🍥｜anime

[supple raptor]

@tepid geyser I would recommend grabbing this one specifically as a fairly easy to use and flexible model. https://huggingface.co/SweetLuna/Aurora
You want are the AuroraONE_F16.safetensors model from the Aurora One folder, which goes in the folder I previously mentioned. You should also grab the AuroraNegative.pt and KHFB.pt files in the AuroraEmbeddings folder, these go in .\embeddings and to use them, you simply add the filename without the file extension to your negative prompt. Here is an example of a negative prompt that I use that uses these embeddings (KHFB, AuroraNegative:1.1), (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2)

[tepid geyser]

thanks so much for your help!!

let me try to make somethign happen

wish me luck

[supple raptor]

To download from huggingface, first click on Files at the top. Then find the file you want to download, in this case the model is inside the Aurora ONE folder so you click there. Then click the small download icon next to the file you want to actually download.

[tepid geyser]

i've been scarred so much

i have seen more body horror in the past 30 minutes of trying to make this work than in my entire life

[supple raptor]

I suspect you'll start getting much better results with this model and setting your Clip Skip to 2.

[tepid geyser]

what are the parentheses for in the prompts?

like some prompts dont have parentheses and others do in examples ive seen

[supple raptor]

This is probably the best explanation for that. Essentially, parenthesis make the prompt pay more attention to what is inside of them. #🍥｜anime message

[tepid geyser]

ah okay thanks!

[tepid geyser]

MY FIRST DECENT LOOKING ONE

[tepid geyser]

quick question if you have time, this model says it is "versatile" and shows a lot of different artstyles. to make it a certain style is there like a setting or value? or do i just have to find the right prompt to make it look that way?

[supple raptor]

It's done through prompting. usually I use the danbooru tags for different artstyles, such as graphite \(medium\) for a pencil drawing style. It will also respond to words like realistic when you want a more realistic looking image. Do you have the "tagcomplete" extension yet? https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git

[tepid geyser]

I do, but im not sure how to use it, the tagcomplete extension that is. is it really as simple as typing and if it recognizes the tag then itll help me fill it?

[supple raptor]

Yep, but look at the number next to the tag. The number represents the number of images tagged with that tag on danbooru's website. In general, tags that have less than 1k images will have little to no effect in your prompt. You can try those ones, but you may need to find other ways to describe what you want. Tags that have 5k or more images tend to be quite reliable. Character tags have a wide range of minimum images they need to work, but usually 10K or more are pretty reliable. Some characters work with far less though.

[tepid geyser]

thank you, i'll study it more

[supple raptor]

Here's an example prompt I made based on your image using mostly danbooru tags: 1girl, serafuku, in a classroom, brown eyes, leaning on a desk, black hair, blue skirt, white shirt, sailor collar, absurdly long hair, chalkboard in the background, dappled sunlight, light rays, extremely detailed, intricate

[tepid geyser]

woah

[supple raptor]

That's using the AuroraOne_F16 model too.

[tepid geyser]

wait so these are all mostly danbooru tags so it makes the generation lean more towards images that contain those elements?

damn, okay i'll try to figure that out thanks for the example!

[supple raptor]

Pretty much, yeah. The AI is trained using text and image pairings, our anime models are almost all trained on images tagged with danbooru tags. Which allows it to learn pretty effectively which part of the image is associated with most of the different tags.

[tepid geyser]

how important is sampling steps above 20?

[supple raptor]

Depends on the sampler. I use DPM++ 2M Karras and almost always use exactly 20 steps.

[supple raptor]

15 steps is often enough for this sampler, but 20 is a pretty safe number to ensure that it gets enough steps.

@tepid geyser do you know how to use PNG Info to view other people's prompts?

[tepid geyser]

no, i did not know that was a thing

[supple raptor]

Yep, click on one of the images that either I or kether posted and then click "open original" or "open in browser". Then go to your WebUI and click on the PNG Info tab. Then drag that image from the browser window into the source field in WebUI and it will display the prompt as well as the the parameters used to generate it.

[tepid geyser]

holy crap

thank you! that's a great resource

and also this gives me a new perspective on how long prompts can be

👀

[supple raptor]

It will only work if the person uploaded the image with the metadata intact, but many of us do, and it can be a very good way to learn how to prompt better by seeing what other people do and then trying to modify their prompt to get what you want.

In that example, most of it is the negative prompt. Only the first 3 lines are the actual prompt.

[tepid geyser]

oh wow

i didnt realize

thanks for the resource!

any tips on how to avoid weird hands and stuff like that?

[supple raptor]

This negative embedding can help a bit: https://civitai.com/models/56519/negativehand-negative-embedding
Put it in your .\embeddings folder and then add the filename without the extension to your your prompt. The other thing that is the biggest help is using tags that define a specific position that the hands should be in. Hands are weird and can make a lot of different shapes which often makes the AI confused when it has to try to come up with a shape for them on it's own. Tags like waving tend to work pretty well. Luck is also a big factor. Many of us that post images with good hands probably have several with bad hands for every one that looks good.

[tepid geyser]

alrighty, thanks!

[tepid geyser]

wait do i want to add the negative hand to my regular prompt or negative prompt?

im assuming negative right

also does the order of prompts matter a lot? like lets say character prompts vs physical objects vs scenery ordering

[supple raptor]

negative prompt for that one.

[supple raptor]

Somewhat yes, prompts are interpretted in blocks of 75 tokens. Tags closer to the start of a block of 75 tokens are inherently assigned more importance than tags closer to the end. Tokens are shown by that number in the top right corner of your prompt box. You want to make sure that tag is all within the same block of 75 tokens (sometimes words can be made up of multiple tokens) otherwise the AI can't understand the full tag. For example blue hair is 2 tokens which are most likely "blue" and "hair" if you added that when you already have 74 tokens then blue would be in the first block of 75 tokens and hair would be in the next so it would know you want blue in the image and that you want hair in the image, but it likely wouldn't get that you want the hair to be blue. chalkboard is also 2 tokens so it could also be split as "chalk" and "board". You can add BREAK in all capital letters at any point to fill the current block of 75 tokens with empty tokens and force the next token to be the start of a new block.

[tepid geyser]

thanks, that's very informative!

all this time i've been iterating something and i finally got something that looks decent!!

this is so exciting

it's so cool that i can generate images of this quality locally for free

[supple raptor]

Oh I like that!

[tepid geyser]

i dont have to rely on tools like midjourney

[supple raptor]

Yep, the power of open source tech is awesome!

[tepid geyser]

thanks for all your help today :)

[supple raptor]

My pleasure! The more people I help, the more people I have posting images for me to enjoy XD

[tepid geyser]

oh real quick, how do i delete saved styles?

i saved one way back when i had no idea what i was doing and want to clear it

actually im not 100% what styles do still

but

[supple raptor]

Styles are just a way to save all or part of a prompt so you can easily add it. I like to use it to save my negative prompts and a few prompts that I commonly use to test models. To delete them, you have to open the styles.csv file in the root directory and manually delete them.

