based models are preeetty good with hands, give them a try if you can run it

[zinc flax]

Holy shit this is nai?

[nova remnant]

sleepti nai inspector on the case

[nova remnant]

Does it not look very aesthetic ?

cx

Whats wrong with it

[zinc flax]

It's super clean

[nova remnant]

really

I thought it was like bad

lmao

(Compared to other stuff I see from nai)

[zinc flax]

You've seen the bad gens then lol

[nova remnant]

Very subjective

looks kinda weird to me

[zinc flax]

Though tbf it's pretty hard to get good gens from nai if you don't know the artist combination

[nova remnant]

damn

Those are very nice

I think you use a lora in there, right ?

Whoever provided some of those training data must be an awesome person!

jhdsgjsgdcjsdcsc

ong

theyre so cool

theyd be cooler if they provided more kgsdsc

[nova remnant]

[unkempt cape]

naiv3 so peak

[junior quest]

got a question, when we train a model or lora or whatever, lets say we have images that exceed 4096(max bucket res), so whats happening if we just trained with those images?

[bitter latch]

hmm this came out interesting

[green temple]

cool

[bitter latch]

new merge has my NOMD anime, rpgtoolsv4 and realcartoon, and changed seed by 1

there we go

[green temple]

broken but ok

[bitter latch]

interesting style

changes his up with some idea's

[green temple]

i didn`t even know how i got it 🤣

but cool

[green temple]

the face 🤣

cool

[green temple]

my honest reaction

[bitter latch]

changed it up to a more drawn style

[nova remnant]

[bitter latch]

oh.. forgot to change back to random. oops

tries 2048x2048 no upscale

oh boy this is gonna take a moment or 10

no nono , nitemare, not sharingthat. ick

goes back to bed

[green temple]

let`s goooooo 🤣

🤣

[wispy canopy]

Darkalfa+ELLA? didn't know my old SD1.5 model coudl do stuff like this 😮

[queen void]

A little bit late, but hai :)

[magic wave]

Hai haiii

What you been up to Nerf? :o

[nova remnant]

Too cute

Put a maid uniform on smol

[green temple]

there were no clothing prompt lol ,

[nova remnant]

She has the little maid headdress already

Get her to work x3

[green temple]

wtf

[green temple]

stole a prompt

original

[nova remnant]

no sad smol

[queen void]

Nothing much. University stuff and waiting for 5090

[vital raptor]

I really want the 5090 to have enough vram to be a viable upgrade to a 4090 for AI stuff, but on the other hand i don't want to spend another $2k 😄

[magic wave]

Fair :o

[green temple]

ooohhh

[bitter latch]

we sorta had same idea haha

[golden crow]

\

[wooden coral]

I feel this a lot

[bitter latch]

yeah they could get me a car with that

😄

not a good one tho

[bitter latch]

finally , she's on top of the tank. sheesh

[bitter latch]

ponders different theme. hmmmms

[bitter latch]

hmm ya know I am hungry. brb

[gloomy scroll]

Same. For some reason genning hamburger images isn't helping

[supple raptor]

I've seen predictions that it will have either 32 or 36 gb. Price is probably going to to be increased considerably again though so only spending another $2k for it probably isn't realistic

[vital raptor]

36gb isn’t even worth it for me. I’d need at least 48

[supple raptor]

it's not getting 48

[bitter latch]

puts some instant ramen in some water to cook

[nova remnant]

It will get 48

[supple raptor]

3gb chips exist now. it will probably get 12 of them which gives 36GB while the quadra/tesla get 24 of them for 72gb

[bitter latch]

suppose to be hamburger burglar costume....

[bitter latch]

ok enough of that

Ramen time

there thats better, now to eat my ramen

[bitter latch]

hmm mostly drained instant ramen with seasoning packet and a little bit of thousand island dressing mixed thru... not bad

[nova remnant]

grabby ears

[green temple]

oh nice

[zinc flax]

she angy

[gloomy scroll]

[bitter latch]

joins in

it tried....

oh .. uh oh

Prof X is that you?

[bitter latch]

got an upgrade

[tired crest]

Hello!

