🍥｜anime

[bitter latch]

hmm didn't work out.. almost tho

There

haha a poster for a new game system?

[gusty palm]

im scared to try relatives of comfyui i think ill install a111 standard i had good results with that, UX forge had so many features i didnt need, and i couldnt get i2i to work the way i wanted it to

[gusty palm]

altho now that i know where it stores temporary files i might just try it, i mostly just wanted to avoid bloating my hdd with all the stuff it needs to download

[bitter latch]

goes to bed.. THUD

[gusty palm]

https://www.instagram.com/reel/DB9_BNPyfmY/?igsh=MWM2aWVpbTljMnlycw==

tarek.em

He just wants some cuddles #catsofinstagram

Original video: @wulong1997 on TikTok

#weirdcore #weirdpets #creatures

Likes

1050833

Instagram

this typea this is what I wanna figure out

is the mov2mov extension all it would take ?

[near elbow]

looks oily enough to me 🙂
"(((oil painting thick brushstrokes expressionist))) anime miku"

van gogh

[supple raptor]

eh.... not really. it sort of tried on the left, but ultimately just added some lines that only sort of got the idea, meanwhile it continued to do it's standard generic anime style for miku. btw. emphasis doesn't work on flux. the ((( and ))) there are just being tokenized

[vital raptor]

it's fine until you notice the character is not the same style as the rest

[supple raptor]

it's not a bad looking image to be fair. the model just isn't nearly flexible enough in style without loras for me to be interested currently.

[supple raptor]

i would probably be less disappointed in flux if i hadn't seen what the earlier versions could do

[vital raptor]

did they 'refine' it for censoring or something?

[supple raptor]

that being said there are some efforts going on to de-distill flux.snell which would allow proper finetuning of it and could result in some outstanding models in the future

[supple raptor]

idk what exactly happened with it, but earlier versions were much more flexible in style and also seemed to have a little better character knowledge

[gusty palm]

https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell

is this wat yalla re talking about

[supple raptor]

sadly SD3.5L is dead in the water for multiple reasons imo despite having a lot of potential

[supple raptor]

yep

[gusty palm]

do u need like a big vram to run it

or does the safetensor file size not matter

cuz its like 20gb i was trying it in the web thingy last night but wasnt sure if i wanted to try running it locally

i liked the results the web thing was giving me but u have to wait like 5 minutes between each render

[vital raptor]

depending on your hardware, you might have to wait 5 minutes anyway

it takes almost a minute for a 4090 to do a flux image with 2x upscale

sdxl is ~10 seconds for comparison

that's with the dev version of flux though, they all have their own vram requirements

[gusty palm]

i think i like dev better

[supple raptor]

realistically, anything below 12gb it's going to be painfully slow to run even with the quantized versions.

[gusty palm]

ah ok, ive g ot a 1060

[supple raptor]

dev is definitely better than schnell. the main reason that the community is interested in schnell is the licenses. dev doesn't allow commercial use which makes it unusable for many of the people/groups who actually have the resources to finetune it

[modest fog]

Waiting for pony v7

[gusty palm]

this schnell i like it

[supple raptor]

it's being trained on a base model that has a lot of problems with some of the same and a whole new set of mistakes as v6. don't hold your breath. might as well just use noobai if you have to have e621 training (or even if you don't since it's the best anime model available that just also happens to have furry training)

[zinc flax]

Noobai is already pony v7

Both are bad lol

[modest fog]

Noob is ugly

[zinc flax]

And ugly

Though everything is ugly by your standards so idk

[supple raptor]

noob knows damn near every artist on danbooru and a huge number from e621. if you can't find a style in noob that you like, that's a skill issue.

knowledge wise, it's far ahead of even NAID3

[modest fog]

Yeah don't like just throwing artists into prompts

Everything I made with noob was utter chaos

[supple raptor]

just throwing an artist into a prompt is boring. the interesting part is mixing multiple artists to make completely different art styles

[modest fog]

I want some 2.5d stuff for it

[zinc flax]

Idk, still have that 3d pony style

It's not subtle either

[modest fog]

Does auraflow even matter that much

We are using sdxl

[vital raptor]

i managed to get some good results from noobai after i found some artists that worked, i'd have to spend a lot more time to get something i really like, just like how i spent a lot of time running stuff through pony

i pretty much just decided that there wasn't much point in me finding the right combos for noobai when pony already does what i want

[supple raptor]

auraflow is overcooked on ai outputs. there's a reason nobody else is showing interest in training it

[modest fog]

I didn't use it enough

Looked like a simpler sd3 to me at the time

@green temple

Bloo smol

[supple raptor]

literally over 100k different styles to pick from and mix before even factoring in loras

[zinc flax]

I'm just too biased to illust/naiv3 artist styles then

[supple raptor]

wait, were you using noob v1?

