🍥｜anime

[untold glacier]

[untold glacier]

[subtle crown]

whos dave

[quiet stone]

🙂

Cat

[wispy shard]

perfect

[merry tendon]

i NEED you model

[queen void]

new LoRA :)

[wispy shard]

lol

[merry tendon]

THEN GIMMEEE, looks looo fucking sick

holy

[queen void]

xD

[merry tendon]

upload when?

[merry tendon]

bro you need to upload now im gonna die

holy that looks sick

[queen void]

its indeed a rly nice LoRA

[merry tendon]

@queen void when will you upload??

[queen void]

idk man

xD

I literally just trained it

[merry tendon]

is it done?

can you send me it in private ore here? no images or docs just plain lora

[queen void]

that might actually be one of my favorite LoRA's lol

[merry tendon]

yes mine too

thats why i wanna have it so bad

[merry tendon]

aahahhha

i am gonna go crazy

[wispy shard]

waifu informatic

[merry tendon]

i need this

[elfin bolt]

ur reaction has given me inspiration on what to generate next

[merry tendon]

bruh what XD

[merry tendon]

upload

it

please

i

need

it

in

my

life

[merry tendon]

what?

[compact warren]

Non

[merry tendon]

no

[merry tendon]

dont let me hang like this bro

[merry tendon]

will you upload today?

[queen void]

well its 1 am so, yes today, but not now :/

[merry tendon]

can you dm it to me when you upload it?

[queen void]

sure

[untold glacier]

Ok peeps. Anybody know how to neg prompt this type of hair shape between the eyes? hair_between_eyes isn't really taking care of the problem.

[merry tendon]

can you dm it to me when you upload it?

[untold glacier]

#🍥｜anime message

[merry tendon]

oh sorry didnt see it

[compact warren]

I think ur outta luck trying to get something that specific. The between eyes tag is literally the only thing

[merry tendon]

something with "bangs" probably

[untold glacier]

I wouldn't say it's specific. 😅 It's a general look to the hair. Guess Aurora is slightly overtrained on that type of hair.

[merry tendon]

crossed_bangs is the tag

[untold glacier]

Yeah. I guess I need to look up alternate booru hairstyles that may go in a different direction. Thanks.

[merry tendon]

crossed_bangs is the tag

just looked it up

[untold glacier]

Was referring to hair in the area above the nose, and space above the eyes.

[compact warren]

I was using that for a lora i was training the other day, that one usually details specifically and blatantly crossed bangs

Like across the face type bangs

[untold glacier]

Example: A similar but different bang style with the same problem.

[merry tendon]

yea thats crossed_bangs

[merry tendon]

ok idk what you want then 😅

[compact warren]

U want it lower?

[untold glacier]

I'd love to have these gone.

[compact warren]

Inpaint 😐

[untold glacier]

Spaces above the eyes. But it's likely just the model.

[merry tendon]

every model has that problem

[untold glacier]

You haven't heard of all the times I've complained on how inpainting doesn't work for me, lol.

[compact warren]

Probably not, im not here as often

But i get that

[untold glacier]

Either way, thanks for y'all's suggestions.

[merry tendon]

try "(widow's peak1.2)" in negative

[compact warren]

latest case of crossed bangs on my end https://media.discordapp.net/attachments/1071038975072026676/1102007945421656154/29341-1643437190-best_quality_digital_art_textless_version_battleship_water_oni_1girl_red_eyes__pale_skin_colored_skin_1.2_horns_singl.png?width=801&height=1203

https://media.discordapp.net/attachments/1071038975072026676/1102007925674872882/29339-2860235976-best_quality_digital_art_textless_version_battleship_water_oni_1girl_red_eyes_white_skin_pale_skin_colored_skin_1.2_h.png?width=801&height=1203

[untold glacier]

What an extreme example, lol.

[compact warren]

yup

wild oni hair

[untold glacier]

neg crossed_bangs though is definitely helping further. Thank y'all!

[merry tendon]

try "(widow's peak1.2)" in negative

[untold glacier]

Such a low number of tagged images though. We sure that works?

[merry tendon]

no

thats why i said "try"

[untold glacier]

[compact warren]

What did i do wrong

[untold glacier]

Lol. No, I'm getting a preview image for the BagsUnderEyes LoRA. Your suggestion works perfectly.

[compact warren]

Oh that oughta be neat

Wonder how itd mix with crazy expressions lol

Sleep deprived waifus ftw

[supple raptor]

someone talking about crazy waifus in here?

[magic wave]

Try harder?

[wispy canopy]

@merry tendon why the what did i dooooo?!?!?!!

[magic wave]

[merry tendon]

crying because it looks beautiful

[untold glacier]

[wispy canopy]

It's my llm bot, enhanced with my randomizer prompt, doing song lyrics! lol

[wispy canopy]

stablevicuna 13b is pretty damn good

[merry tendon]

you rich my boi?

how much vram does it cost?

[wispy canopy]

about about 4 for SD, 8 for the LLM

[merry tendon]

bruh

[wispy canopy]

i specifically went searching for a used 3090 lol

[merry tendon]

so only 8gb vram for llm?

[untold glacier]

[wispy canopy]

8-9.5 not counting inference or tokens

[supple raptor]

Which llm model are you using?

[wispy canopy]

stablevicuna 13b

[merry tendon]

LOOL i found my yandere lora on the frontpage of civit

[supple raptor]

yandere angels anyone?

[wispy canopy]

lel

[subtle crown]

the white horse thing with pink hair?

[merry tendon]

what horse bro

[merry tendon]

this is the main image

[magic wave]

[subtle crown]

then i was looking at something else lol

[untold glacier]

[magic wave]

[supple raptor]

'tis a masterpiece

[merry tendon]

what?

[supple raptor]

Your lora

[merry tendon]

you are using it?!

[supple raptor]

Yeah. I grabbed it like 2 hours after you uploaded

[merry tendon]

am i finally popular mom?

[merry tendon]

thanks for using it, do you have some suggestions on what could be better?

[supple raptor]

The irises can get a little messy at times. Other than that it's been working well for me even when mixed with other LoRAs. I like it around 0.8 strength, sometimes just a bit lower.

[supple raptor]

I used it to make this profile pic for one of my chatbots.