[bitter latch]

hello

[steady grail]

yo

just played a session of angel of dusk.
my god that game <.<

[steady grail]

bullet hell games breaks my brain in 2 >.<
but the story is so accattivating

[bitter latch]

the graphics.. ick

just no thanks

[bitter latch]

put'em on a street and he turned..hmm german?

[steady grail]

yeah it's very disgusting eheheh
considering the lore, it's not a surprise XD

[bitter latch]

lore?

I just did goblin soldier industrial punk

[nova remnant]

@nova remnant finished the final (for now) version of the eyeshadow lora

im still playing with weights and styles

but it seems to accept style loras without too much bleed now

[steady grail]

cute the goblin

[terse ore]

Is there any good workflow for img2img backgrounds? Trying to put blender renders through SD but the original renders look better :/ but I want anime style but they're not

[steady grail]

where, there is this node that can gen characters in transparent bg

i wonder ife there is a concat for noise

[steady grail]

try with this workflow. the top part generate the bg and the bottom part generate the depth controlnet of the character. everything is fused and generated on the ksampler on the bottom right.
there, you can try disconnecting the noise and add an image connected to a vae as noise, delete the ksampler at the top and add a vague prompt of the bg inserted at the top

[steady grail]

cute

[junior quest]

how to fix hands?

I mean when training a checkpoint model, not like generating images

[vital raptor]

extra smol

[nova remnant]

You're in eyeshadow gang now

[gloomy lodge]

their is a cool robot under your skin find it

[nova remnant]

Keep the bottom intact

And I'll find another thing

[bitter latch]

here have a new phone

goes back to bed

[nova remnant]

really cool image

[unkempt cape]

artifacts heaven but still look kinda cool

[nova remnant]

squish

[gloomy scroll]

You mean like this?

[unkempt cape]

i dont get the joke

[unkempt cape]

I in need of ketchup rn squash

[unkempt cape]

ah ok

[junior quest]

a question, whats the difference between finetune and training?

@supple raptor

[gloomy scroll]

Model finetune you have existing model you finetune for your concept dataset and model training you are making new model from dataset.

Now we wait for Nubby's correction.

[steady grail]

so, you are not fining the model to force it in doing your bidding?

[junior quest]

ouh so finetune means, improving dataset?

so how can we fix hands and other parts? like without using a hand fixer or anything?

[gloomy scroll]

Not dataset but teaching model to do new concept or do something just better

[junior quest]

eh how do we do that? trained a 1.5 model, it has bad hands and stuff, need to fix them

[gloomy scroll]

Don't rly now how to actually do training/finetune

[junior quest]

I see

[untold glacier]

You must pay Nubby in feet

Only then will he be summoned

[untold glacier]

[supple raptor]

Training is the general term for any type of, well, training. When creating a model from scratch, training is generally done in stages, finetuning is how you would describe the final stage or sometimes stages.

Finetuning can mean many things, but most generally it is taking a pretrained model which is more generalized and further training it on a more focused dataset to make it better at the thing you want it to do.

Here's an example of the simplified process to train a completely new model from scratch. First comes foundational training this usually involves an absolutely massive dataset to teach it as many concepts as possible. Next you further finetune the result of that initial training to improve understanding of the most common concepts. Finally you might do another finetuning pass with a focus on the artstyle of the model.

Basically everything the community does in regards to training could be considered finetuning.

[supple raptor]

yeah you pretty much got it

[gloomy scroll]

[supple raptor]

Finetuning always comes with some trade-offs. Making the model better at the things you want it to do will make it worse at some other things and sometimes in unpredictable ways, but it's better to have a model that's good at creating desirable content over one that's just average at making both desirable and undesirable content.

[junior quest]

oh so I did fintuning? got anythingv3 and trained on it, but got bad hands. hmm how do we fix those?

[grave snow]

https://tenor.com/view/starwars-obi-wan-now-thats-a-name-ive-not-heard-in-along-time-gif-11814407

[supple raptor]

It's hard to say. among other things it could be an issue with dataset tagging, the resolution of the training images, or it could be due to bad hyperparameters being used during training. If the dataset included many full body images, then you may need to change the dataset to include more close-up images which will show what hands should look like better, the trade off to this is you will be increasing the bias of the model towards portrait type images. Remember, the model doesn't see the same thing that you see when looking at the images. It is looking at latents which are basically an 8x8 pixel section of the image condensed down into a single value string to represent what those pixels show. This means that a hand that is small in the training data looks largely like an amorphous blob to the model and it's up to the VAE to try to make sense of that blob.