[tepid geyser]

ah okay, thank you!

she lost her blade

LOL

there we gooo

[tepid geyser]

love the ambiance!

[lament osprey]

Epic

[supple raptor]

noise offset loras help a lot for dark/night time images. this is the one that I'm using https://huggingface.co/closertodeath/ctdlora/blob/main/loraslayernoiseoffset.safetensors

how much vram does your gpu have @tepid geyser?

[tepid geyser]

thanks for the extra resource!

I have 8gb of vram, but because it's an AMD gpu it's not supported well so i have some trouble doing bigger tasks

[supple raptor]

Oh, yeah. No, xformers on AMD unfortunately which is a huge vram savings. I don't know if she's around right now, but I know that @trim current uses an AMD GPU so she might have some tips for making SD run better on them.

[trim current]

Uh linjx

[tepid geyser]

!!

oo it autosaves all your images

that's good to know

[trim current]

COMMANDLINE ARGS= line and change it to
set COMMANDLINE ARGS=--medvram --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-

Ba0ck to bed

[tepid geyser]

thanks for the tips! ill try those settings

[supple raptor]

https://tenor.com/view/good-night-night-sleep-well-sleepy-sweet-dreams-gif-21090687

[wispy canopy]

what is your name? "MA NAME"

[umbral depot]

Morning

[tepid geyser]

rapunzel?

[tepid geyser]

did you do text to image or image to image?

[hidden grove]

text2img only. Lora "Rapunzel Disney Princess, Tangled by YeiyeiArt"

[tepid geyser]

oh wow there's a lora specifically for her that's cool

[tepid geyser]

wait a second

her hand is coming out from her elbow

[supple raptor]

She's just missing her left hand and it's her right hand that you can see. Laying on her harm.

[tepid geyser]

i choose to believe that her left hand is just inside the sleeve

:D

[supple raptor]

She chewed it off in her sleep like she is about to do to her right hand

[tepid geyser]

oh no

[supple raptor]

[tepid geyser]

i like this one but she has a third arm

here's another iteration, but for some reason adding "only two arms" made it ignore "black hair"

[glacial bane]

When will genderbend lora be uploaded

I just lost a good seed

Anyone know how to get it back

It was literally perfect on the preveuw

But my pc turned off and when it was finished it didn't even show

[tepid geyser]

man me too, i just lost the seed that gave me the two images above and everything else is so boring and bad

[supple raptor]

you can drag the image from the folder it's saved in on your PC into the PNG info tab to get that seed back.

[tepid geyser]

ah!

fantastic

[glacial bane]

What

I check png and press what

[tepid geyser]

if you have the image that was generated, you can check the metadata on it

[supple raptor]

If you haven't generated anything since, you can click that blue arrow icon to have it autofill the prompt and all of the parameters of the last generation including the seed.

[glacial bane]

Oh thank god.

Wait what if i tried generating something before you told me this?

[supple raptor]

Then it would be lost to the void unless it happened to be one of a batch of multiple images (and not the first in that batch) in which case you could figure it out since the seeds are sequential in batches.

[glacial bane]

Oh sad

[tepid geyser]

took a few tries but i think it turned out okay

:D

another version but with wonkier fingers

[glacial bane]

How the hell

Blud heow

[tepid geyser]

Prompt: ((2d1.5)), masterpiece, best quality, naruto (series), ((uzumaki naruto)), ((genderswap)), ((1girl)), cute, beautiful, medium breasts, blonde hair, ((blue eyes)), ((orange jumpsuit)), ninja, ninja headband, konohagakure symbol, arm up peace sign, hand on hip BREAK Konohagakure, village hidden in the leaves lora:add_detail:1.5 lora:flat2DAnimerge_v10:1
Negative prompt: (KHFB, AuroraNegative:1.1), (Worst Quality, Low Quality:1.2), border, nsfw, skimpy, grayscale, multiple_girls, (watermark:1.2), photorealistic, bad anatomy, bad hands, mutated hands and fingers, bad feet, username, text, error, blurry, jpeg artifacts, cropped
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 46334899, Size: 600x800, Model hash: 1b5f8211ec, Model: AuroraONE_F16, Clip skip: 2

the different images were iterations of these, sometimes just changing the seed

[sage kiln]

Where are we going mecha-kun?

[still night]

[tepid geyser]

damn, love the detailing inside the machine

do you think increasing add detail to 2 would make the machine even more complex inside?

[sage kiln]

Maybe but overdoing it gives you unwanted artifacts

[sage kiln]

But the mecha lora gives really good settings

[tepid geyser]

damnnnn i'll definitely be looking into that!

[glacial bane]

@tepid geyser you want 2D?

[pastel adder]

Hi, does everyone use Stable Diffusion checkpoint v1-5-pruned-emaonly.safetensors?
Though I am not sure if it affects the generated image.

[tepid geyser]

for anime there are better checkpoints

https://huggingface.co/SweetLuna/Aurora

such as this one!

[tepid geyser]

wdym want 2d??

but probably yes

[pastel adder]

Thank you!

[tepid geyser]

glad to help!

[tepid geyser]

not as good, but i did my best

[pastel adder]

Hey, thanks for this. I got extremely better result.

[fallen sundial]

hello

is there someone has a strong gpu ? I want to ask for creating an image for me cuz I don't have really strong gpu

[zinc flax]

Use google colab

[fallen sundial]

can I get the link ?

[zinc flax]

Google nocrypt colab

[fallen sundial]

how to use google colab exactly ? is there a video?

[glacial bane]

Bro didn't even use a lora.

[zinc flax]

the power of waving

[cyan burrow]

she's got like a pencil neck, careful it doesn't snap from the weight of her head

[compact warren]

@supple raptor tried reference only, it doesnt seem like its reflecting the image i give it

[compact warren]

Tried it with some 3d style images (koikatsu ofc)
Images just seem like theyre generated under normal t2i parameters

[wispy canopy]

interesting outcome...

[compact warren]

Why can’t i generate anything so highres and detailed 😭

[wispy canopy]

make image 640x640 -> upscale to 1280x1280 -> feed image to controlnet tile upscale, change prompt to 'lineart, pastel colors', change resolution to 1920x1920

[compact warren]

Controlnet has an upscale mode?

[steady grail]

hallo

[wispy canopy]

seemingly, that tile resample can be used for it

[compact warren]

Am confused

[wispy canopy]

og image -> this one i fed to controlnet tile resample

my actual prompt in the resampled image is detailed lineart, pastel colors, and just increased the resolution

was just testing what it actually does

[fallen sundial]

is that you who made the model ?

and also what vae you use ?