[merry tendon]

thats really nice to hear, the irirses are because there are many "retracted" irises in the dataset, but i just needed to add it to my lora 😉

[wispy canopy]

dang, stabilityai made a good vicuna model 🙂 it's pretty damn good at interpreting lyrics

[merry tendon]

looks sick as hell, if you get some good picks you can dm them to me, i will retrain the lora someday. i could use some pics.

[supple raptor]

Sure, I'll start a folder to collect some in and maybe try to fix up a few that are close

[merry tendon]

thanks, if you collected around 20 or more i would appreciate if you could give them to me, because(even if you would think otherwise) there are not many good yandere pics that are suitable for training.

[merry tendon]

rate my controlnet image

[real nexus]

HOW DID U MAKE THIS

ITS SO CUTE

[merry tendon]

this i think https://civitai.com/models/53294/twitch-emote-maker-lora

[real nexus]

ME AND WHO

[real nexus]

lmao

my own model

thank you for using it well youre doing it justice

❤️

[merry tendon]

lol didnt notice

[gloomy scroll]

That image should have pnginfo

I think I used part of your prompt

[wispy canopy]

oh right, you need that extension for it, right?

[real nexus]

yeah im glad people use the stuff i make makes me happy to see results

[gloomy scroll]

Didn't discord go back on that update

[real nexus]

i think they did cuz mine worked

[wispy canopy]

they did? 😮

[merry tendon]

i can agree, it makes me very happy if someone uses my loras. Btw, your lora is great❤️

[gloomy scroll]

Not sure since I used stealth pnginfo but I head someone say that they did

[real nexus]

thank you!!

[subtle crown]

you can view png on discrod its not blocked

[gloomy scroll]

I only used prompt from your image but not the actual lora

[real nexus]

i just realized u are the one who used it thank you but also thanks everyone<3

[wispy canopy]

llm wants me to prompt this? aight. can do 😄

[gloomy scroll]

But for that image I used one of my model merges _ANE_Pruned-Aurora-clip and just some style lora merge I did

[real nexus]

oooo i see

[gloomy scroll]

Since that model is pretty decent at making chibis

[gloomy scroll]

If someone interested here's stuff I have uploaded
https://huggingface.co/Surficialruby

[magic wave]

Harime Nui simping time

[wispy canopy]

❤️

[merry tendon]

she´s like 11

[untold glacier]

Not in that image she ain't.

[merry tendon]

model?

[wispy canopy]

my hentai model XD

which... is actually good at non-hentai

[merry tendon]

can you dm link please?

[magic wave]

You were saying?

[untold glacier]

[subtle crown]

well not like she needs a head to get ahead

[wispy canopy]

don't make me go get my thic waifus...

[magic wave]

[trim current]

wizardlm?

[untold glacier]

Finally got this LoRA to work semi-decently... very finicky though. For some reason the pov hand is always of a darker skin though

[wispy canopy]

nah, stablevicuna 13b and my lyrics->visual keywords prompt

although, i noticed it has some issues, so i'll have to redo that llm prompt again. AGAIN

[trim current]

very ni ce

[wispy canopy]

❤️

[magic wave]

Clyde bullies

[gloomy scroll]

[wispy canopy]

lol

[terse ore]

How consistent does a sequence of images need to be, to be "good enough" for a webtoon?

[wispy canopy]

i think... very?

lots of loras/embeddings

[untold glacier]

By "sequence of images" do you mean panels?

[wispy canopy]

and then it's still fighting against SD

[subtle crown]

if you want consistant images you need to train lora

[terse ore]

Yeah, panels. I was thinking making stuff in 3d and then putting it into img2img with lower desnoise settings to keep consistency

[terse ore]

I'm still experimenting with settings but it seems like it is not too bad; it takes out the obvious 3d look and keeps color the same

It could be a style where the character's hair stands change here and there (as long as the basic style is the same, like bob for example)

And then inpainting and manually cleaning eyes and hands

[gloomy scroll]

[lyric zealot]

why is my abg script not working?

[wispy canopy]

#2 is crying because she saw ur browser history...

[terse ore]

Do you think these are consistent enough for a webtoon? I know I would need to fix some things manually like the hands and small things on the clothes but I thought it might be okay to not have to get every detail right, like in the background for example

The hair is basically the same but not exactly the same

Ie it could plausibly be different due to the wind, or something

[wispy canopy]

colors seem to be shifting, as some elements, how their lighting is applied

[lyric zealot]

depends on how you want your webtoon to look right?

like in your 2nd example, there's close to no shadow

so if you are intending to go with no shadow webtoon, then it might even be easier to make background separately

for the characters wise, i think their colors and hairstyles look alike enough

all in all, if i was given a webtoon with these two panels, i would still be fine with it

obviously it's my own opinion tho

[terse ore]

I see, thanks. This is great feedback. I need to learn how to deal with light and shadow correctly now.

[lyric zealot]

not sure if you've seen it, but there's a controlnet module for that

haven't play with that yet, so can't advise on that

[terse ore]

Ah yes I remember, I posted that one on the chat. I need to use these tools. The example images I posted were only img2img upscale, so it can definitely get better with more tools

[magic wave]

Is anyone else's wildcards manager tab broken?

[gloomy scroll]

Maybe I need to make one of these too

[lyric zealot]

mine's broken for god's know how long

[lyric zealot]

you should!

[magic wave]

I was gonna try to use em

[magic wave]

How do you use it when it broken?

[untold glacier]

Wow. This is supposed to be the ThisIsFine LoRA, but it didn't work in the best way possible.

[lyric zealot]

I kinda remember how the one i usually used, type __ for autocompleting wildcard

[magic wave]

Imma try that

[lyric zealot]

so using that, if you know what you want, you can "search" for it

[lyric zealot]

this do be fine tho

[wispy canopy]

oh yeah. if you copy the stuff to thte wildcard folder

[lyric zealot]

I think you already start with some right?

[wispy canopy]

none.

the dynamic prompts folder has a collection

[lyric zealot]

for real?

[wispy canopy]

but it's not in the wildscards folder yet

[lyric zealot]

ahhhh

[wispy canopy]

so you copy it from collections to wildcards

and then you do

[lyric zealot]

i guess good point, so @magic wave you need to make sure you actually have stuff in your wildcard folder

[gloomy scroll]

I never got dynamic prompts wildcards working correctly so I'm just using some old wildcard extension

[lyric zealot]

don't fix stuff that's not broken

[untold glacier]

Such versatility. I love it.