[junior quest]

I see, then I have to change my dataset

[supple raptor]

Also, for future training attempts, anythingv3 is largely considered a bad model to train on. It's already very overfitted and just overall not very good by today's standards

[junior quest]

so what are the good models?

[gloomy scroll]

Have some flowers

[supple raptor]

Most people opt to train on the original NAI model for SD1.5 training.

[junior quest]

I trained NAI and anythingv3, I noticed eye pupils and stuff works better on anyhingv4, thats why I choosed it

[supple raptor]

I assume you're training LoRA or LyCORIS right?

[junior quest]

no no, I did train lora, and then I noticed about those pupils and stuff, so I got anythingv3 to train a checkpoint also

[supple raptor]

You might consider trying a more modern model to use for the base instead. I would think one of the Based models would probably work better than AnyV3. TBH my guess would be that the hyperparameters probably just needed tweaking on your LoRA attempt though. Training a full checkpoint is generally not worth it unless you're doing very large scale training.

[nova remnant]

Nubby going insane on the keyboard

[gloomy scroll]

[junior quest]

hmm @supple raptor how about if I had images like this? about 50-100?

current dataset is included with about 1500 images

[supple raptor]

yeah having some of those might help. don't overdo it or you'll introduce the close-ups as a bias, but I would think with that dataset size, it shouldn't be a problem.

[junior quest]

naisu

[steady grail]

maybe tokenize a correct hand pose could help

the problem of the mangled hands is that the model gen hands in a random pose

[nova remnant]

1500 booba pictures

I approve if so

[junior quest]

souka, but most of them have

50/50 sfw and nsfw

I guess 50

[steady grail]

you should focus you time in improving booba gens!

most of the people does thaat, that's why today we gen fantastic boobs, but mangled hands

[junior quest]

we need to fix hands to grab b-

[steady grail]

https://tenor.com/view/im-glad-you-understand-eric-cartman-south-park-splatty-tomato-thank-you-for-your-understanding-gif-20405952

[nova remnant]

https://tenor.com/view/off-work-wake-up-poke-shaking-gif-17882870

[junior quest]

bad hands

1.5 SD

ahh I didnt had any wings in dataset

[vital raptor]

i've never seen a model that doesn't have bad hands

some can do pretty good, but most of the time inpainting is needed

[steady grail]

tokenize a precise hand pose require too much time

[terse ore]

Made this render in blender:

Put it thru SD:

Looks like it got worse 😔😭😢😞

Been trying to make decent img2img backgrounds from blender renders but they all turn out looking pretty bad. It works on characters but not BGs

[supple raptor]

that generation is 4608x3072. i'm amazed it looks as decent as it does. that's roughly 14x the number of pixels of the native res assuming SDXL or 54x if SD1.5

[terse ore]

Well I made the render at 1536x1024 and after a couple of passes and upscale it became that. It isn't the best one I've made but I just wanted it to look more anime. It looks more like a webcomic (because I guess that's what it basically is, a 3d asset with automatically drawn lines in the 3d software). I think the "correct" way is to inpaint each part of the image manually but that takes forever if you want to make panel-by-panel.

[bitter latch]

falls into a chair

On a space station field trip

[supple raptor]

I would img2img the whole thing with some sort of controlnet (you'll need to try them out to see which one works for you) at native resolution (1344x768 for SDXL) to start with, then upscale later after you're mostly happy with the style of the image. High denoising is going to look bad at high resolutions.

[terse ore]

Should I make the render (input image) at 1344x768?

[supple raptor]

Pprobably doesn't matter much. WebUI can handle downscaling that much without issues. The little bit of noise that it creates will go away when you img2img it anyways.

[terse ore]

I see, so basically render it at that res and then put it through a couple of ksamplers. What resolution should it be at that point, or just keep it at 1344x768?