[compact warren]

What does the tile resample setting look like

[wispy canopy]

did just use the default settings, nothing changed

[steady grail]

why upscaling like this?

[compact warren]

I will try to remember this

Multidiffusion seems hard to grasp

[wispy canopy]

multidiffusion?

[compact warren]

The other go-to upscale method these days

Idk how to get it working properly

[wispy canopy]

og, upscale 2x, upscale 3x, using tile resample

[compact warren]

Ive never heard of tile resample before now

Only tiled VAE

Or tile diffusion

[wispy canopy]

tile resample gets some pretty detailed results though

[compact warren]

So many methods to do one thing in particular, it drives me nuts
Idk any upscaling methods besides the normal hires fix

[steady grail]

a ghost O.O

[wispy canopy]

highres is just doing the diffusion over, using a base noise pattern (renoised image)

tile resample is using the specific image as a base, and it uses your prompt to imagine the things it can't sample properly from the image? i dunno

[compact warren]

I value simplicity but there is no such thing with ai

[wispy canopy]

and then there's comfyui xD

[steady grail]

oh, so, it's a way to rescale and add the lost details

canny is not better in this?

[wispy canopy]

canny is edge detection

and using a 'smart' way to use those lines in what they should represent, when combined with the prompt

[steady grail]

add the original image on top to have the colors and canny down, use tiled diffusion and voilà, you get a very good upscaling of the original

[pastel adder]

Hi, how do I avoid this result?

[steady grail]

but i will try the 2 metods and see the differences

[steady grail]

a ghost!

[pastel adder]

[wispy canopy]

you're doing a single shot, no highres upscale generation?

[pastel adder]

I have no idea what you mean, sorry. But here's what I used:

[nova remnant]

that stuff happens sometimes when you gen larger resolutions

[pastel adder]

oh, what would be the recommended resolution?

[nova remnant]

it's usually better to gen at a lower resolution like 768x and then highres. fix to the desired resolution

[steady grail]

closer to 512x512

the closer is, the less mutations you get

[wispy canopy]

yeah... the thing about SD -> it's context aware in an area of 512x512 , how it was trained

so when you go for big resolutions, it simply isn't aware it's generating 2 heads

that's why most of us use highres fix for that

so you first make a 512x512 image, (or 768x768) and then upscale that with highres fix

[steady grail]

i'm developing a theory that it is aware of what it is doing, but it is punishing us for forcing him do extra work

[pastel adder]

Understood. Thank you!

[wispy canopy]

Here is an example which explains it pretty well: Both have the same aspect ratio, however because of the difference in prompt, on the #2, it started 'repeating patterns'. However, the prompt on #1 gave the AI enough information to know how do the composition without having to duplicate things...

[steady grail]

tick!

[wispy canopy]

the prompt: absolute thiccness, fashion, business meeting, presentation

LOL

[warm plinth]

Could you please tell me how to begin the fun? is it happening on a specific channel here ? how to start producing pixtures? is it being recalled by some sort of command?

[tropic wraith]

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐｜pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610⁠ for more details! You can also stop by ⁠#1025467151206854736 for any issues you experience while using DreamStudio or ⁠#🤝｜tech-support for any problems you encounter while installing it locally!

[wispy canopy]

#1072229020520947753 #1072220168534642768

ah, the command /faq even exists now? 😄

[shut wind]

yes just must be used as solo 🙂

[warm plinth]

oke thanks bro

[wispy canopy]

i know 🙂 that's why i used it in sentence, to not trigger it

and hi @warm plinth , hope you'll stick around ^^

[warm plinth]

No worries I will stay around

[pastel adder]

Where can I find a guide for upscaling images? I think I found it.

[fallen sundial]

hello how to change a pose of this image by stable ?

I want him to be holding book

[steady grail]

if openpose doesn't work, try to do a sketch by hand and than use canny

[fallen sundial]

OH

I forgot open pose

[trim current]

I wake up and this is rhe first thing i see

[wispy canopy]

https://tenor.com/view/chefs-kiss-french-chef-perfect-dish-excellent-food-perfection-gif-20341505

[trim current]

XD

[wispy canopy]

on a side note, cartoons like that fucked up our hands xD

(nah, there's a whole lot more going on)

[trim current]

@tepid geyser for amd youre best off swapping to linux. Dual boot or wslz preferably dual booting. Linux uses rocm and it changes from It/s gens to s/it
512 takes 10ish seconds with rocm vs 2mind with vulkan

They adding rocm to windows but fuckin knows when

[fallen sundial]

@steady grail btw I can't use control net

[trim current]

Could be tomorrow could be in 3 yrs

[steady grail]

why?

[fallen sundial]

it's an big extention so my pc can't handle it

[steady grail]

you have a gc 1660, right?

[fallen sundial]

yes

6 vram

[steady grail]

better than mine, your problem i'm pretty sure is with the last xformers

i have 4 g and i can use it

[fallen sundial]

how ?

should I reinstall xformers 0.16 ?

[steady grail]

it's not so easyy

[fallen sundial]

so what to do ?

[steady grail]

follow this #🤝｜tech-support message

[rich tartan]

uh, i would like to hear them too

[steady grail]

than git pull in git bash here

i posted the right hash after that post

[rich tartan]

@trim current what optimization are you using? For amd/directml

[steady grail]

the one i used

[trim current]

I am not

Lol

Whatever i posted was when Indid run ML

[wispy canopy]

this looks like my vhils prompt 😮

[trim current]

Backtrack my reply to nubby and fenrix

Youll see the flags i posted for AMD ML usage

But in all honesty for amd i think jts better to use linux. Rocm blows it out of the water its fucking insane

And laughable

[steady grail]

i used this e55cb92067094564c259b515b8a6ec3869654083

[wispy canopy]

lol, prompt only absolute thiccness causes.. things.

[trim current]

Lmfao

Goddamnit eface

XD

[rich tartan]

thanks

[trim current]

Yw

[wispy canopy]

one CHEEK gets accentuated

[trim current]

I still use most flags on linux just to make it stable

But like example difference.
ML 512 takes at most 1min with tiled, im limited to 1.5 higres to 768
On rocm is 15seconds and i can highres x4

Too gay and early in the morning for this amh

I need coffee

[wispy canopy]

LOL xD

[steady grail]

add a coffee in the image

[wispy canopy]

i'm in awe of the image quality, 512 non upscaled.

[fallen sundial]

what is this .-.?

[wispy canopy]

who cares about waifu ass when image quality is on the line!

[rich tartan]

dam, that's optimization asf, one last question, did you install linux on a vm?

[trim current]

I wouldnt

Personally dual boot

I got 2 ssds

Some ppl swear by wsl

[rich tartan]

oh, i understood now

[trim current]

But id just dual boot

[rich tartan]

interesting

[fallen sundial]

are you serious ,-. ?

