#🍥|anime
1 messages · Page 242 of 1
Sometimes I get confused cuz it feels so much like sd15, but after checking the node settings it was sd3 after all...
This is her if all you have is tomato juice.
@lyric zealot 

cute eyes bleed over into the icecream
2 eyebrows 
2 sets of them
ugly 

thats why people should avoid anime
I think those are her eyelashes
listens to fireworks from around him.. damn they are noisy

Did they? That's actually huge if true
Astra still hasn't begun training v6.9 so if the license has changed then v7 on SD3 could happen, it all depends on if SD3 is easier to train than SDXL I guess

AND they said they'll update the base model with more refinement
👍
That's actually huge omg
I just hope they don't censor it this time
Yeah makes sense I guess

Fishnets came out blurry for some reason but they're actually quite accurate
I wish I could draw that lmao 😭
The only thing I fixed were just the eyelashes a lil bit and I added a lil mole 

||mole fetish||
It is what it is, it looks cute 
This cat needs to be arrested
hes learning
Learning fraud 😂😂
You can see it in his smile
sneak sneaky cat
lick

How to reduce color bleed?
regional prompt!
BREAKs (maybe they help, never got much out of them anyway)
Well I mean it's just 1 character in a white background. But it adds bits of blue on her T-shirt (from the blue jeans)
regional prompt!
you can do regions for hair and stuff
so other parts won't have the "blue" or whatever in the prompt
And this works with comfyUI?
I have no idea how to comfy, I think it does but with complexity just like anything in comfy
cause I know some people use regional prompt thing in comfy
I see I'll check it out. I would have expected it to work without needing it though because it's just one character
like @nova remnant said, BREAK might also be worth it to try for this case
Yes ty 😊
stoopid

working 😔
finally found some time to get back into ai
took me a bit to get my setup all well, setup lmao
SEND ME ALL OF THIS idk if this is the only one you made but SEND PLS dms

Lord have mertcy

too much works not enough time for us 
you need a dose of bonk
Leme drink my cola.
With zero sugar
😔
finally trying some sdxl stuff
BREAK is not a built in function for comfy. To achieve the same effect you would either combine 2 different prompt nodes, or use something like SMZ nodes which support A1111 prompt syntax
loooool never
im just going to make my laptop power through it
its a tank
your laptop will evolve from so much beating 
its my hope that if i abuse it enough itll learn to ascend from laptop to GOOD laptop 
make any good goth girls?? 👁️ @nova remnant
no, been a while
what model are you using 
#🍥|anime message```yaml
best quality, masterpiece, score_9, score_8_up, rating_safe, bold line,
1girl, solo, black hair, blunt bangs, sidelocks, twin braids, red eyes, gothic, collared shirt, makeup, half-closed eyes, closed mouth, smile,
(bee (deadflow):0.8), ilya kuvshinov (style), wlop, realistic,
Negative prompt: bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, (worst quality, low quality:1.1), jpeg artifacts, signature, watermark, username, white background, simple background, (gpo, aca, aer, api, fla, gcx, hll, hnj, gpc, fii, fey, fbv, evg, iew, ifl, igh, iwj, iwp, ixb, ixe, ixz, jaf, jbm, jfb, jsf, jyk, kmz, ksh, kxg, kzg, lbv, xsl:0.5)
Steps: 24, Sampler: Euler a, CFG scale: 7, Seed: 4102779530, Size: 768x1024, Model hash: 89e7c7518c, Model: tPonynai3_v55, VAE hash: 66f78ad136, VAE: sdxl_vae-fp16fix-blessed.safetensors, Denoising strength: 0.4, Clip skip: 2, Hires upscale: 1.5, Hires steps: 10, Hires upscaler: 4x_fatal_Anime_500000_G, Version: f0.0.17v1.8.0rc-latest-269-gef35383b```
there you go ^ 
thank you :D
hey whats that score_9, score_8_up thing
it's prompting convention for pony based models
wdym 
it basically has the same affect as doing something like best quality, hires, masterpiece or whatever for 1.5 models
oh bet
sdxl models (including pony) don't need those tags generally, but pony specifically was trained with the score tags accidentally, so they're required to get the best quality from the model
oof
can you give me a quick rundown on how to use them properly
i want to make sure i do it right
i might have a lora problem 
the pony page on civitai has more explanation, but basically you want to add score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, to your prompt
also there's source_pony, source_furry, source_anime, source_cartoon for specific types of content and `rating_safe, rating_questionable, rating_explicit'
I would say start with all of them, and then test it for yourself
depending on what kind of style you like, you might not need them all
the model also knows a lot of artist tags, so you can get vastly different styles from the same model just by changing the artist name in the prompt, and using some or all of the score tags
dunno if that artist is known or not, but here's an example of the same prompt/seed but with 2 different artist tags (eba on left, nemu on right)
those are ones i knew off the top of my head that were obviously different
another artist, everything else the same
oh sorry didnt see your question but am glad Captain already answered it
this is with all 3 artists 😄
well, for the score_tag, most of the time people kept all of them, but you can play around with it, even putting some of them in negative might change the style
look at those metadata of others of style you like, to see how they prompt
how do you prompt artists??
i just put the name in the prompt
eba, solo, (child:1.2), petite, thin, tan, (fox:0.6) ears, pink eyes, long twintails, white hair, fluffy tail, hoodie, shorts, outside, night, happy, score_9, score_8_up, score_7_up, source_anime, this is the prompt for the first one i posted, 'eba' is the artist name