[supple raptor]

keep it at 1344x768 until you're happy with the style then upscale at low denoising once you get the style down.

[terse ore]

Thanks I'll try it

[terse ore]

The eyes are so far apart they play baseball with each other

[steady grail]

i suggest depth cn

i love that cn XD

[steady grail]

the last 2 images are fisheye POV ?

[lyric zealot]

[junior quest]

@supple raptor lets say we dun have certain things in our dataset, but we train or finetune it on pre trained model (pretrain model is able to do those things that dataset doesnt include). so the question is finally after we got a model out of that dataset, is it able to do the things that dataset didnt include?

because the model I created with anything3 lacking so many things compared to anything3

I did ask this question cuz I saw a SDXL (pony based) model that only trained with 200 images, BUT still able to do pretty much everything that pony can do. maybe its because its sdxl and this is sd 1.5?

[magic wave]

Prettey

What the PNGInfo on that? :o

[green temple]

why booba 😭

[zinc flax]

Why not lol

[green temple]

why yes

[nova remnant]

Grabby

[bitter latch]

comfi ui pnginfo is a mess oi

[bitter latch]

wait why are there two?

[honest palm]

biomechanical style illustration beautiful woman, colorful melting human head. acrylic bubbles and flowers, ferrofluids, water distortions. looking up, intricate abstract. intricate artwork. beeple. blend of organic and mechanical elements, futuristic, cybernetic, detailed, intricate, dark theme, in awe:1.2, confident, surprised:0.7, natural skin, style of Stan Manoukian

[bitter latch]

sorry no bot here but thanks for prompt

[bitter latch]

very weird but ok

[green temple]

noice

[bitter latch]

hmm lets try a more anime model

kinda disturbing

tweaks it a little bit

oop I made it worse

better

the new borg?

ok NOW she's a borg

[bitter latch]

upscaled 4x WOW

[green temple]

5000X7000💀

[bitter latch]

uses his vivid pastel style on the prompt he's created

[bitter latch]

hehe yeah

hmm little bit diff

[green temple]

wtf

[bitter latch]

shrugs

[bitter latch]

here have a wallpaper

hmm if I added HR gigar to the style.....

or love craft

hmm.. nah

better not

a new queen?

one more and i'm going back to bed

[magic wave]

Trained a LoRA on my own style :o

[bitter latch]

finally

I would watch that anime trek

ok back to bed.. thud

[nova remnant]

Your own style ? o:

[green temple]

hmm...

what do you think guys

[nova remnant]

effect on the first one is cool

but a little bit too strong I think

If you're using a lora o:

still smol

[green temple]

trying multi diffusion

[magic wave]

Yeh, my own art :o

[nova remnant]

smol brings flower

[nova remnant]

smol leaves

[supple raptor]

this is where finding the right hyperparameters for training comes into play. ideally a finetune shouldn't lose too much of the capabilities of it's parent's model and should be able to generalize it's finetuned data across concepts from the original model. larger datasets also help in this regard.

[supple raptor]

generally speaking, models like this that are only trained on a couple hundred images are trained as a LoRA which is then merged with the model for release.

[pale current]

原来是乡里大佬谢谢！😄

[lyric zealot]

sorry for the late reply, but here goes score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, rating_safe, 1girl, pink hair, ponytail, forehead, freckles, hairband, sleeveless shirt, looking at viewer, source_anime Negative prompt: (worst quality, low quality:1.2), 3d, censored, abdominal muscle, artistic glow Steps: 24, Sampler: Euler a, CFG scale: 7, Seed: 379054278, Size: 768x1024, Model hash: 90c2c6dc18, Model: tPonynai3_v35, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Version: f0.0.17v1.8.0rc-latest-269-gef35383b

[magic wave]

Ooo, Tponynai v35? I've been using v41 :o

[lyric zealot]

i think i wanted to use v41

but mistakenly used v35

[magic wave]

O

XD

Thats fair

xD

41 has the best anatomy so far of those models I think :o

[lyric zealot]

i think i tried to swapped model, but didn't realised it didn't switch

[magic wave]

Ohhh, makes sense xD

[stable crane]

anatomy you say

[magic wave]