[latent vector]

for some reason aom3 is better at producing furries than actual men in suits so i just have to stick to this for a while

[gloomy scroll]

@lyric zealot I'm just wondering did you sleep already or do when are you planning to sleep?

[lyric zealot]

actually, i just woke up

might be going out to cycle with friends later, so needed that rest

[gloomy scroll]

I slept well last night too

But... That means I wasn't too tired today

[lyric zealot]

that's a good thing right?

[gloomy scroll]

Well I'm 4h behind you

[lyric zealot]

well, it's a early Sunday morning! Time to get breakfast and chill

[lyric zealot]

hold on, my note says 5hours

[pastel bluff]

smol sus post on pixiv man

[lyric zealot]

guess, i need to update the note

[lyric zealot]

I'm lazy man

[lyric zealot]

Are you still getting a lot of ko fi?

[gloomy scroll]

wait its 6h? I'm too tired I rly can't count hours anymore???

[lyric zealot]

yup, definitely 6 hours

it's like 3am for you right?

[gloomy scroll]

4

[lyric zealot]

ahhh, i believe it's the daylight saving thing

it's always off by 1 hour

[terse ore]

Have you tried anythingv5?

[gloomy scroll]

Oh yeah we have that useless thing

[lyric zealot]

is that thing active now or something?

[untold glacier]

Meh. Terribad with LoRAs and other networks.

AOM3 is as well compared to AOM2

[lyric zealot]

anythingv5 is real?

[gloomy scroll]

We moved clocks some time ago but no idea what direction since all of my clocks move automatically

[gloomy scroll]

I believe so

[lyric zealot]

i thought it was anythingv4 that was released recently

[lyric zealot]

lmao, that's kinda weird tbh

[gloomy scroll]

That was Andite's troll

[terse ore]

Yes that was my experience too and I kept trying when I was new nothing worked. Wasted so much time on it

[lyric zealot]

isn't there a real anythingv4?

or they just went and skip that?

[gloomy scroll]

I think they skipped it and said that it has nothing to with them

[untold glacier]

Anyv4 made by someone other than the person who made Anyv3. Anyv3 guy went and made Anyv5.

Anyv5 is on the Anyv3 page.

[lyric zealot]

well, not like im gonna use anythingv5 anyway

just curious about it

[latent vector]

isnt that a troll

or something

or v4

i forgot

[gloomy scroll]

I wonder what happened to Andite since I haven't seen them on any server for long time now

[gloomy scroll]

v4 & v4.5 were trolls

[lyric zealot]

he dipped from the AI scene right?

[terse ore]

Me too but I remember seeing it so yeah

[latent vector]

makes sense

but v4 wasnt that bad at generating simple images tho

[supple raptor]

He got banned from Touhou, then hung out here briefly and disappeared since then.

[scarlet kelp]

Yandere Lucy

[untold glacier]

v4 wasn't a troll, just different. v4.5 was a troll.

[gusty escarp]

v4 and v4.5 were both by andite so both trolls

[latent vector]

so

[copper marsh]

v4 and v4.5 were both by andite so both trolls

[latent vector]

which

was

[gloomy scroll]

v4 was andite's model too

[untold glacier]

v4 worked well. v4.5 didn't.

[latent vector]

or wasnt

[latent vector]

v4 and v4.5 were both by andite so both trolls

[gloomy scroll]

[untold glacier]

Troll or not, v4 was great. v4.5 wasn't.

image not related to topic

[lyric zealot]

lmao

[latent vector]

what model do yall use for faces

[untold glacier]

Aurora

[gloomy scroll]

If you sharing more ping me so I remember to check when I wake up

[latent vector]

oh damn

[untold glacier]

Good... night?

[latent vector]

its a nice model

[lyric zealot]

for sure for sure!

@gloomy scroll goodnight btw

[pastel bluff]

yes

[lyric zealot]

daymn!

[pastel bluff]

get that bag bro

[lyric zealot]

I'm too lazy for that

[pastel bluff]

i see u are too rich

[lyric zealot]

nobody's rich enough

[pastel bluff]

rich enough that extra 400a month is not worth

[lyric zealot]

not true, but just really really lazy

will do that someday tho, for sure for sure

[pastel bluff]

japanEse ppl are really generous

ngl

i have 250 subs now

[lyric zealot]

that's why realistic model with Asian stuff works so well

probably not for anime tho

[pastel bluff]

lol anime is more popular by far

realistic is niche

[lyric zealot]

actually true

but maybe people that are into realistic are more willing to pay for stuff like that

might not be true tho

im not sure tbh

[lyric zealot]

Cursed

[subtle crown]

big brain

[lyric zealot]

actual 5head

[rapid shuttle]

I love this style

[rapid shuttle]

This is fine

[rapid shuttle]

How did you fix this?

[lyric zealot]

^ @gloomy scroll almost forget

[rapid shuttle]

I'm just getting started

[supple raptor]

You wake up to find your new self-appointed guardian angel standing over you

[untold glacier]

"work" is a bit strong of a word, but just making sure the following was included in the prompt: pov, pov_hands, (finger_in_another's_mouth:1.3), (open_mouth), (fiday:1), fang,
As well as using high strength weightings on the LoRA. Around 1 or so.

[pastel bluff]

pro

[lyric zealot]

just using lora lmao

[rapid shuttle]

radical

[pastel bluff]

so humble

[lyric zealot]

That’s actual fact tho, just using people’s Lora

[pastel bluff]

https://tenor.com/view/peachandgoma-peachmad-couplegoals-peach-and-gif-22394109

[trim current]

Controlnet? Thats amazing

[slate narwhal]

No, just pure txt2img

[trim current]

thats really cool

[slate narwhal]

Thanks 💖

[slate quarry]

wow, it look very cool

[trim current]

wasnt the style im after but its nice

[keen hollow]

evenin'

What's cookin'?

[supple raptor]

https://cdn.discordapp.com/emojis/998672574345121855.gif?size=64

[latent vector]

nah cuz what she reading

[slate narwhal]

it is a newspaper prompt

[supple raptor]

might be time for vedal987 to run.