[zinc flax]

The weird thing about noob is that unfinished 3d render look

[zinc flax]

Tried all of noob actually

[modest fog]

Furries and their 3d stuff

[supple raptor]

v1 got overcooked. 0.75 is a lot better

[zinc flax]

I'll give it one more chance with the final vpred version

[supple raptor]

0.5 is more similar to illustrious, but increased stability

[vital raptor]

i was actually thinking of running some tests on noobai compared to results i get in pony (autism) because pony has "thousands" of artist/styles but when you really nitpick the minor details, there's actually very few distinct styles i could find

[zinc flax]

It's hard to explain but even from 0.5 it loses that "rough realistic coloring" ish

[supple raptor]

it's just western artists, which have a much lower average quality

[modest fog]

Not anime?? Banned

[modest fog]

Nyoo, I mean 3d stuff like render

Pony can go heavy into that it looks so weird

[supple raptor]

tough to really directly compare because they both require significantly different prompting

[jagged field]

[zinc flax]

Should've used kemono instead of western furries

[zinc flax]

Why do we need e621 again lol

Do people need to prompt sex that in depth

[vital raptor]

yeah, i don't mean 1 to 1 comparisons, just running through a list of artist tags with the same prompt/seed to see how different they really are (or not)

[supple raptor]

their tagging system is a lot more detailed and just overall better than danbooru. it isn't without it's merrits

[zinc flax]

Couldn't they just yoink the tagging only

And keep using danbooru for the dataset

[supple raptor]

possibly, but then you have a very confusing dataset that nobody knows how common tags are across the dataset

[zinc flax]

And then we'll have tag bleed if they don't curate both danbooru and e621 tags properly

Nobody wins lol

[vital raptor]

it doesnt even work with all prompts i noticed, seems more tokens makes it worse, but last night i was changing out artists with a prompt i had and about 30 of the 35 artists i used produced the same (or very similar) results as not using an artist at all

[zinc flax]

Also pony hashed most of the artists there so it's not surprising if there's no difference most of the time

Some are still there though, one example would be mdastarou

Great artist

[vital raptor]

some of the tokens I used seemed to be influencing the style quite a lot as well, so only 'strong' artist tags made a big difference

[gusty palm]

flux dev kinda thirsty

ux-forge has a cool feature styles-integrated where u can pick like a style and it inserts the words into the prompt

[gusty palm]

heres the list if u want it for inspiration or anything

[gloomy scroll]

Nom

[jagged field]

SD3.5 Large

[keen hollow]

style/lighting & shadows on bottom left is cool

[keen hollow]

what's your go-to negative with noobai?

[modest fog]

Sooma usually leaves png info

And nubby

[modest fog]

Tried to reply to this

[green temple]

everyone should know the smol prompts

[modest fog]

And big too

[green temple]

[modest fog]

[green temple]

big-smol

[modest fog]

Bloo

@green temple give more smols

Only smol can make day better

[green temple]

indeed

[modest fog]

Big

[modest fog]

@green temple wtf

I slept for like 10 minutes

I thought I was sending stuff

Was smol in dream

I'm just bored

[green temple]

u were smoling

[modest fog]

Not me

I had my phone in my dream in the same position I'm in

I thought it was real lmao

[green temple]

lol

[modest fog]

I usually know I'm in a dream when I'm in one

[green temple]

oh 😂

it happens

[modest fog]

Sleepy

[modest fog]

Why this smol in 240p

[green temple]

[gusty palm]

I like this

[steady grail]

wow, cool style

never saw Rei with such frizzy hair, but it's cool indeed

[steady grail]

i just had in mind a new experiment on comfy

[brittle rampart]

try it. why not?

[steady grail]

your image inspired me!

[brittle rampart]

I'm glad

[jagged field]

Hello

[near elbow]

strange, with forge there is definitely an effect with (())...

same prompt and seed, just "((((((question marks))))))" instead, does have an effect

and i love how all the anime gals are using macs 🙂

[brittle rampart]

not true

[bitter latch]

hmm not what I expected

[brittle rampart]

awesome ears!

[bitter latch]

I said small pointed ears not elf ears..sigh

maybe I should say vulcan

[near elbow]

"anime style, a laptop" - looks very macbooky to me 🙂

may be a flux bias as well

[near elbow]

okok...

[jagged field]

[near elbow]

oops, gotta run - cu

[bitter latch]

ahh ha more anime

the upscale of the DTG extension did it again... 😄

why does it keep adding eggs?

[steady grail]

a bias of the model, or a token leaking

[bitter latch]

yep vulcan ears work better

[supple raptor]

there's an effect because you're adding additional tokens to the prompt. the effect is not emphasis though it's just extra text that the model doesn't understand

[jagged field]

in Moochi i have 160gb of vram and it says out of vram like wtf

[bitter latch]

you has to much vram and it doesn't believe you and errors?

[brittle rampart]

need to share!

[jagged field]

Okay wait

[jagged field]

I'm trying to do 61 frames

[steady grail]

eh? why?

they are not usually set at 60?

[jagged field]

no by default the workflow comes at 37

but when i go up to 61 and above i did 121 it comes to the vae decoding part and gives me this error

I'm using the Moochi vae decoder

[brittle rampart]

oh well!!!.. maybe you can just share some extra memory?