[rich tartan]

thank you again

[steady grail]

i used it to downgrade stable diffusion and use the old xformer

[trim current]

Yeah one drive windows the other linux

No worries

[wispy canopy]

yeah, ur probably right 🙂

[trim current]

Lmfao

[wispy canopy]

and i'm also one of the dude who does mention that to people.

[trim current]

Yeah but only when its excessive i find

[wispy canopy]

however, i specifically go for nippleslips and cameltoes

[trim current]

Straight up nudity is when i say smth

[fallen sundial]

I should type this in cmd ?

[wispy canopy]

straight nudity is easy no

but cameltoes CAN happen, and i do call them out

[trim current]

Yeee

[wispy canopy]

however, if someone has issues with ... VERY skimpy skin+thicc, i'm not above taking my image down 🙂

[sick idol]

[wispy canopy]

there was no nudity, cameltoe, or even sexual intent. The image itself was purely in jest of the prompt only being 'absolute thiccness`. However, i respect you calling it out, and remove it accordingly. we good? ^^

[steady grail]

right button on stable-diffusion-webui than git gui here, and type git checkout e55cb92067094564c259b515b8a6ec3869654083

[fallen sundial]

That's why I hate anime girls, people made anime worse with these girls
anyway I don't want to talk about this

[fallen sundial]

Okay , thanks

[steady grail]

delete the venv and stop using --precision full --no-half --no-half-vae and restart stable diffusion

[fallen sundial]

I use no half becasue I get errors

wait.. I should detete venu?

That's means I should wait for downloading...

I use these cods

[steady grail]

yep

it's necessary

with that xformers you don't need it any more

[fallen sundial]

what now I should delete ?

[steady grail]

use this --always-batch-cond-uncond --opt-split-attention --xformers --medvram

[fallen sundial]

are you sure ?

I don't know if I may get another problems

[trim current]

thats a far reach

i hope your back is ok

[steady grail]

you have more gpu than me, if it works with me, i don't see why it shouldn't work with you

[fallen sundial]

I don't get what you mean ._.

[trim current]

its okay

[fallen sundial]

okay

[wispy canopy]

the guy/girl said doesn't want to talk about it. so i'll shut my mouth, but i share your sentiment

[trim current]

yeah thats all i had to say about it and im moving on xP

[wispy canopy]

less discussion, more waifus!

[fallen sundial]

why I don't like a guy ? my character is a boy, I see people don't know I'm a boy

[trim current]

thats pretty

[steady grail]

the grain is pretty

[wispy canopy]

we're on the internet. as far as i'm concerned, everyone is ungendered.

[trim current]

pretty much

[steady grail]

too bad we have that woman in the middle

[trim current]

i go neutral till told otherwise

[fallen sundial]

where is the git gui ?

[fallen sundial]

oh

[steady grail]

you are in stable-diffusion-webui

[fallen sundial]

yes

[steady grail]

you should go one directory before and press on the directory with right button

[magic wave]

Wait

[fallen sundial]

so you mean this

[magic wave]

Isn't git checkout used in root folder

[steady grail]

i use git bash here

[magic wave]

The root folder is stable-diffusion-webui

[steady grail]

dunno the difference

[fallen sundial]

but nothing happens.

[steady grail]

oh, so with get gui that's what it does

[rich tartan]

i needed to look twice too

[fallen sundial]

so what should I do ?

[fallen sundial]

for me I don't get anything in git

or python

[steady grail]

close it and go on git bash here

you will open a prompt in git

[steady grail]

perfect

close it

now sd is downgraded

delete the venv and restart

[fallen sundial]

you mean get ready for the slowest internet.

[steady grail]

i know man

[fallen sundial]

you mean restert this ?

restart*

[steady grail]

yep

[fallen sundial]

is there a way to share internet ?

[trim current]

she look sad but is pretty

[wispy canopy]

here, NSFW fields of gold!

[fallen sundial]

I should get ready for the next battle.

[steady grail]

tick!

[trim current]

torch 2.0 is faster

[wispy canopy]

yes. many ticks. SO. MANY. TICKS. do not walk through these fields -> these fields are LYME COUNTRY

[stone onyx]

innocent, blush, embarrassed

[steady grail]

yeah, but the old xiformers isn't compatible

[trim current]

am takin abit of break from sd atm, (knowing ill be back in later as much as i try)
my model merge is coming nicely but still looks bleh and ocd is like FIX IT FIX IT FIX IT FIX IT

an trying a new method

[stone onyx]

I got those breaks as well

[wispy canopy]

but. will it beat darkalfa?!

[stone onyx]

I'd take a break for 5 days then come back

against my will

[trim current]

hoping it wil be good enough for ppl to use

mood

[wispy canopy]

darkalfa is doing worse than alfahentai 😦

[trim current]

rn its coming out too cel-shaded

[wispy canopy]

people are horny xD

[trim current]

new method is creating the base with 1.5 sd but merged with add-details

1 is too high, will need to remerge lower and try again

[stone onyx]

this sounds like a mess

I'm never gonna get into mixing models

[wispy canopy]

good. keep it that way 😛

i'm already hating on darkalfa because i identified flaws 😛

[wispy canopy]

last image was on alfahentai. this is on darkalfa. that wheat is VERY NSFW

[steady grail]

i think you should censore that wheat in the second one XD

[wispy canopy]

yeah, you are probably right.

[brittle jay]

hello everyone

[trim current]

It is 😄

[magic wave]

Cool style with reference controlnet :o

[nova remnant]

What model?

[wispy canopy]

DarkAlfa

My own model mix

[magic wave]

This controlnet is fonky, but its dope

[magic wave]

Thicccc

[wispy canopy]

lol. medieval thiccness 😄

[sullen hollow]

this is a medieval goddess

[sullen hollow]

colors on point

why is the wheat nsfw i dont get it

[hollow meteor]

oh yeah. That wheat would get me fired. v. nsfw

[sullen hollow]

?

[wispy canopy]

because the wheat is VERY good looking

[sullen hollow]

oh

[wispy canopy]

when anyything gets high enough image quality, it could be considered 'nsfw' because it's just that damn good

i mean. it's just that, a wheat field, but damn, does it makes you want to be there ❤️

[rich tartan]

Faces on point, ngl, you got luck

[sullen hollow]

XD totally nsfw wheat

[wispy canopy]

she has concerns...

[wispy canopy]

SCIENCE!

[rain valve]

oh it's SCIENCE power

[wispy canopy]

if anything, i am really proud how pretty the images come out of my model 😄

[brittle jay]

eface, how bad is my cow submission from 1 to 10?