[past gazelle]

hello, i have been trying to recreate this format - full body shot with zoomed-in face in the background. is there a particular name for this style, or any prompt that can help?

prompt i used to generate these two:

(finely detailed beautiful eyes and detailed face)lighting, perfect face, latex suit, full body
Negative prompt: sketch, duplicate, ugly, huge eyes, text, logo, watermark, monochrome, worst face, (bad and mutated hands:1.3), (worst quality:2.0), (low quality:2.0), (blurry:2.0), horror, geometry, bad_prompt, (bad hands), (missing fingers), multiple limbs, bad anatomy, (interlocked fingers:1.2), Ugly Fingers, (extra digit and hands and fingers and legs and arms:1.4), ((2girl)), (deformed fingers:1.2), (long fingers:1.2),(bad-artist-anime), bad-artist, bad hand, extra legs, EasyNegative
Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 7, Seed: 3068623869, Size: 512x768, Model hash: 969673ad74, Model: darkSushiMixMix_colorful, Denoising strength: 0.5, Clip skip: 2, ENSD: 31337, Hires upscale: 2, Hires steps: 20, Hires upscaler: R-ESRGAN 4x+ Anime6B

the closest i could get are these, using reference sheet in the prompt

[magic wave]

Try openpose controlnet

[terse jacinth]

a wild smile

[magic wave]

Also what model is that?

[past gazelle]

i'll have to learn that , thanks

[magic wave]

A wild Floydian

[past gazelle]

dark sushi mix

[magic wave]

Its not too hard, theres a few extensions to do it, and a app, though I'm not sure if the app is available anymore

[magic wave]

I'll have to check it out

Seems like a good model to train a LoRA off it

[past gazelle]

thanksss ❤️

[past gazelle]

how do u get these zoomed-in shots? can u share the prompt plz

[untold glacier]

It's already in the img PNG Info

Negative prompt: (EasyNegative:1.2), ((broken feet:1.3)) ((bad-fingers:0.8)) ((text:0.8)), (((character design:0.7))), ((2girls, 3girls:1.0)), Reference sheet, ((watermark)), ((no body:1.2)), ((no hands:1.3)) ((basic background:0.8)), ((three legs:1.0)), (machines:0.6), ((extra limbs:1.3))
Steps: 28, Sampler: UniPC, CFG scale: 8, Seed: 1912475651, Size: 568x640, Model hash: 31c9be0ee8, Model: Defmix-v2.0, Denoising strength: 0.6, Clip skip: 2, Hires upscale: 2, Hires upscaler: Latent (nearest)```

[past gazelle]

ahh my bad, will check frm next time. thankss

[untold glacier]

Hey @cunning gyro how are you getting your Superscale 178000_G to work? It's a RealESRGAN model right? I've tried placing it in both my ESRGAN and RealESRGAN folders and SD is still not seeing it. Are you using something other than Auto1111's build for SD?

[cunning gyro]

Nope, A1111 (now vladmandic, but still). IT goes in models/esrgan

[untold glacier]

Hmm. Wonder why it's not seeing it then.

[cunning gyro]

if you just installed it, try refreshing the upscaler list?

[supple raptor]

patch notes in Auto1111's Release Candidate branch imply that the RealESRGAN directory may be broken in some manner in other builds. He seems to be working on a fix. Not sure if the Release Candidate branch is truly stable or not.

[untold glacier]

I've been restarting SD completely 😅

[cunning gyro]

o

[untold glacier]

Wait, found it! Talking to y'all was magic apparently.

[untold glacier]

Hmm. Good to know. Thanks Nubbs

[cunning gyro]

https://i.gyazo.com/269db703e942edd8acf97bfe07a0fa79.png

[untold glacier]

Could account for why it was then wasn't able to see it

[cunning gyro]

is how i have it

[untold glacier]

Yeah, it's able to see it in my ESRGAN folder

I wanted an anime-friendly painted-like look without the darkness that Fatal provides, and I remembered you talking about it.

I look forward to giving it a spin.

[cunning gyro]

I've been starting to question a few of my underlying assumptions as to how some of my workflow works, and I think its possible NMKD may not work like I thought, but let me know how you feel about it

[untold glacier]

Will do!

[slate quarry]

https://tenor.com/view/小猫吃惊-what-amazed-gif-15665604

[untold glacier]

Took the time to load it up. Only had to try it once.
Compared with something like Ultrasharp or Fatal, it had a heck of a lot of noise and visual artifacts comparatively...

[cunning gyro]

yeah so i forgot but i actually did some tests with it earlier when trying to scope out some changes to my mix

Left is NMKD, right is Fatal

you may have to blow it up really high to see the differences, lol

but there is more noise on the NMKD one, but it ends up giving it a "canvas"-y look i guess

Whereas if you zoom in on the Fatal one its kinda pixely at places

[untold glacier]

Yeah, harder to tell with those, especially as they're either different seeds or different prompts. And is that straight ESRGAN upscaling or with SD upscale as well?

[cunning gyro]

nope, just img2img with postprocess upscale

[random valley]

bruh, prompt and model. Now

[untold glacier]

Gotcha. SD must not like what NMKD does to the image then. I usually test upscalers like this in the context of SD Upscale, as that's almost exclusively what I use them for.

[cunning gyro]

Still hard to tell. left is nmkd, right is fatal

[south sable]

https://tenor.com/view/suzaku-kururugi-gun-point-aim-gif-13318800

[cunning gyro]

[untold glacier]

It's right in the PNG Info

[south sable]

https://tenor.com/view/cyber-gif-20990497

[untold glacier]

https://tenor.com/view/rascal-does-not-dream-of-bunny-girl-senpai-mai-sakurajima-anime-point-gif-17742393

[cunning gyro]

stable diffusion create a dress without a long train train challenge (impossible)

[untold glacier]

Yeah, look at the crispness of the eyes in the NMKD one. Slight, but more noise. SD must just not like it.

[south sable]

https://tenor.com/view/scaleway-scale-scale-it-gif-20254562

[untold glacier]

https://tenor.com/view/spy-kids-lemme-zoom-in-on-that-zoom-watching-you-spy-gif-5212510

[south sable]

https://tenor.com/view/take-it-to-the-next-level-rich-benoit-rich-rebuilds-escalate-it-step-up-gif-26371439

[untold glacier]

https://tenor.com/view/boom-gif-5722488

Love the vector art LoRA though.