[jagged field]

Oh i see xD

[jagged field]

I've changed it to tiled vae decoding it doesn't give me this error

[steady grail]

i'm saying that usually, videos run at 60fps

[jagged field]

no it's the video frames size

more frames = more seconds

i did like 13 seconds with Moochi and i think you can do more if you want

[steady grail]

ohh, how long is the video

not how many frames in a second

[jagged field]

Yes but this is defined by number of frames per steps i think i'm not sure

[desert island]

about 2 seconds

[jagged field]

I'm talking about the video length yes but idk know what 61 refers to if not number of frames

[steady grail]

maybe a video 61 seconds long?

[jagged field]

wtf ?

[desert island]

it is the number of frames

[jagged field]

61 seconds LOL

[steady grail]

so i was right

it's the fps

[jagged field]

yes i told you this

[desert island]

no In the documentation it says that that is the number of frames

[steady grail]

fps is frame per second

that means that there are 60 frames in a second

that's the standard in modern monitor

there are faster monitor

but the standard is usually 60

[desert island]

that's not what that node is refering to.

[steady grail]

you 2 have conflicting info, i can't gen video, so, i can just interpreter what you are saying

the node is clearly generating a 61 blocks long noise

[supple raptor]

sounds like it has a fixed fps and the frames parameter adjusts the number of frames generated, and in turn, the length of the video

[desert island]

https://youtu.be/mG1YqJK1p0g?si=-lBMNII8JKFNSdzK&t=493

[jagged field]

it doesn't know will smith rip

[steady grail]

it's weird

it's obviously slower than 60 fps

maybe that's 30?

[jagged field]

that's 49 frames 25 steps and only at 8 fps

you control fps at the video combine node

[steady grail]

oh

[desert island]

seriously? I just sent the video with all this information and linked right to the spot in the video that talks about this.

[steady grail]

why i should study video making when i can't use it?

[jagged field]

someone sent me a workflow for mochi for i2v

I'm gonna try it out

[desert island]

Then why are you spreading misinformation?

[steady grail]

i have some notion of video rendering

and his standards

and indeed that video doesn't look fluid

there was a reason if standards where 30 and 60 fps

[jagged field]

we don't do 60 fps video

[steady grail]

i suppose is too hard for now

[jagged field]

because to do a 60 fps video you have to have a a lot of frames for that

for example cog videox5bi2v doing max of 49 frames

if you calculate it 49 / 8 fps = 6 seconds clip

[bitter latch]

yeah.. 60 of them.. who knew

[jagged field]

you can make the model generate 60 frames and put the video on 60 frames but it will be 1 second only or 2 seconds clip

and you will pay a lot of money for that

[steady grail]

i know there was an app to fill artificially the empty space between frames to make video more fluid

[desert island]

you can use 8fps if you're on 3's

[terse ore]

For noob XL I understand it does much better with artist tags. But I don't know artists. So I will have to browse around and pick artists I like and then use that as a tag?

[steady grail]

they converted a video from 30 fps to 60 fps

it was a very interesting argument

[supple raptor]

these were made on illustrious so won't be completely accurate for noob, but will give you an idea of what to expect https://mega.nz/folder/Vd8yTCTL#FEpX0epT6cAcMZQI-QuBAA/folder/pF13hC7C

[desert island]

https://www.youtube.com/watch?app=desktop&v=1QC7mroZ9MA&ab_channel=DevonKong

[steady grail]

damn, so many memories

i love to see how animations progressed with time

[modest fog]

Grabby

[clever shore]

Looking into this, is there a guide or recommended args for the trainer? I found some for colabs but they're confusing to follow. Also, since you said to use e621 tags, is there auto tagging for that specifically? I remember using sd webui extensions in the past with booru tags.

[keen hollow]

evenin'

[near elbow]

old school

[green temple]

`

[modest fog]

[green temple]

"evil" in the prompt, so she is evil

[modest fog]

I put good girl

I'm the pink blob on the back

Fat

[sinful fulcrum]

Whats the consensus on the stable diffusion model that does anime pics the best?

[green temple]

[vital raptor]

either a pony based model or illustrious/noobai

[green temple]

📝

[modest fog]

https://tenor.com/view/rrane-battal-rrane-battal-kızgın-battal-kızgın-gif-25355985

[spare glacier]

long angle full body image character sheet of Ravana, Huge muscular body, Indian mythology, The legend of Ravana, with a dark evil background of Lanka city of gold, hyper realistic, in the style by Cedric Peyravernay, 32k, --ar 8:9 --stylize 750 --niji 5

[keen hollow]

Nice, I like it.

Been messing around with this noob thing, and ya'll are right--using artist names has a significant impact.

Exact same prompt, only changing artist.

also "superhero landing" is a fun tag 😛

[green temple]

👍

[keen hollow]

running an artist grid right now... tempting to run like 2000 artists 😐

[keen hollow]

Oh hello.

I see what you did there.

I like that weird lightsabery one

[desert island]

yeah I wasn't sure what was happening there but the interpretation of motion was kinda neat.