[wispy canopy]

uh... i'll just go with "i'm sure you won't win"

[brittle jay]

nice

[wispy canopy]

i know it's gutts

but... i'm kinda mystified what you were trying to attempt

[brittle jay]

gotta say though, probably one of my favorite submissions thus far

[brittle jay]

I think my image looks bad quality wise (looks like someone who only drew for a few months or like a year total drew it) but I really liked how I managed to make the idea I had in mind

[wispy canopy]

that's actually more important imho 🙂

[brittle jay]

yes

[wispy canopy]

i've done done some submissions i knew wouldn't win

but were more close to my heart than making something 'that woudl be more probable to win'

[brittle jay]

what cfg?

[magic wave]

Trying more reference-only controlnet

[trim current]

This is my first submission. I know i wont win pow

[wispy canopy]

^^

first pow is always scary

[magic wave]

Using doujinshi pages from nhentai kinda works

Not sure what would work best tho

[stone onyx]

so CUUUUUUUUUUTEEEEEEEEEEEEEEEEEEEEEEEEEEE

[stone onyx]

first POW always hurts

[trim current]

[magic wave]

Ayyy

[sage kiln]

In a chassis of steel, she ponders her plight, trapped in the chaos of relentless fight;
Her purpose in war, a cryptic lure, begging the question - for what does she endure?

[stone onyx]

AAAAAAAAAAAAAAAAAAAAAAAHHHH

I'm stealing ur prompt

gimme PROMPT

[magic wave]

Its thanks to the new controlnet :o

Here, this the pic I used for it

[wispy canopy]

WHO IS USING MY LLM PROMPTING STYLE -> YOU NEED MY MODEL!

[sage kiln]

It's just flavour text haha

Also that is chatGPT

[wispy canopy]

no. that is a prompt!

Prompt: Nestled amidst the tangled roots of a massive,ancient tree lies a small clearing,illuminated by the soft glow of bioluminescent fungi clinging to the trunk. Amidst the underbrush,a curious sight catches the eye - a sleek,metallic feline perched atop a fallen log,its eyes fixed firmly on its surroundings. The Mechanical Cheshire Cat,once a mere housecat,has been transformed by unknown means into a living weapon,its body adorned with gears and circuitry. One half of its body is painted in bright blues and yellows,while the other half is coated in matte black. Its long,slender tail ends in a deadly barb,perfect for snaring prey,
2x controlnet upscale (tile resample + reference)

i have a 1x controlnet upscale as cow submission, but i might even like this one more 😮

[stone onyx]

THANK YOU

[wispy canopy]

(using the og as reference keeps the end image coherence even better)

[brittle jay]

what is that reference_only?

[sage kiln]

Wtf it actually is

[brittle jay]

why so saturated then?

[magic wave]

New controlnet update

[wispy canopy]

i know an LLM prompt when i see it

[magic wave]

Came out with v1.1.153 I think

[brittle jay]

what it do?

[magic wave]

https://github.com/Mikubill/sd-webui-controlnet/discussions/1236

[brittle jay]

so this is like img2img but better?

[wispy canopy]

basically, it takes the 'elements' of the image, and uses it as reference

[wispy canopy]

no.

[magic wave]

I'm not too sure

But

[brittle jay]

this is what people think img2img is supposed to be

?

[magic wave]

It does things

[wispy canopy]

@brittle jay it uses elements of the image you use, for 'inspiration'

be it style, or subject

something like the t2i style

[brittle jay]

yeah so what people think img2img is supposed to be

[wispy canopy]

in smile's case, he used black/white art, so that style gets applied

[magic wave]

I mean, not only that, cause I did have to put on (greyscale:1.4) and use black/grey/white colors for the hair and eyes and stuff

[wispy canopy]

in my case with the mecha cat, i used the reference style of the og non-upscaled image, so that it could keep image coherence better

[magic wave]

But it also mimicked the contrast

Cause normally the hair would be done darker

But it mimicked it all being like all the same washed out black

And I feel like it probably mimics like the line style too and color composition too

[wispy canopy]

OG - > Tile resample Upscale (non reference) -> Tile resampe upscale (with reference to OG)

[magic wave]

Generated pic & controlnet pic, it is trying to do something with like the black at the bottom

[wispy canopy]

make sure your resolutions align

[magic wave]

Too much work tho

[wispy canopy]

fair.

[magic wave]

But after tweaking the controlnet pic it works fine

[wispy canopy]

you the og artist?

[magic wave]

Thats a picrew thingy :o

So I could change it easily

[wispy canopy]

picrew?

[magic wave]

Also for my COW submission I did a Cyberpunk 2077 screenshot as the controlnet pic

[magic wave]

https://picrew.me/en/image_maker/257476

[sage kiln]

Wait so is the right the controlnet pic and left your generation?

[magic wave]

That is correct yes

[wispy canopy]

reference?

[sage kiln]

damn

[magic wave]

Reference preprocessor ye

[wispy canopy]

yeah, that one will be seeing a LOT of use in the future

[magic wave]

Yeee, its cool

[magic wave]

Yeah, its really good XD

[wispy canopy]

not sure if it going to be for me, because ... i need to click stuff

[magic wave]

I mean, you could like

Have a bunch of random pics of like random themes

And randomly apply em

Depending on what you feeling like

[wispy canopy]

my job is insert someting insane into an LLM, then ctrl+c->ctrl+v into SD, press generate, enjoy pretty pictures

[magic wave]

THOUGH

I also wonder

If it works with noise

Or gradients

[wispy canopy]

dude. how many do you think i have? i've literally engineerd the random prompt xD

[magic wave]

Yeah but I mean like, to slap onto the controlnet xD

[wispy canopy]

now. if that controlnet could be fed automagically..

[wispy canopy]

too many clicks.

lol

[magic wave]

:O

[wispy canopy]

paging @ocean reef seems my model has hidden husbando for ya 'jack frost'

[magic wave]

I think I'm going to try to see if something like this works tho

[magic wave]

Very peggable, I approve

[wispy canopy]

uhh... ok 🙂

[magic wave]

XD

Also

[wispy canopy]

whatever floats your boat ❤️

[magic wave]

I think my idea works

And I think it looks sick as hell

[wispy canopy]

https://tenor.com/view/good-evil-star-wars-emporer-palpatine-darkside-gif-4624104

[magic wave]

I'm going to try rainbow hair and rainbow eyes next

[wispy canopy]

somewhere long all the models, there has been training on jack frost, but it doesn't really recall what it should REALLY look like. this is my model performing in peak performance (it doesn't know the character, but a vague recollection)

[wispy canopy]

if you use that, you should use my 'scientists' from earlier today 😄

[magic wave]

It looks really weird but also dope

[magic wave]

Scientists?

[wispy canopy]

#🍥｜anime message science.

[magic wave]

Good idea :O

[wispy canopy]

curves

so. many. curves

[magic wave]

Infinite curves

[wispy canopy]

i dunno what u prompted though. i simply prompted scientist and used a reference one of my absolute thiccness prompts

[magic wave]

Oh, I tried to replicate the same charcter you did xD

[wispy canopy]

used my prompt?