[south sable]

I forgot about it. now to try it out >.>

[untold glacier]

I don't know if I got it before it disappeared, but I couldn't find just now on Civitai.
If you want it and can't find it let me know.

[south sable]

Sometimes I just download things and never touch them T_T

close but weird hand. still nice ish

[untold glacier]

Great flat shiny style.

[cunning gyro]

this one? https://civitai.com/models/34321/vector-graphics-art-style-no-lineart

[untold glacier]

There it is!

[cunning gyro]

i saw that earlier, lol

[untold glacier]

No wonder I couldn't find it again. It's a LoCon. I forgot to add that back into my search settings.

[south sable]

https://tenor.com/view/anime-buttons-switch-controls-gif-17290671

trying it now 😄

[untold glacier]

[south sable]

I didn't hit enable

https://tenor.com/view/dead-oof-fall-anime-oh-gif-19412815

[untold glacier]

Color me impressed. I gave Interrogator the image on the left and it gave me the image on the right! 🤯

[south sable]

I do love interrogate

[wispy canopy]

good morning and... eh.. disturbing morning?

[untold glacier]

[south sable]

[untold glacier]

[south sable]

[mild tree]

Gif mania today

https://tenor.com/view/yugioh-anime-omg-wtf-cant-unsee-gif-5159766

[wispy canopy]

lol.

[untold glacier]

Sorry sir, you missed it. Gif mania was from 6pm-7pm GMT.

[wispy canopy]

does her gasmask have a gasmask?!

[queen void]

maybe

[random valley]

canvas zoom is the one crapping out

[untold glacier]

Regardless, I deleted it at the time and Controlnet was the one that was making my UI not work. It wouldn't even install correctly when I tried to reinstall it from scratch, citing an access error.

[nova remnant]

i honestly don't know what people see in anime2

too much red

anime2 is too saturated and anything vae isn't saturated enough

i think clearvae is a good balance

[untold glacier]

Never heard of clear. I used blessed2 or anime2. The main problem that I deal with when working with anime models is that they're often difficult to saturate properly, or get enough contrast. anime2 helps immensely with that.

[nova remnant]

this was generated using walnutcream blend

[untold glacier]

I use Aurora.

[supple raptor]

All of the [redacted] based VAEs give me a hideous hexagonal texture on any backgrounds that are detailed. I can't stand any of them. I'll take oversaturated images any day over ugly textures.

[untold glacier]

Even without upscaling? I've found usually such texture patterns in the gens are the result of overzealous upscaling processes. Or SD Upscale interacting with them in a way that is detrimental to the end result. Anime6B would be a lot less likely to have this problem as opposed to say, Ultrasharp.

[pliant epoch]

yeah everyone has personal preferences but I don't like the dithering effect that every vae besides anime2 has

[untold glacier]

It definitely depends on the networks and models being used. There are times when anime2 is blurring my lines in a simple flat color anime gen. Other times I'm getting bad patterns or desaturated results using blessed2. Clear looks interesting though--I'll have to try it out and see its worth as a tool in my box.

[supple raptor]

I almost never gen without hires fix but the same thing happens regardless of upscaler and anywhere from 0.4-0.65 denoising with every sampler I use and I've never seen a single image from anime2 with the same problem.

It'd be nice if mse could do anime eyes decently because that's the best out of all of them colorwise.

[untold glacier]

I see the same problem usually with tennis court nets, fishnets, up-close pantyhose, etc. But that's usually it. SD generally has a problem with those with the inherent moire effect anyway.

[supple raptor]

Nah it's a totally different problem than you're talking about. This is fine hexagonal textured dithering pattern across the entire background. It's almost like a screen door effect

[nova remnant]

can 1.5 based models generate at 768x768 well?

[untold glacier]

Depends on the model, but often yes.

Hires fix is always our friend though. It's good to keep at least one side around 512 and then upscale from there. If you're using a 16x9, 4x3, etc.

[nova remnant]

i'm going to try it with walnutcream

[supple raptor]

Usually they don't start breaking until either both dimensions go above 768 or one dimension goes above 1024 for modern models.

[nova remnant]

people say latent sucks bit it's pretty good to me

[untold glacier]

Your hires fix is showing something weird. Uncheck and recheck it and it should be good to go.

[nova remnant]

wat

[nova remnant]

oh

[untold glacier]

Your width and height is the same in the interface, and yet it shows as something different in the resize.

[supple raptor]

You're using super high denoising so it covers up the blurriness that latent gives.

[nova remnant]

eeeee

[brittle jay]

Me when I have phobia of redacted vae

[supple raptor]

Just a visual bug in my experience. No impact on generations.

[untold glacier]

[nova remnant]

i originally used 0.5 but i noticed that 0.7 gave more appealing results

[brittle jay]

Latent doesn't suck, just upscalers are most of the time better

[cyan burrow]

multidiffusion

[nova remnant]

i got the 4x-animesharp one but i couldn't find that other one that nubby talked about

[brittle jay]

Multi diffusion is hacking

[cyan burrow]

Multidiffusion is great

[brittle jay]

https://upscale.wiki/wiki/Model_Database

[supple raptor]

Latent can be good. When it works well on an image it can be the best option, but it can often be the worst.

[brittle jay]

Fatal?

Nmkd?

Download both

[nova remnant]

idk why i thought i could upscale 768x768 at 2 when i can't even do it with 512x768

[brittle jay]

6gb vram😎

[nova remnant]

i have 8gb

OutOfMemoryError: CUDA out of memory. Tried to allocate 2.25 GiB (GPU 0; 8.00 GiB total capacity; 5.01 GiB already allocated; 470.40 MiB free; 5.60 GiB reserved in total by PyTorch)

[supple raptor]

guess the upscaling method

[nova remnant]

is stable diffusion under utilizing my gpu?

[compact warren]

does anyone else have the extension for this? i have it installed but i dont see it

[supple raptor]

I have it. Have you completely restarted SD since you installed it?