[keen hollow]

yeah the motion blur is cool

I'm running Tifas through a bunch of artist tags right now

I need someone that's already done a huge one and picked out the best though. 😛

[desert island]

can't wait to see what you post 😄

[keen hollow]

ah it's hard to post because the grids are humungo

(and i've been doing the web-page grids from swarm... not even sure how to share those)

[green temple]

trying to get ruby from jewelpet twinkle , pls help

[desert island]

this? https://jewelpet.fandom.com/wiki/Ruby

[green temple]

yup

[green temple]

i got it , but it`s a bit sus

[keen hollow]

There we go... there's the grid haha

Does this work? http://www.fragpub.com/grid/gridtest

would be cool if that's all it took to share a swarm grid

[green temple]

[green temple]

we want a grid for smol

[keen hollow]

Sure - what's the prompt?

and what's a smol? lol

[modest fog]

@keen hollow nai is dum dum, doesn't know what superhero landing is I think
@green temple smol

[keen hollow]

hmm it works for me

[modest fog]

Nai as in novel ai

not noob

[keen hollow]

oh

[green temple]

child, loli, kind of artists

any prompt

[keen hollow]

I try to stay away from those, they get weird.

[green temple]

ok np

[keen hollow]

ha... is there a particular character?

[green temple]

no

just the style

[green temple]

this is "smol" 👍

[vital raptor]

aged down sometimes works enough to avoid the unwanted gens while still getting smols

[keen hollow]

How about Yuffie, also from FF7?

[green temple]

nice

[vital raptor]

on autismmix (pony) i've been using loli, aged down just because 'child' tends to gen a second character that is a child instead of making a younger character for whatever reason

searching through danbooru tags, most images tagged with child has multiple characters though, so that's probably why

[green temple]

oh yup sometimes, i think "solo " could avoid getting a second character

[vital raptor]

it helps, but it wasn't helping enough so i switched the tags and it's more consistent now

[green temple]

nice

[vital raptor]

and as long as I prompt for some kind of clothes, i don't end up with a bunch of nsfw gens

[keen hollow]

so noob contains a vae baked in I suppose... wish I could turn that off

[vital raptor]

you can just use a different one

[keen hollow]

it's so bright

[vital raptor]

there also a way to strip the vae out if you really needed to using something like super merger *a1111/forge extension),

[keen hollow]

i tried setting to 'none' but that didn't work 😛

meh, not that big of a deal

[vital raptor]

well, you have to have one, it can't just be none

so either built-in or a specific external one

[keen hollow]

gotta find a nice external one that's more matte/muted

[modest fog]

plastic hair

[vital raptor]

i was thinking of that the other day, my current artist tag mix makes the colors pretty saturated

[modest fog]

That doesn't look too bad

[vital raptor]

same image with a 'pastel' vae

[modest fog]

Food

[vital raptor]

a little too faded tbh

lol, my upscaler makes it even worse

[modest fog]

@green temple she fights back

[green temple]

smol fighter

[modest fog]

Evil smol

[vital raptor]

😄

[modest fog]

Angy

[vital raptor]

fighting is such a heavy token, it changes the images a lot

[modest fog]

Depends on the artists you're using I think

Some just put a lot of stuff in the image

Evil'er

[modest fog]

chicken fingers

[modest fog]

Is that ear just pointy ears?

[vital raptor]

yeah, not sure what's with this particular artist mix, it almost always does 3 fingers

[vital raptor]

yeah

[modest fog]

@green temple click to see cute smol

[modest fog]

I am manually doing the super hero landing pose

@green temple your kryptonite

[green temple]

big

[modest fog]

Pillows

I'm fat

Can't move after eating lunch

[green temple]

@modest fog were is bloo ?

[modest fog]

I'm evil sad cat

🔪

[modest fog]

Some.. Shark airplanes?

had to cut the image

But cute

NAI dumb the more nsfw negatives I put the more nsfw it gets

Stupid

@green temple bloo

(what if she ate poisonous frog?)

[modest fog]

tentacle monster and hungry flying giant bird ready to eat food on the sea

[green temple]

[modest fog]

Smol food

A normal gen, miracle

@green temple how bloo is this

[green temple]

30%

[modest fog]

Needs more blue

Time to blue everything

@green temple
Runs away

Your payment for making all blue

[green temple]

why payment is my life

[modest fog]

No life, just a piece

[green temple]

[desert island]

If Attack on Titan & Solo Leveling had a baby.

[keen hollow]

manually?

[modest fog]

yeah type all the prompts that make her do the superhero landing pose lmao

[keen hollow]

not just 'superhero landing' ?

[modest fog]

nai is dumb, cant do it

[keen hollow]

ah yes you mentioned that

[modest fog]

[keen hollow]

these artist tags are a blast

[rough oxide]

yeh artist tags r fun

[modest fog]

https://tenor.com/view/theyre-the-same-picture-the-office-pam-the-office-us-gif-20757621

[modest fog]

Not that arcane character

[rough oxide]

i never saw arcane

or played league

but i like jinx design both reg+arcane

[modest fog]

Good

I'm too old for those things anyway

[keen hollow]

evenin'

[keen hollow]

in the grass

[vestal ferry]

https://cdn.discordapp.com/attachments/1106777714284822530/1304882332876017734/0lp91r4mtlmd1.gif

[green temple]

don`t eat

[vestal ferry]

https://tenor.com/view/creepy-smile-anime-grin-cat-gif-17290912

maybe

[vestal ferry]

send more

https://tenor.com/view/트릭컬-리바이브-트릭컬-코미-코미-backflip-computer-gif-13479146599765963639

[green temple]

oh , u like it ?