[magic wave]

What is your prompt? :o

Just scientist and nothing else? :o

[wispy canopy]

the OG thicc scientist, or these?

[magic wave]

The OG one I suppose :O

[wispy canopy]

A plus-size woman stands confidently amidst a chaotic laboratory filled with strange devices and equipment. Her short, spiky hair is dyed bright blue, and she sports thick glasses perched on the end of her nose. She wears a stained white lab coat over a tight-fitting purple dress, revealing ample cleavage and curves. One arm is raised, holding a test tube aloft as if ready to perform some experiment. The background is cluttered with junk and debris, adding to the overall sense of disarray. Despite the mess, there's no denying the intensity of focus on the scientist's face as she goes about her work. BREAK lineart

[magic wave]

[wispy canopy]

https://tenor.com/view/good-morning-do-it-gif-24435329

[magic wave]

Aurora won't understand that tho

[wispy canopy]

you're right. it won't

that's why i hate that model xD

lol

can i introduce to darkalfa? 😄

[tepid geyser]

that is extremely tempting. I used to mess around with linux partitions before so I know how to do it, it just…. ughhhhh annoying

thanks for the tip though! the time difference is insane….

[wispy canopy]

if ur on windows -> wsl2

[magic wave]

It tried xD

But, I'm too used to booru prompting by now

[wispy canopy]

i have my LLM on ubuntu in windows

[tepid geyser]

for the purposes of stable diffusion making the linux partition on a HDD should be fine right? as opposed to SSD.

i just have more room on the hdd

[wispy canopy]

THE NAI HAIR AGAIN

[tepid geyser]

you have an llm locally on your machine?

[wispy canopy]

yes

[magic wave]

It do like that a lot

[wispy canopy]

the specific reason i went and bought a used 3090, for the vram xD

[wispy canopy]

i'm so glad my model killed that shit off for good

[tepid geyser]

wait what’s wsl2?

[wispy canopy]

it's gone. no more underglow hair. EVER. done.

[wispy canopy]

WSL = windows subsystem for linux

you can run a unix kernel NATIVE inside windows

[magic wave]

Is it bad?

[wispy canopy]

i HATE the NAI glowing hair with a vengeance

[magic wave]

I mean, the only thing I don't like about booru prompting is that its really depending on booru tag amount

And booru images being tagged correctly

Cause its really hard to do like tomboy style hair

[tepid geyser]

wait and would wsl2 allow me to use rocm like lizzie was saying? or at least get better performance using AMD?

[ocean reef]

handsome! ❤️

[magic wave]

Ohhhh, the uhhh "inner_colored_hair" tag basically xD

[wispy canopy]

i've been using it to use the GPTQ binaries that are available for linux, but not for windows, so i can fit that 30b model in my vram

[magic wave]

Also, more science

[tepid geyser]

you’re very welcome! i know before i was told about the different ckpt my outputs were so trash too LMAO

so much body horror 🫣

[wispy canopy]

so yeah, it shoudl be able to do rocm inside windows (because you are actually using linux)

[wispy canopy]

yes. my model can only do it if there's an actual light source that supplies the light to do it

[magic wave]

Fair

Question is

Can your model do tomboys and femboys

[tepid geyser]

basically name a popular male character and add ((genderbent)) and “[size] breasts” to the prompt LMAO

[wispy canopy]

why not just download the model 😄

i need more downloads anywy lol

[magic wave]

Send the link

O?

[wispy canopy]

it's NOT good at booru prmopts, i'll give that up front

[tepid geyser]

do you have a guide to set me in the right direction for setting this up by chance?

[wispy canopy]

wait whut?!

lets try this one https://civitai.com/models/64843/darkalfa-effortless-high-quality

[magic wave]

https://civitai.com/models/64843/darkalfa-effortless-high-quality This?

O

[wispy canopy]

it's not the best at booru prompts, but at least you don't need masterpiece either 😄

(that's right, IGNORE ALL quality prompts)

do use my negative prompt tho

[tepid geyser]

cute

[magic wave]

Civit is lying

[magic wave]

Fair

[wispy canopy]

i'm REALLY bummed it does worse than my accidental hentai model

[magic wave]

People horni

[wispy canopy]

where is the art

like my nsfw wheat fields!

[magic wave]

Fair XD

https://youtu.be/FyKPsua6Br8 Also, vibes

O

I accidentally pasted the screenshot here too

XD

[sage kiln]

Send it

[fallen sundial]

hello

[wispy canopy]

https://tenor.com/view/do-it-gif-26049397

[magic wave]

Doesn't look like it did anything tho

[fallen sundial]

I deleted venu and I got this after a lot of downloading

[trim current]

Cos youre rebuilding venv now

[fallen sundial]

so how to fix that ?

[sage kiln]

Try deleting venv and doing it again

[wispy canopy]

seems to have bugged out on git clone though

[steady grail]

for me, trying to restart after the error, would have done the trick

[fallen sundial]

I also get this

still

[steady grail]

every time?

[fallen sundial]

yes

[steady grail]

-you deleted the ven, restarted the program from the bat and you got an error?

[fallen sundial]

yes

[wispy canopy]

@fallen sundial i see you have spaces in your folder names

[fallen sundial]

now I'm restarting again

[wispy canopy]

don't do that

python doesn't like that

[fallen sundial]

yes

huh

[wispy canopy]

e:\ai-stable-diffusion\stable-diffusion

rename it so you dn't have spaces

[fallen sundial]

I cant change

[wispy canopy]

you have a folder or library open

[steady grail]

oh... i forgot why i removed spaces from names, that was the reasonù

[fallen sundial]

the name

[steady grail]

you are using a file inside the directory

[wispy canopy]

or the file is loaded and doesn't want to be unloaded

it happens

in that case, you need to restart

[steady grail]

i use unlocker in this cases

[fallen sundial]

Okay

btw Ai is very useful

I use ai for creating anime drawings for my youtube videos

and people thinks this is my drawing

but I try to mention that is not my drawing

[compact warren]

Just let them believe what they want ig

Until causes actual trouble for you

[sage kiln]

Until someone does hand spotting

[fallen sundial]

not as you thinking,they say these drawing are cool

[compact warren]

Well thats fine then

[compact warren]

Or background analysis

[fallen sundial]

what these messages mean?

do I have a problem?

[steady grail]

no

wait

[brittle jay]

guys maybe move to #🤝｜tech-support

[fallen sundial]

yes right

[wispy canopy]

the whole prompt: 👩‍🏫

[stone onyx]

CHontono!

wor!

[magic wave]

I got a idea

[magic wave]

@wispy canopy You think merging Aurora with Chillout mix would be good?

[wispy canopy]

only one way to find out!

[magic wave]

Tru

[wispy canopy]

(insert any of the 1000 do it gifs)

[gloomy scroll]

Good luck

[wispy canopy]

be amazed the crisp font tho!