[nova remnant]

it seems like it's only using 5gb of my vram

[supple raptor]

Looks like you're using about 2.5GB of vram for something else. The most common culprits are hardware acceleration in your browser and/or discord, wallpaper engine, and nvidia broadcast.

[nova remnant]

ic

[compact warren]

i honestly dont remember installing it actually

[nova remnant]

i turned off hardware acceleration on my browser and discord

[supple raptor]

Should help a lot

[nova remnant]

i guess i probably need to restart SD too so it can re-allocate

[supple raptor]

Yeah, I've found that SD sometimes kind of shits itself after getting an OoM error so I usually just restart after it happens.

[nova remnant]

does adding --xformers to args automatically build and install it?

[supple raptor]

Yep, but if you've manually installed pytorch 2.0 you'll need to manually install xformers as well so that you get the right version.

[nova remnant]

well i didn't do that so all good

[supple raptor]

Here's an example of the [redacted] vae texture I was talking about. This is blessed2 to be specific, but it happens with all of them.

[nova remnant]

are you talking about the "scales"

[supple raptor]

yep

[nova remnant]

i didn't even notice until you said something about it

[supple raptor]

Sorry, you'll probably never be able to unsee it now.

[nova remnant]

i think it makes the background look like it's painted on fabric

disabling hardware acceleration on discord and chrome only increased the reserved vram to 6.25 GB

[nova remnant]

oh, edge has hardware acceleration on too

[lyric zealot]

morning!

[nova remnant]

(i use chrome for my daily use and edge for my college work)

[supple raptor]

Ah, yep. Pretty much every browser has HW acceleration turned on by default.

[gloomy scroll]

[nova remnant]

i wonder if the spotify desktop client uses hw acceleration too

[supple raptor]

https://cdn.discordapp.com/emojis/1101433307343892490.gif?size=64

[nova remnant]

aha

[supple raptor]

Open task manager, go to details tab, right click any of the column labels, click "Select columns", and enable "Dedicated GPU memory", sort by that column high to low and now you can easily see all of the processes that are using vram as well as how much.

[gloomy scroll]

Is that win11 feature?

[supple raptor]

I use Win 10, not sure if 11 has it too, but I assume it does.

[gloomy scroll]

I just can't find it in win10

[nova remnant]

you have to right click then click "select columns"

[gloomy scroll]

I kinda expected that to mean something else

[supple raptor]

Sorry, I missed that step.

[nova remnant]

windows 11 task manager is funky

i have windows 11 on my laptop and it looks like this

[supple raptor]

I've used win11 for like 20 minutes while I was setting up my mom's printer for her. I was not impressed.

[nova remnant]

i changed things around to make it more like windows 10

i had to edit the group policy to make it so you only need to right click once in the file explorer to get the full menu

[supple raptor]

It was mostly the control panel and settings menu that I found annoying.

[nova remnant]

yeah i don't really like how they make it hard to get to the in depth settings

or how you click on something in control panel and it opens the settings app

[supple raptor]

Yeah, I've hated everything about the settings menu since they started implementing it.

[nova remnant]

i looked at the details tab

the only 2 things with significant dedicated gpu allocation other than python are explorer and "dwm.exe"

[wispy canopy]

dwm is the process that's drawing your desktop, so you can't really go without it

[nova remnant]

this is what it looks like after i get the cuda out of memory error

wait no it went down

[supple raptor]

Yeah, that's what I have too. I haven't found a way to reduce dwm vram usage.

[nova remnant]

python would be at like 7,300,000

[wispy canopy]

you can't reduce dwm, unless you decide windows doesn't need any way to draw to your screen. lowering your resolution, less virtual desktops, less color space, that kind of shit helps

but it's marginal (mb's vs required gb's)

[nova remnant]

would turning on medvram actually let me generate higher resolution images in exchange for time?

[supple raptor]

If they hadn't taken away our ability to reduce desktop animations, I imagine we could reduce dwm vram usage quite a bit.

[nova remnant]

here we go, this is what it looks like directly after the out of memory error

[wispy canopy]

yeah.

it's a marginal speed hit for pretty substantial memory buff (reduction)

[gloomy scroll]

Wouldn't tiled VAE help with this also?

[wispy canopy]

it would. but tiled vae is pretty bad in initial txt2img

[nova remnant]

i'm trying to generate 768x768 at x2 upscale with hires fix

[compact warren]

moves to disable

[wispy canopy]

make sure that the tiled diffusion /vae only activates on the upscale part

[nova remnant]

i don't know how to do these things

[wispy canopy]

(check your console and see to it that it doesn't trigger on the initial pass)

[nova remnant]

i don't even know how to get tiled vae

[compact warren]

How do u control that

[wispy canopy]

vladmandic has it out of the box, extension for a1111

[wispy canopy]

check the tile sizes -> if the tiles are 'big' enough, the image can fit in a 1x1 tile, so the console will mention it will not split the image

[compact warren]

By saying that unnecessary warning?

[nova remnant]

this one?

[wispy canopy]

yes

[supple raptor]

Using --no-half-vae so that you can use [redacted] vaes is probably killing your vram usage. Mine tops out at 6.2GB usage with just --xformers enabled doing 768x768 with 2x upscaling.

[nova remnant]

i'm not using no half vae

oh nvm i am

i forgot that i was

[supple raptor]

That's with 2 LoRA's too so it would be even lower without those.

[wispy canopy]

this warning: if this triggers only on the upscale image rendering, you're golden. if it triggers during initial txt2img you're in for a world of weird shit

[nova remnant]

i have disable safe unpickle on because i was trying to use old .ckpt models and they weren't working

probably not the best move in hindsight

[gloomy scroll]

Does webui use deterministic xformers or do I need to update manually?

[wispy canopy]

xformers by nature are non-deterministic

you'd have to switch to pytorch 2 and go for --opt-sdp-attention

[supple raptor]

You have to install torch 2.0 and manually install xformers.

[supple raptor]

not any more.

[wispy canopy]

they fixed that?

[gloomy scroll]

Ok can't bother to do that

[nova remnant]

can i change the order of these extra things

i want the aspect ratio helper closer to the top

[queen void]

interesting

[supple raptor]

yep. the current build is deterministic.

[wispy canopy]

the result of txt2img with wrong tiled diffusion (on purpose)

[gloomy scroll]

What is that page?