[vestal ferry]

yep

[green temple]

thx

sure

[vestal ferry]

you are using a lora style

or only artist styles

[green temple]

no only artist style

[vestal ferry]

try this one

[vestal ferry]

(toi8), yd_\(orange_maru\), clearite, as109, rhasta, binggong_asylum, ebifurya, mochirong, enn_matien, yanje, ask_\(askzy\), (terufuu),

https://tenor.com/view/maygi-smile-cute-vtuber-anime-girl-gif-9573395351183115989

[green temple]

👍

i know 2 of them are sus but i will try

[vestal ferry]

hehe

did you get a nice result?

[vestal ferry]

mmm

its better with yours

[green temple]

yup

[vestal ferry]

https://cdn.discordapp.com/attachments/1106777714284822530/1304882030596722790/updated-doro-gif-the-rest-comment-v0-m5izf32htlmd1.gif

@green temple

look

[green temple]

nooooo only cuteness is allowed

the style is nice tho

[vestal ferry]

well im trying to use a better upscaled number

like 1,4 or 1,6

[green temple]

i don`t use upscale

too much time

[vestal ferry]

i will try one without using

[green temple]

ok

[vestal ferry]

damn

i dont like it at all

[green temple]

lol

[vestal ferry]

https://tenor.com/view/uzaki-laugh-xd-gif-18137240

[green temple]

i dunno why when i use upscale the image becomes burned

[green temple]

i will try to use upscale on this

[vestal ferry]

oki

[green temple]

why is she laughing at me

[vestal ferry]

because its evil hehe

https://cdn.discordapp.com/attachments/1249717548778262618/1282208825507844118/39313390C6F7A2061B8BBFB5E99F9333.gif

[green temple]

this is with upscale 512x768 x2

weird, it`s not burned

[vestal ferry]

trying to use upscaled 1,6 and 1000x980

[vital raptor]

different upscaling methods can cause different results, and there's different upscaler models as well

[green temple]

okkk

i usually upscale with "extras " if i want to

but when i generate with hires fix on , images usually getting burned

[vital raptor]

with a1111/forge i used hiresfix with 0.2 denoise and the 4x-UltraSharp model

[green temple]

i use 4xanimesharp

maybe 0.2 denoising works well, i will try 👍

[vital raptor]

i tend to keep it low, otherwise it changes too much of the image, I also only did like 7 steps

[green temple]

this is with 0.5

[vital raptor]

https://civitai.com/models/116225/4x-ultrasharp if you want to try the upscaler i use

[green temple]

oh yeah i have 4x-ultrasharp, but i usually use anime sharp

[vestal ferry]

can you gave me the normal prompts

the general ones

[green temple]

metadata is in the image

png info will work

[vestal ferry]

oki

do you wanna see my style

[green temple]

if nsfw don`t send pls

[vestal ferry]

here cant show nsfw

[green temple]

kawaii

[vestal ferry]

the hands are off

[vital raptor]

the first set or the second? 😄

[vestal ferry]

the second one is more visible

the first the hand of the doll is wrong

[green temple]

what`s the artists u used here ? same artists ?

[green temple]

?

[vestal ferry]

nope

my private lora

https://cdn.discordapp.com/attachments/1106777714284822530/1304882044337262592/updated-doro-gif-the-rest-comment-v0-amrtonfhtlmd1.gif

[green temple]

oh ok

[vestal ferry]

but the difference its from the negative prompts

and this

8k, (high detailed skin), (masterpiece), (best quality), (ultra-detailed), beautiful color, amazing quality, very aesthetic, illustration, absurdres, newest, perfect composition, realistic, cel shading,photorealistic,Cinematic Lighting,beast quality,HDR, highres,

negative prompts

lowres, worst quality, normal_quality, old, early, (outline, muted_color, pastel_colors:1.2), signature, username, logo, (extra:1.1), (missing:1.1), bad hands, 4_fingers, 6_fingers, mammal,anthro,ambiguous_form,feral,semi-anthro, (scan artifacts, scan, jpeg artifacts:0.9), film_grain, confetti, (backlighting:1.1), (silhouette:1.1), pale_skin, hair_intakes, colored inner hair,unfinished, downscaled, pixelated, lowres, artistic error, fewer digits, extra digits,

https://cdn.discordapp.com/attachments/1106777714284822530/1304882318900596776/doro-nikke.gif

[green temple]

i will try them with my prompt

[green temple]

actually u don`t need all that negative prompt , try to simplify it

[vestal ferry]

https://cdn.discordapp.com/attachments/1106777714284822530/1304882318900596776/doro-nikke.gif

you can simplify it

[zinc flax]