some of the tags i use, like the weighted 'child' tag, solo, etc are because of the influence specific artist tags can have on thing
like, some artists will make a chibi cat girl by default, so if you like the style but want something else, you have to prompt more heavily to fight it
or they'll make explicit content by default, so you have to tell it what clothes to use 😄
nooooo i hate prompting clothes 😭
just depends on the artist
im no fashionista, i never know what type of clothes to prompt
i usually keep it simple, this is just hoodie, shorts for clothes
most of the artist tags i use frequently actually do have clothes by default, but i have seen some that don't, so i figured it was worth mentioning
ill probably find some wildcards for clothes and let it do all the clothes for me loool
yeah that works too
i should make wildcards for all my loras
rn im just trrying to get all of my loras to show up 💀
do you use a1111 
Experimenting with a new Retro 80's style
It's a bit rough atm
colors are reminiscent of the 80's, but doesn't fit 80's anime style at all if that's what you were going for.
Huge
Particularly the colors, the actual art style isn't pure 80's though and I'm aware of that, but I quite like how it looks. I'm gonna try and see if I can get something more pure but I'm pretty much good with this
it's a nice look. re-reading my message it seemed negative and that's not what i intended.
Ah, suppose I did take it that way... sorry lol and I understand
Booba
no reason to be sorry. i caused the misunderstanding. keep up the good work. i'm always interested to see your new posts here. they tend to be pretty cool
Oh what? That makes me happy lol
best style 
Really liking this new checkpoint merge
Very very beautiful anime picture.
thx !
lucky gen
Sd15?
sdxl but sd 1.5 can do it too
(sd3 medium)
lovely, this is NAI?
Yup 
great stuff
I always love this kind of stuff. Very satisfying details.
glad you like it !
We are evolving more and more
Nice outfit bro its very unique today

I like the concept of it

thanks friend
No keep it up

Anyone use sd-cn-animation? I need help every time I try to create text to video I get this error: An exception occurred while trying to process the frame: Device index must not be negative I don't know how to make it work it generates the image but won't continue after that.
I n automatic1111 btw
try asking in #🤝|tech-support
that extension has been abandoned and only works in older versions of a1111
not exactly what I wanted but , interesting
chills out to "Pink Floyd - Sorrow (PULSE Restored & Re-Edited)"
guys can i use "tagger " extension on forge ?
Thank you 
I really do suck at choosing outfits for these characters tho 😭
shes lost in the desert will you help her?