[nova remnant]

must be moderated by UD mods

[wispy canopy]

yeah, it's very provocative nerf, totally too lewd 😛

[queen void]

Civit

[supple raptor]

that's tiled diffusion, tiled vae can be used independently of that and won't give that sort or weirdness even when it activates on the initial pass.

[queen void]

https://civitai.com/models/54424?modelVersionId=58787 Well whatever, its uploaded xD

[wispy canopy]

i haven't found a use for tiled vae yet tho 😦 does it help? i ran out of 10gb on my 3080 regardless wether i used it or not

[nova remnant]

free gpu

ok anyways what do i change here

to make it not horrible

[supple raptor]

Yeah it can significantly reduce vram usage. default tile sizes are too high though. Encoder should be around 768-1024 and Decoder should be around 64-96.
This image was done with 512 encoder tile size BTW.

[wispy canopy]

seems i'm spoiled by my 3090 now 😦

[supple raptor]

try these settings

[nova remnant]

should you change it based on what the target resolution is?

[wispy canopy]

i'm so going to hate summer.... the 3090's fans are perfectly positioned for the airflow in my case. the case exhaust is working REALLY well and... i'm literally heating my house with inference xD

[random valley]

How does one do Dark night images that look bluish?

[nova remnant]

i probably need to replace the exhaust fan on my PC case
it has been dead for over a year

[supple raptor]

I never really do unless I need to make it smaller to avoid OoM.

[nova remnant]

all my 1070 has is a blower shroud

[wispy canopy]

i can really recommend https://github.com/Rem0o/FanControl.Releases

[random valley]

I can't seem to get that blue night look with minimal lighting

[nova remnant]

also i got this when tryint to use tiled vae with those settings you suggested
RuntimeError: The expanded size of the tensor (3) must match the existing size (8) at non-singleton dimension 1. Target sizes: [1, 3, 96, 96]. Tensor sizes: [8, 96, 96]

[wispy canopy]

that way you can get your case fans to go max power when your gpu gets hot -> more cool air in, more hot air out = cooler gpu

[gloomy scroll]

noise offset lora and theme blue?

[random valley]

like prompt (theme blue)?

[gloomy scroll]

Yeah but I rly have no idea if it does anything

[wispy canopy]

blue theme, yeah that would work

or blue hue?

[nova remnant]

so what do i do about this? it only happens with tiled vae enabled, even if the image is small enough to where it says tiling is unnecessary

[wispy canopy]

i think it was caused by another extension

cutoff or controlnet? i had that too at times

[nova remnant]

i don't have those enabled

i don't even have cutoff

[wispy canopy]

blue theme vs blue hue, something like this?

had one each for red, blue, yellow ❤️

[supple raptor]

I've never had that error.

[gloomy scroll]

Is your width and height dividable by 8?

[wispy canopy]

combine with cutoff to make sure colors are isolated properly (pink shirt, black hair, colored hues)

[nova remnant]

yea i was trying 512 and 768

[queen void]

now combine it with my LoRA :D doit

[wispy canopy]

link plox

[gloomy scroll]

Yeah 1.0 was too much

[nova remnant]

these are the settings i'm using

[wispy canopy]

@queen void pina artstyle?

[nova remnant]

the vram estimator is buggin out because i set the resolution higher than 512

[queen void]

yes

[wispy canopy]

of. disentegrated the image

lemme try on another prompt tho

my model is pretty clear about what it likes and dislikes, and this kind of deformations -> stuff that's incompatible is going on

[wispy canopy]

0.8 on a image that is inherently unstable, actually stabilized it

cool how the red/blue lightning get a different interpretation from the yellow lightning xD

[queen void]

nice

try alittle lower though liek 0.6-0.7 with these models

[nova remnant]

THE JAR

[brittle jay]

School be like

[wispy canopy]

@queen void pina vs no pina

[brittle jay]

Looks a bit overfitted/overtrained or something

Like the 2 tries looks too similar compared to the 2 tries without it

[queen void]

model depended kinda. It was trained on aurora

[wispy canopy]

my model is pretty distant from aurora tho

[brittle jay]

Oh, makes sense

[brittle jay]

Probably why it's like that then

[nova remnant]

tiled vae does not work if i have hires fix on

[brittle jay]

Guys what is tiled vae?

[wispy canopy]

lemme see what my prompt does on aurora

[wheat falcon]

It does normaly

[brittle jay]

And how to use multi diffusion thingy?

[wispy canopy]

i feel my model has a quirky prompting style, good at doing descriptions, but lackluster when doing it booru style

[wheat falcon]

Tiled vae is part of the multidiffusion extension, and can save you a lot of vram when generating highres images by tiling the vae

[brittle jay]

So like. It seperate image into tiles and then works on them one by one? All at the same time?

[wheat falcon]

It takes a bit longer but you wont get out of vram errors anymore

[brittle jay]

So it works on them one by one?

[wheat falcon]

Only for the vae

[brittle jay]

Ye

[nova remnant]

i keep getting his
RuntimeError: The expanded size of the tensor (3) must match the existing size (8) at non-singleton dimension 1. Target sizes: [1, 3, 88, 88]. Tensor sizes: [8, 88, 88]

only happens when i have both hires fix and tiled vae enabled at the same time

[brittle jay]

Oh yeah btw CS1o, for a long time now I wanted to get the latents before they get transformed into an image. How can I do that?

[supple raptor]

I would make sure everything is updated.

[nova remnant]

it has been a while since my last git pull...

[brittle jay]

A1111 is back

[nova remnant]

hmm

[supple raptor]

Yep, he finally made a dev branch. The RC branch looked like it just got everything thrown into that the dev branch had though from a quick look so I'm unsure how much testing he actually did with the dev branch.