That's a lot of quality prompts

[green temple]

yup

[vestal ferry]

the signature

or logo is missing in negative

https://cdn.discordapp.com/attachments/1106777714284822530/1304882332876017734/0lp91r4mtlmd1.gif

or make another one

[vestal ferry]

yep

[green temple]

this is your negative 🌚

[vestal ferry]

https://tenor.com/view/nikke-doro-dorothy-pipe-bomb-gif-4643514375044855612

(watermark, signature:1.3),

used this

[supple raptor]

Signatures and logos are basically a non-issue. Just use this to remove them https://github.com/Sanster/IOPaint

[modest fog]

Use smol magic on me turn me young again

[modest fog]

monster

[vestal ferry]

you save me

https://tenor.com/view/creepy-smile-anime-grin-cat-gif-17290912

[modest fog]

Ugly

[modest fog]

I'm irreversible

[vestal ferry]

https://tenor.com/view/doro-dorothy-nikke-goddess-of-victory-nikke-creature-gif-17030571333828640261

[supple raptor]

There's also a HF space for it which is slow, but fine for a single image occasionally. It takes like 2 seconds when you run it locally on your GPU so it's worth it to install if you're going to use it much. https://huggingface.co/spaces/Sanster/iopaint-lama

[vestal ferry]

i want to try to take of extra fingers

[vestal ferry]

https://cdn.discordapp.com/attachments/1106777714284822530/1304882044337262592/updated-doro-gif-the-rest-comment-v0-amrtonfhtlmd1.gif

[modest fog]

Just draw the hand again

Manually

[vestal ferry]

https://tenor.com/view/nuh-uh-punch-gif-14008503396845178733

[modest fog]

[modest fog]

Ugly vtuber

[green temple]

https://www.photoroom.com/tools/remove-object-from-photo i sometimes use this , try it if u wan`t

[modest fog]

Grabby

[vestal ferry]

https://tenor.com/view/pain-sad-anime-loli-gif-20376542

https://tenor.com/view/onimai-anime-sleep-mihari-oyama-gif-27609322

[green temple]

indeed all vtubers are ugly

[modest fog]

Smoking bad

[green temple]

smol smoking !!

[modest fog]

Ban him

[green temple]

i like the prompt lol

[vestal ferry]

https://tenor.com/view/loli-anime-waiting-excited-gif-7426168

[keen hollow]

isn't this just stable diffusion inpainting? How does this differ from selecting a model and doing the same inpainting in another interface?

[bitter birch]

somehow underrated style

had to upgrade my ram up to 32GB just to run this model 😭

[modest fog]

furry

[green temple]

no furries allowed

[green temple]

the second burger

[desert island]

3rd Burger

[desert island]

I love putting in my random prompt and just seeing where it decides to take me, or what the RNG gods inspires it to make 😄 . I'm really liking sd1.5 based models and stuff now. I had always assumed sdxl was superior so I hadn't really played with these models. Pleasantly surprised.

[bitter latch]

my inet's been out since yesterday,ugh

[bitter latch]

I have discovered TIPO extention, make a prompt with regular english

[desert island]

I love this castle in the background, cat girl might be to risque though.

[bitter latch]

copies words and see's what tipo does with it

[desert island]

pretty 😁

[green temple]

big...

[desert island]

smol...

[bitter latch]

oops it crashed

[bitter latch]

Pizza?

yes please

[bitter latch]

meh TIPO takes the fun out of it

[supple raptor]

It also supports stable diffusion Inpainting, but that's not the part we care about. The Erase models are what we care about here which are very small models that are specifically made for removing objects from images. No prompt is needed, they're style agnostic, matching the input image quite well, and inference much faster than Inpainting with an SD model would be. With the LaMa model loaded, all you do is drop your image onto the canvas and draw a mask over the area with the problem object. About 2 seconds later it's done. You can then draw another mask if there's a second problem area and repeat until all problem areas have been removed.

[bitter latch]

Mortal Kombat Undead Jedi?

[keen hollow]

ah interesting....

[keen hollow]

played around with the huggingface a bit, pretty neat

[supple raptor]

yeah it's the most reliable (and fastest when run on gpu) option for removing watermarks or random small objects/shapes from images that i've found. inpainting can work, but requires a lot more luck since removing something requires high denoising if you want it to match up with the rest of the image.

[modest fog]

Smool

[gusty palm]

i like this style too

what kinda tags would help achieve it

[bitter birch]

Nah, it's a model. On civit.ai, called Kiwi Mix XL

[bitter birch]

Nah, Kemono. As an Asian version of furries

[gusty palm]

ooo

i duno if i can run xl models ive only ever tried 1.5

im gonna clean install auto111 today i think

decided i dont like forge-ux

[bitter birch]

Before XL, my laptop runs 1.5 model smoothly without any memory out. When I use the XL, it took a lot of time, and I had to upgrade my laptop for that :v

[bitter birch]

I've been using Auto1⁴

[gusty palm]

its kinda like a semi flat almost handdrawn reminds me of kakigori or cookiecutter

[bitter birch]