[stone onyx]

It actually is crazy how clean it looks

[magic wave]

Thank

[magic wave]

I mean, it do, I love Kimante tho

[wispy canopy]

yeah, Kimante is best subject!

But my model = crisp lines city ^^

oh, and actual functional microphones 😄

[magic wave]

O, you rite :O

[wispy canopy]

ignore background singers, got this one on lead singer right too

even fingers are right 😮

she's hooked up on copium tho, as clearly visible

[modest arrow]

I can haz cheeseburger?

[magic wave]

Is she? :o

O ye she is

Or she has something sussy going on

[cyan burrow]

@wispy canopy non waifu, hentai and porn or those that got lucky early on when civitai was much smaller and still have a massive following now can get away with non hentai/waifu/porn

[magic wave]

I kinda vibe with chillout aurora tho

[cyan burrow]

@wispy canopy the rest of us though, well we should be happy with 500-1000 downloads

[wispy canopy]

alfahentai 1.0 and 1.1 got 1k downloads in total

[cyan burrow]

indeed

[subtle crown]

hassakuHentai is way better if you want to do hentai, in my opinion

[wispy canopy]

v1.2? yes

v1.0 -> no

which is the one i used for alfahentai actually

[subtle crown]

yes 1.2 😛 its what I use

[sage kiln]

First try at Multidiffusion with region prompts. Holy shit this is insane

[wispy canopy]

oh shit, she's about tho smite someone with a sun?!

[magic wave]

Also @wispy canopy you know where to get that offset model to merge it? :o

[wispy canopy]

the darkness offset?

[magic wave]

Yeee :o

Cause LoRAs are not that efficient :o

[wispy canopy]

left hand why... she was the perfect sun ❤️

[subtle crown]

try this one https://civitai.com/models/48139?modelVersionId=63006

[magic wave]

Oh thats a new one

Cause I have epinoise, sd 1.5, and locon, and only using those 3 combined it gets actually dark xD

[wispy canopy]

https://civitai.com/models/15879?modelVersionId=18744i used this one

ok. time to turn to bed. have this image on the lyrics to Cepheid - Goddess (and be jelly of my transparant fabric!)

[magic wave]

epinoise vs locon vs sd15noise

[wispy canopy]

might want to remove #2

[magic wave]

And all of them at once is the only way to actually get dark

[magic wave]

#2?

[wispy canopy]

oh 3

she has euhm.. the pokies

[magic wave]

Uhh, I mean barely, but sure :o

[wispy canopy]

i know, it's nitpicking

[magic wave]

Fair :o

[wispy canopy]

are you specifying WHAT should be dark/black?

applied locon at 0.6 but i didn't speicfy what should be dark, so background gets it

aight, now i'm off for real!

[magic wave]

O

Good nite :O

But ye I was

xD

[fallen sundial]

new animation , but without removing the face or editing frames

I will send it

here

wait a minute

or I will send it with editing

Wait me to finish it

[trim current]

making pretty wides and i wasnt tagged

[stone onyx]

Sori

I added little lizzie to the prompt tho

[trim current]

[stone onyx]

I still need to use ur prompt

Been slacking

[trim current]

me too dw

[fallen sundial]

is there someone can use sd rn ?

[lament osprey]

for what?

[fallen sundial]

for upscaling an image

4x

I need upscaling image

[lament osprey]

u cant do it>

?

[fallen sundial]

I can't use sd rn

it needs downloading

[lament osprey]

i think there are upscaling websites are out there, u dont need sd to upscale

[fallen sundial]

and my internet is good enough , so I would like to ask someone

[lament osprey]

so confused

[fallen sundial]

I can't find better than sd

other websites needs credits and like these and I used all of mine

and yes there are others but sd really upscale well

[fallen sundial]

@subtle crown could I get your help with sd ?\

[fallen sundial]

if someone of you guys don't mind helping me with sd

I just need upscaling

and also creating image if you don't mind

[lament osprey]

why can you not do it again?

[fallen sundial]

I need to download sd again

And I don't have a good internet rn

[lament osprey]

can't you wait till you have good internet?

[fallen sundial]

I can, but I need the images for my video editing

[magic wave]

May have new OC

[subtle crown]

sucks cant use any loras with WD, couse they all 1.5 lol

[magic wave]

Yehhhh

[terse ore]

Are there any samplers specifically better for anime?

[nova remnant]

a lot of models recommend DPM++ SDE Karras but in general the sampler isn't going to have an impact on the anime style, moreso the overal composition of the image

[untold glacier]

Erm, I'm seeing a visible toe of a camel there bud.

[magic wave]

I'm pretty sure thats reflection

I made a better one tho

And this one is pretty great, but reflections are a bit weird

O wait this is what I used for reference, I think I see why the crotch is like that now

[weak moss]

Does anyone know what this necklace is called?

[nova remnant]

bead_necklace?

[untold glacier]

That's a bit bloody for here bud.

[nova remnant]

she just had an accident with the ketchup

oh really, sorry

[untold glacier]

I'd at least put spoilers on it for those who aren't the biggest fan

[nova remnant]

sure thing

[untold glacier]

Thanks

[nova remnant]

LOL

[untold glacier]

Me rn in this channel

[wispy shard]

[untold glacier]

Cmon peeps

[upbeat wasp]

anyone tested out the new waifu diffusion beta? Downloading two of them now.

[modest anvil]

is there a way to generate diffrent colored hair with one prompot like red hair than black hair than brown

[magic wave]

https://tenor.com/view/perfection-x-men-michael-fassbender-magneto-reaction-gif-5305242

[upbeat wasp]

Need an extension that lets you specify prompt regions and split it into three prompts.

[modest anvil]

yea i tried that extension but shits crash my stable

[upbeat wasp]

Gonna have to outpaint instead I guess.

[nova remnant]

you can use tags like multicolored_hair and streaked_hair with the desired hair colors but it's not going to be consistent

so replicating the same pattern in multiple images is going to be tough

I'm sure it's fairly simple with inpainting but I don't really use inpainting so I can't help you with that 🙃

[untold glacier]

Less horny, more peace and tranquility...

[lucid rapids]

anyone know how to fix this?

[nova remnant]

did the model you use recommend a vae? that usally fixed the problem if i ever got those purple spots

[magic wave]

That is a no VAE issue, most models need a VAE

https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt
Get this, put it in in the "Stable-Diffusion" folder in the "models" folder
Rename the file type from ".ckpt" to ".vae.pt"
And then you can go to settings and apply it from there

[keen hollow]

evenin'

[lucid rapids]

ty

[magic wave]

Evening

[keen hollow]

how ya doin

[chrome nexus]

How do I create the illustrations?

[keen hollow]

most folks are using stable diffusion 'round these parts

[tropic wraith]

@chrome nexus

FAQ: How do I generate images? Is there a bot on the server?