Yis :o

[stable crane]

here I go again, downloading another model

[magic wave]

xD

[stable crane]

I've still been using autismmix_confetti

[magic wave]

I've been mostly using AutismMix_DPO :o

But T-ponynai3 has a cool style :o

I also heard of some people trying to do like a merge of ArtiWaifu, Autism and Animagine, but dunno how thts going :o

[magic wave]

Mikoo in my style

[terse ore]

@supple raptor I did straight img2img with the resolution you recommended, changed some prompt stuff, and changed the model as well. Even without upscaling I'm surprised it looks this good. I'll try upscaling stuff tomorrow.

[bitter latch]

she's coming for us all

she got an upgrade

[bitter latch]

thats better

she grumpy

[bitter latch]

love it when it does this

hmmmm

changes models

hmm pointy eared vulcan still gives elf ears.. so vulcans are space elvs?

ok to bed.. thud

sneaks one in

ooh small pointy ears! finally

closes the apps

[nova remnant]

sleep is for the weak, come back

@nova remnant ik you dont use a1111 but do you know anything about regional prompter

[lyric zealot]

Im not @nova remnant but what do you wanna know about regional prompter?

[nova remnant]

how to make my images come out horribly when using a character lora with it

rn i have cs10 teaching me a different extension thats similar but id like to learn this one as well

[lyric zealot]

you meant not come out horribly right?

[nova remnant]

ye

im high

words aint my strong suit rn

[lyric zealot]

regional prompter have two basic generation mode, attention and latent. If you are using lora, you will probably want to use latent mode instead. If you use attention mode, the lora will be applied to all regions regardless of where you put it

[nova remnant]

ive tried latent and attention, debug, messing with lora stop steps

its all hit or miss

but nothing turns out the results i need

its all lowquality, or looks nothing like the characters i prompted, or it merges them into a monster

[lyric zealot]

can you get something really basic with regional prompter first?

like maybe region 1 with a girl, region 2 with a boy

and check your cmd to know if it's working properly

[nova remnant]

yeah i did that first

it came out fine

its only when i decided to put my oc with a character plus that characters lora

did things start messing up

im pretty sure it doesnt like character loras but i need it to like them 💀

[lyric zealot]

do you need your lora to apply to only one region?

[nova remnant]

yessss

to one character and one character alone

[lyric zealot]

then you have to limit yourself with latent mode. And from my previous testing, latent mode with regional prompter isn't that good

[nova remnant]

yeah

latent caused the quality to drop into hell territory

[lyric zealot]

there's another one i used to use

[nova remnant]

sometimes it had some good stuff but then it ruins itself midway through

[lyric zealot]

i think it's this https://github.com/opparco/stable-diffusion-webui-two-shot

[nova remnant]

ill try this one as well

would you be willing to walk me through it

[lyric zealot]

there might be a better fork, or a better updated fork

[nova remnant]

itll be a hassle because im being walked through on how to use another region prompter like extension in tech support

but id apprecaite it

[lyric zealot]

well, i mean try that first, since you won't have to install anything new

[nova remnant]

once im done ill ping you, if thats alright

[lyric zealot]

sure, just that i might not be here tho

Since im gonna go eat soon

[nova remnant]

understandable

[white zenith]

magic miku

[green temple]

nice

[nova remnant]

No it never worked on my PC

[green temple]

ask @brittle rampart 👍

[brittle rampart]

No need! I'm stupid girl!

[green temple]

forgot to @

[nova remnant]

Smart Smol

[nova remnant]

Book about delicious carrots

[brittle rampart]

100500 carrot recipes

[nova remnant]

My favorite recipe: peel and eat

[green temple]

book about catzu 👍

[nova remnant]

👍

[nova remnant]

me as a recipe

I'll be cooked

[green temple]

joining her

[brittle rampart]

We don't know how to draw...

[nova remnant]

The chef

big ears

[nova remnant]

@green temple why you upside down looking at me

[bitter latch]

sigh, didn't get much sleep. blah

guess misery does love company

😄

huh.. adding art by Akihito Tsukushi, makes a big diff

[bitter latch]

huh I did not rquest a border. but ok