Here's my laptop specs tho:
RTX 3050 Laptop 4GB vram + 32GB (previously only 16GB)

[gusty palm]

https://civitai.com/models/107315?modelVersionId=154555

i have a 1060gtx and 16gb system ram

[steady grail]

i can run it and gen 1024x1024 in 6 min, but the loading of models is a big turn off

[gusty palm]

honestly been wondering if theres a way to make stuff take longer but achieve better results in more time with less energy use all at once

i dont like how my gpu goes from like 40c to 60c in like a matter of seconds

[bitter birch]

60°C is normal I guess, my laptop GPU would be running around that too

[steady grail]

well, until it doesn't touch 80 100 °, it's not a big deal

[vital raptor]

60C is easily acceptable

[gusty palm]

it goes a bit above that like i know 70 is not meltdown but i dont like it getting that high basically

the defaults on my gpu software are set to like 80~c

[bitter birch]

Perhaps if it's too loud, you could set up yourself on how the fan handle

[merry tendon]

helu

[bitter birch]

My laptop would run in low power mode, and still have the same result even when the fan speed is low

[vital raptor]

you could underclock your gpu, it would take longer, use less power and run cooler

i don't know that I would recommend it, but it's possible

[merry tendon]

very cool

[bitter birch]

Peak best use of AI is to help me drawing lmfao

[merry tendon]

currently making concept art for my game, game will use art from paid artists :)

[gusty palm]

thats a good idea

[bitter birch]

we can visualize what we want to the artist by AI, and we still pay 'em to make one, which is peak using them 😂

[merry tendon]

yepp, i also just want good, non slop art. And i am willing to pay for my game

[merry tendon]

do we know eachother?

[bitter birch]

what game d'ya develop?

[merry tendon]

name sounds familiar

[merry tendon]

mostly everything a placeholder in this scene, gonna be about bascially a girl obsessing over another girl, psychological horror

[gusty palm]

i sat next to you in juggling class

[merry tendon]

sent the wrong vid

[gusty palm]

idk lol nice to meet u tho

[bitter birch]

peak using XL model

when it was 16GB

[merry tendon]

using xl too

[bitter birch]

you have more VRAM, mine was 1/3 of your RAM 😭

[gusty palm]

i will try an xl model later then an see if it works

[bitter birch]

but I have 32 now

[merry tendon]

rate char 1-10 pls

[gusty palm]

if i have 16gb of ram currently in 2 sticks, and wnted to upgrade. would i need to get exactly the same kind

[bitter birch]

for the game, isn't it?

[merry tendon]

yep

[bitter birch]

how many RAM slots do you have?

[gusty palm]

currenyly using 2/4

[bitter birch]

yes

please use the same kind, same brand

well, same specs

[gusty palm]

ok

[bitter birch]

for the game, it fits well, I'd give it 8/10

[merry tendon]

well she is kind of psycho

lol

[bitter birch]

me personally, not into goth girls (it looks goth, but if not, ok) but for the game, it's good

[merry tendon]

its a design choice, makes sense in the story

[gusty palm]

goth girls>

[merry tendon]

man.. poor miyu, kaede(goth girl) gonna do unspeakable things to her.

[gusty palm]

wat game

[merry tendon]

game i am currently working on

[bitter birch]

vs the 1,5 model

[gusty palm]

she got the size 16 retro jordans

[green temple]

big jellyfish

[green temple]

1.5 style looks better in my opinion, ( only the style not the character )

[zinc flax]

Still cute

[green temple]

ugly

[zinc flax]

Some may not know if you're joking or not lol

[green temple]

ofcourse i`m joking lol

[bitter birch]

lmfaoooo xD

[terse ore]

(For inpainting), what SDXL model gives the most detail but also uses a style that has discrete color (solid colors for light and shadow, no gradients)

[gusty palm]

https://github.com/anapnoe/stable-diffusion-webui-ux

think ill try this even though i just tried the forge variant and didnt like of it, but i think it was just the forge portion i didnt like

i could never get tiled diffusion to work in forge the same way i was able to with auto111

so hopefully it works in this

[green temple]

[past junco]

I'm inviting my fellow anime ai generation lovers and enthusiasts to playtest a game I made in TTS which makes use of anime generated artwork. If you are interested and have TTS, creep right on into my DM's

[gusty palm]

i have small tts

wats tts

i might try

[modest fog]

They sting

[modest fog]

1.5 models still look best

Hands

[gusty palm]

i wouldnt mind if my gf had 4 or 6 fingers

[steady grail]

always nothing that can't be fixed and img2img makes wonders

expecially ipadapter

[modest fog]

I wouldnt mind If I had one at all

[gusty palm]

how to learn to use this

[steady grail]

it's full of guides online

[gusty palm]

i havent mastered controlnet but i feel like im missin out

[steady grail]

they burn the image pretty easily, so, you need to be precise when setting his weight

[modest fog]

Huh

Sleepti blocked me

Why the f

[modest fog]

I have been thinking about it

I think it's table top simulator

[gusty palm]

oooh

[modest fog]

I'm 90% sure

[gusty palm]

i do have that

[modest fog]