[nova remnant]

is the dev branch like nightly builds or something

anyways i did a git pull and all my extensions are updated, i still get the error

[supple raptor]

That's basically the intention

[wispy canopy]

i have no idea why 'regular' models hate my prompts xD

[wheat falcon]

Idk sry

[queen void]

i like it though

[wispy canopy]

seems my model doesn't like the lora that much 😦

[nova remnant]

so i guess no tiled vae

one result

https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111/discussions/152

this might be the fix

it worked

the solution was to replace the vae_optimize.py file with a different one

[cyan burrow]

Try this prompt: man of liquid metal chrome, red glowing mechanical eyes, vivid color, shimmering metallic short hair, brutalpunk fashion, transmogrifying arms into blades, slices nosferatu demons, 3d digital illustration, digital art, [futurepunk:dieselpunk:0.60], dystopian fantasy art style, dark fantasy mood, cyberpunk aesthetic, hard lighting, filmic angle, low contrast, medium saturation, 8k, HDR

Though I dunno what weights fit or negative. Still in hospital bed but my prompt eye is still trying stuff

[pastel bluff]

BASED

Gantz author likes AI, doesnt care his works has been used as learning material

[nova remnant]

he's posting ai generated girls on his twitter

[pastel bluff]

yeah

hes using my model

i wonder if i can get to say hi

[brittle jay]

Epic

[brittle jay]

Omg I would be so extremely happy and honored if I were you

[pastel bluff]

gantz is pretty goated

[random valley]

Why are my images looking at the ant's eye view?

[brittle jay]

I just did (hands up:3) in negative

and it came out fine

this is what happens when negative prompt becomes too long

[sullen hollow]

when youre at your 5000th render and the hands are still poorly drawn 😎

[sullen hollow]

those are all embeddings?

[scarlet kelp]

text embeddings

[sullen hollow]

ok will try them out 😀 thanks

[brittle jay]

EZ

not all are that good

just try them

[sullen hollow]

probably depends on the model and sampling method as well i imagine

[brittle jay]

sampling method?

why would that matter?

[brittle jay]

only model

[sullen hollow]

ok

[brittle jay]

before and after lora. should I release?

(sorry for loli, model does that by default and I didn't put any negative other than 2 embeds)

[sullen hollow]

has definitely a more natural and realistic feel to it. looks good imo

[brittle jay]

thanks that's how it should have been

[real nexus]

not anime but my new model

[brittle jay]

I actually should retrain it though. (this is one of the training images so you see what I mean)

it's unusable beyond epoch 1

and epoch 1 doesn't resemble the training images enough imo

[brittle jay]

very sus

[nova remnant]

civitai's search doesn't seem to be working

[slate night]

Yeah it was giving me trouble yesterday/this morning

It would work from the main page but nowhere else

[nova remnant]

the new layout sucks

[teal pivot]

hiii

[brittle jay]

e dating?

bruh

[stone onyx]

[brittle jay]

what this all about?

this a meme or you actually do it?

[stone onyx]

oh

you mean my profile

[brittle jay]

yes

[stone onyx]

I'll leave it for your imagination

[brittle jay]

come on man

I need to know

[stone onyx]

It's a meme

that went a bit too far

[brittle jay]

a bit too far??

[stone onyx]

[pliant epoch]

There should be an icon near the filter where you can switch to the old layout

[brittle jay]

https://tenor.com/view/plink-wide-cat-plink-cat-meow-gif-27396868

tell

story

story time

[stone onyx]

https://tenor.com/view/cat-lurk-kuss-kus-gif-24016787

[brittle jay]

https://tenor.com/view/1984-big-brother-discord-mods-moderators-gif-20619192

[nova remnant]

Once there was an Ugly Barnacle. He was so ugly, that everyone died! The End.

[brittle jay]

noooo

now I wanna use pyg7b to tell me story time

[brittle jay]

lora?

[ocean reef]

fashionGirl and ExperimentalMakeup, model nabimix

[stable crane]

happy to hear that 🙂

[brittle jay]

lottalewds epico

https://tenor.com/view/give-my-heart-to-you-girl-gif-12972296

[magic wave]

Isn't there like two background removing extensions?

[scarlet kelp]

made with @real nexuss new Model and Jinx Lora

[real nexus]

makima my boo

[magic wave]

https://github.com/KutsuyaYuki/ABG_extension

There is also this in extensions which may be better

I haven't tried it tho

[regal pivot]

Online ... https://huggingface.co/spaces/skytnt/anime-remove-background

[magic wave]

Theres that one too, I think that one works the same as ABG

[magic wave]

Damn thats a nice style, I'll have to try that

[queen void]

mhhh

[brittle jay]

sir

you ones are great

ofc

[queen void]

[brittle jay]

who wants me to dm them image I can't really post here?

[trim current]

That ABR extension

Anime background remover

Huggingface also has a space running it

Oh scrolling up smone already linked it lol oop lol

[magic wave]

Also which version are you using

[ocean reef]

newest 😄

[magic wave]

Fair fair :o

[ocean reef]

I try to update as soon as there is new one 😄

[magic wave]

Fair, they make so many updates tho

[ocean reef]

yeah, sometimes they are not worth downloading after reading description what was changed

[magic wave]

Yeee xD

64 dim? how is it 144 mb then

Also, NAI VAE? so it was trained with a vae?

[ocean reef]

whut, I have lora:fashionGirl_v54:1 only 😄

[magic wave]

Oh

I mean I downloaded it right

[ocean reef]

weird

I've downloaded it in a classic way

[stable crane]

does a1111 have a built-in downloader now? wat

[magic wave]

Weird

[ocean reef]

indeed

[magic wave]

Its the Civit Helper extension

[stable crane]

why tho

click button, save file

[magic wave]

Saves trigger word, link to the page, preview, and some other info I think

https://github.com/butaixianran/Stable-Diffusion-Webui-Civitai-Helper You can check it out here

[stable crane]

oh ok

too bad those things aren't packaged into the lora format

[magic wave]

It neat

[stable crane]

resumable downloads not bad

[magic wave]

Yeeee

[stable crane]

speaking of the lora preview image

I keep accidentally clicking that, and so now all my lora have the wrong preview pics

[ocean reef]

Thats why I always click on the upper part of image 😄

[real nexus]

das a very cute chibi model

[magic wave]

XD

I just click the middle xD

[stable crane]

feature is a menace

[magic wave]

ALSO @ocean reef what weights are you using for the two LoRAs?

[ocean reef]

Usually between 0.6-0.8

If I get annoying result I lower it even to 0.4 xD

[magic wave]

Tried it

[ocean reef]

dat latex tho 😄

[magic wave]

I love latex

[ocean reef]

😏

[real nexus]

i need help

cuz i really dont know the style of this model

is it considered anime? semi realism?