Currently, there is no public bot on the server that generates images. There is an experimental bot available for early server members & Stable Society. You can obtain the Stable Society role by winning our weekly events #⭐｜pow-info #1087493421209485393! However, there are plenty of other ways such as the official https://beta.dreamstudio.ai/ website or by running Stable Diffusion locally using your own hardware! Check out #1080946152318443610⁠ for more details! You can also stop by ⁠#1025467151206854736 for any issues you experience while using DreamStudio or ⁠#🤝｜tech-support for any problems you encounter while installing it locally!

[slate narwhal]

https://huggingface.co/hollowstrawberry/stable-diffusion-guide

[supple raptor]

There's no need to change the extension for VAE's anymore. They can go directly in the .\models\VAE directory as .ckpt, .safetensors, or .vae.pt filetype. The renaming was only needed when we used to load VAEs by naming them the same same as the model with .vae.pt for the extension.

[compact warren]

Tron world

[magic wave]

Ohhh

Thats fair then

Still works tho

XD

[supple raptor]

Yep, I think the renaming trick will run into problems if the VAE is in .safetensors format though.

[chrome nexus]

Does anyone have a One Piece yamato prompt?

[slate narwhal]

yep, tronpunk

[supple raptor]

Can't link it since it's tagged NSFW, but theres a Yamato LoRA on Civitai. I doubt you're going to get very close to her with just prompts since she has less than 1k images on danbooru. Around 3k images is usually the threshold when characters start working on most models.

[subtle crown]

So WD, I hate it, this would pass like ok model 6 months ago, not now...

[supple raptor]

Which one are you using?

[subtle crown]

all of them

[nova remnant]

WD has a very specific style

[subtle crown]

ink and mofu prob the best ones, but its still bad

[nova remnant]

not my cup of tea either

[keen hollow]

let's see some creative images 🙂

[nova remnant]

you first

[supple raptor]

I just started downloading them. I'll reserve any judgement until I've used them for at least a couple of days. OpenCLIP requires different prompting than CLIP, so I expect some skill issues to start with.

[nova remnant]

lol model is no good

[chrome nexus]

I don't have time to install LoRA! Can someone generate an illustration for me?

[nova remnant]

pretty colors

[lament osprey]

got to love automatic1111 saying you somehow run out of memory trying to generate two 512x512 images

[nova remnant]

lol

what kinda gpu

[lament osprey]

pretty sure loras take about 15 seconds to download

[lament osprey]

idk

normally it can do 3 512x768 fine, so i dont know whats wrong right now

[terse ore]

The side of the guy's face is consistently discolored, any way to resolve that? (I use image to image but the original one does not have the discoloration)

[subtle crown]

https://civitai.com/models/65986/ruvimixstyle is not bad if your looking for flat look

[compact warren]

I mean the previews they showed looked pretty mid as well, so idt WD is a go to model for anything still

Its amazing how the ai community stagnated on 1.5 so long that now new models become inconvenient to switch too bc they dont have nearly as much flexible use as 1.5 has been modified to have

[compact warren]

More like 5

[nova remnant]

badoinker

[compact warren]

[subtle crown]

booba

[supple raptor]

Previews, and the 4 images I've generated on a single aesthetic model so far look better than base NAI. Even without modifying my old prompt. Based on the incredibly small amount of testing I've so far, I think it might be the best base checkpoint that exists. It shouldn't be directly compared to models that have been refined through MBW and finetuning for the last 7 months, with a bunch of LoRAs applied on top of that, it should be compared to the base models because all of that additional merging and training can be done on it too.

[compact warren]

I mean, even here, id think NAI is better. WD has this squishiness to the details that i dont like

Also, people keep saying that all it needs for models to be trained on it, and yeah i get that, but where are these models? People said the same thing the last WD release and i still have barely seen anyone use a WD model

People tried training on it at first but then after that i heard nothing about it

[supple raptor]

The last WD checkpoint was ass and pretty much everyone I saw said the same thing. It almost never got anatomy right and would randomly flip between realistic and anime style in the same prompt. The aesthetic version of that one showed more promise, but still wasn't good.

[keen hollow]

ok

time to generate something different

[compact warren]

So u mean to say we’re at a better checkpoint now? (No pun or whatever intended)

[keen hollow]

Pickle salesman.

[compact warren]

Pilch….

[supple raptor]

It seems that way to me so far, but again. I've only just barely started using it. I could certainly start running into major problems.

[keen hollow]

Yep that's the brand. Pilch Pickles.

[compact warren]

Well ill just keep my ears open in case things change this time

[keen hollow]

UFO abduction

[supple raptor]

2 girls, 1girl with blue hair standing on the right, and 1girl with pink hair sitting on the left,
Even if the model does still need more training, OpenCLIP in SD2.x is undeniably better than CLIP in SD1.4/1.5.

[keen hollow]

Choose your fighter

[nova remnant]

damn i like that concept

[keen hollow]

Yeah it was like a megaman thing

people in general didn't like it so i moved on 😛

[random valley]

who got good merges with Aurora

[magic wave]

I merged Aurora with Synergix at 0.5 WS, seems to be able to go good realism

[random valley]

Can you send me the link to the model?

[magic wave]

https://civitai.com/models/54592/synergix Tis one

[random valley]

I meant the merged model.

[magic wave]

Oh

Merge it yourself you lazybones, it takes like 5 minutes to merge the two

[random valley]

fine. I just hate having to deal with large downloads, especially with the colab down bad on fast stable diffusion

[magic wave]

I mean, they both make a good merge

Heres proof of it using my newly made OC

[nova remnant]

any LORA makers in here right now?

[magic wave]

Me, whats yo bussy-ness?

[nova remnant]

I'm pruning tags, adding some, just wondering if it matters if the focus tag is first in the prompt or if alphabetical is fine since the focus tag is in every prompt

[magic wave]

Uhhh, you mean like the keyword?

[nova remnant]

yeah

[magic wave]

I mean

[keen hollow]

you know, i've been farting with this stuff for months and have never ever created a model or lora or anything

[magic wave]

WD tagger puts it at the start by default

[keen hollow]

too much generating to do 😛

[magic wave]

Fair xD

[random valley]

can you use civitai links for nocrypt's colab?

[nova remnant]

guess I'll sort by frequency then, just in case

I'm using dataset tag editor

[magic wave]

Uhhh, not sure, you'll have to test, since I just run it on my pc

[random valley]

nvm, it works

[magic wave]

Ayyy das good then :O

[random valley]

is it a good idea to also merge that model with the epinoiseoffset?

[wispy canopy]

good morning 🙂

[magic wave]

Uhhh, not sure

[magic wave]

Gmorning

Also good morning @wispy canopy

(with extra good)

[wispy canopy]

lel, this early?