My first guess was text to speech and makes no sense lmao

[gusty palm]

yea same lol

[bitter birch]

but is it me, or 1,5 model prompting is kinda hard

[steady grail]

i feel 1.5 more comfortable than xl, but 1.5 is notorious for being more obtuse

[bitter birch]

like, I put look_away, but the character still looks at the viewer anyway

[steady grail]

it's just question of practice

[steady grail]

well, that tepend from the training of the model

[past junco]

Tabletop Simulator

[bitter birch]

I see

[steady grail]

if that action isn't prompted under a different token, that, it simply lack that training

[bitter birch]

I even tried to put v (peace fingers) and it's just not showing xDD

but yea, my model is somehow mix, I merge many models to create my own style

[modest fog]

It's not harder it's just that the models are a bit dumber

[steady grail]

usually, the dullness depend from the clip

[bitter birch]

perhaps if I merge too much models, it get's dumber?

[steady grail]

a wrong unet is much more prone to mutation

[bitter birch]

but somehow behind it's dumbness, I can find an unexpected gem xDD

[modest fog]

I dunno

Me dumb

[bitter birch]

what am I man <w>

[steady grail]

give me the prompt

let's try it on my model

[bitter birch]

anything, but with the "v"

the v fingers

peace

or something like that

[modest fog]

Try peace sign

[zinc flax]

I think you prompt peace sign instead of v on 1.5

[modest fog]

Instead of v or peace fingers

[modest fog]

Stupid

[bitter birch]

it gaves me mid fing instead

[modest fog]

It's trying

[steady grail]

if it react to the token, it may be a mutation

[bitter birch]

💀

[steady grail]

try with a portrait

[bitter birch]

done

but haiyaa

also I just realized, my VAE is inactive

[modest fog]

Still furry

[green temple]

+1

[bitter birch]

at least i'm not one of those most furry degenerates

[modest fog]

@green temple

[green temple]

[modest fog]

I am sad cat

Not even smol can fix me

[supple raptor]

That's fine, your model has one baked in that's being used.

[bitter birch]

hmm, yea..perhaps it's me long time no use the AI lmao

I use it again when I got an XL model

but it's hella slow

I even upgrade my RAM up to 32gb, but the RTX 3050 Laptop is not lying xDDD

[supple raptor]

Also, furry models most likely use e621 tags, which means you should try v sign to get the peace gesture

[bitter birch]

hmm, ok. Thanks for the info

[modest fog]

Which UI are you using

[bitter birch]

Auto1^4

[modest fog]

[bitter birch]

[modest fog]

Slow11111

[bitter birch]

is it?

[bitter birch]

I'm too lazy to change another UI

[steady grail]

that v sign so damn hard to gen

[bitter birch]

not hard, but wasting your time xDDD

[modest fog]

I think with your gpu you should use forge

Looks the same as a1111111

[steady grail]

on 1.5 is pretty hard, hands are simply mangled

[modest fog]

Just better for smol gpu

[steady grail]

they always need a fix

[bitter birch]

I have some plugin installed like Tiled VAE, ADetailer, ControlNet installed

is Forge supports them?

[modest fog]

Yes

It comes by default on forge I think

[bitter birch]

this one is actually good

[bitter birch]

ooouh

but eh... I'm just too lazy to setting up again xDD

[modest fog]

The controlnet guy is the one who made forge

He smart

[bitter birch]

I have those plugins, but never use it, since I have a hard time using it, and no good results

perhaps it's because of the model doesn't support it?

[modest fog]

If you take like 2 minutes to gen something

In forge it might be 1

It's worth installing it

[bitter birch]

well, for 1,5 model, my laptop can do around 30 seconds

for SDXL, up to 3 minutes, and maybe 5 xDDD

[modest fog]

You might be running out of vram for sdxl

It gets really slow then

[zinc flax]

Might as well use a cloud service

Your laptop won't last long

[bitter birch]

it is, what can we expect from 4GB of VRAM ;w;

but it moves to system RAM, that's why I'm upgrading to 32GB

[bitter birch]

I rarely gaming nowadays, so I'll use it for AI

and it only runs on 65c, so it's fine

[modest fog]

You might be able to do sdxl on 4gb with forge

[green temple]

mid bloo

[modest fog]

It's more memory efficient

[bitter birch]

maybe I'll try it later. Thanks for the info

[modest fog]

Big

[zinc flax]

AI is way more resource heavy than gaming though lol

[bitter birch]

I add you in friendlist, in case I'll need oyur help

[bitter birch]

it is xDD

[bitter birch]

yayaka

[modest fog]

@green temple

@green temple 🔵

[green temple]

200% bloo 🥳

[modest fog]

@green temple smol park

[bitter birch]

what's xx?

[modest fog]

Give context

[bitter birch]

hand action

[green temple]

crossed hands ?

[modest fog]

Two fingers crossed?

[bitter birch]

I'ma try

[modest fog]

I have no idea

Could be something nsfw

[bitter birch]

😭

AI nsfw is scary

ngl

[modest fog]

Booba

[green temple]

oooh is it the ... 🌚