wats this model uwu
i wish i could get better sword poses
if you haven't tried it, try "holding sword, sword stance," that helps alot with most of my gens I've noticed

|| 👍 ||
migu
i recognize this pose 🫠
Yeah, it's called "Top down bottom up" i recommend you trying and posting here
😂

HUGEEEEEEE
smol at beach 
lick

I don't know what hte hell she's got but I aint wanting to find out!
tries not to melt
it's 98/real feel 110F
36.6c/ 43.3c
ok thanks, i tried defensive stance on this but ill try that next
hmmm
Great oogly moogly
Its not bad but there are a lot of outfits to think of bro good luck keep it up
could always use a wild card
Smol sus FREAKY ARC??

melting.....
Back at it again with another banger knight anime character that i forgor
Now make her swim
Im your mommy" ahh pose
The "save" button is always on you
ok tries
True, but I also like generating a bunch of different characters with the same outfit like I said 😭
I really like the makima one
she def gonna be a mommy after this 
Makima ones are also my favorite 
Jeanne d'arc alter (avenger)
Oh ty lol i always see it on Pinterest
Had to go look cause I keep forgetting her name 😭
There's way too many fate characters it gets confusing
I watch anime you watch anime
haha it got confused

Man that's basically underwater
I haven't even seen fate I just like the characters 😭
Same😭
Now make gojo satoru a fem.
Jk 
Im just kiddinga
Tomorrow, I've been prompting all day 
People would jump me if they saw it
Why is that?
I'm tired of prompting 😭, need to take a break
Been messing with different prompts all day
Oh i read that wrong
I thought you were gonna prompt all day tmrow
Mb 
Take a break aslong as you need to lol
Who knows, I might do that 
It's fun messing with vibe transfer and inpaint
I thought break
Yk you gotta chill with the breast you sometimes wanna make them even 
Yeah Ik, but still I just know I'll end up prompting for a while
What does that even mean 😭
Too damn BIG
As they should be 
Shiii that's G cup large
I agree but you gotta lessen the size when you're gonna work with an anime character that looks young
Make them even
Balanced
Yk
tries not to melt
I think I prob shouldn't be doing AI art , pumping out more heat. oi
indeed you should do smol characters only 
I wouldn't say makima or arlecchino are young, but even so I don't really see a problem with it
Arlecchino is like 27 or something
Yeah exactly
I do like big melons
Yk just keep doing big melons
Smol has his own style 
I just don't really like making smol characters, just not my thing
ik just kidding

kinda looks like a crime scene, almost
This
I know don't worry 
well, i do enjoy making stuff like that 
Yeah my favorite position
Who doesn't 
Shush ik people pass by my room and see me doing the position
IT IS comfortable okay
do this kind of smol 
Nah uh
4lex would love Nicole
Maybe 


Oh yeah I've seen this character before
She's from that new gacha game
She got chest debuffed after the release
Same company and same community
I play it lol
It's fun
the "3d" ones

crime scene
where the 4D ones
for the 3d one? 
i don't even know how to count cos of how bad the generation was
Is that 3d style trained on overwatch art? I swear that's the overwatch symbol on her shirt 😭
oh lmao, yeah you are right

i guess a lot of overwatch art in dataset tagged with 3d

It was probably trained on all the SFM overwatch images
crime scene "the catch of the day"
thats actually the peace symbol probably leaked from the peace sign in prompt
ahh that's actually more likely
Yeah you're right actually, it just looked really similar to the overwatch logo lmao
Erm akshually 🤓
stops doing art so the room can cool down
but if u want that style just search for yeero on civitai 
Yeah there's some good ones on there, I think Pony can also do really good 3D stuff without loras, never tried it out tho
melts
@glacial bane daily
for you
I will smack the bats out of you
I'll take all 3 
Bro
who gets it? 
Everyone 
ice cream ?
i didn`t 
I definitely cannot link what im talking about here
bro i mean the image not the game 😂
i don't even think i should screenshot part of what im talking about, since it's so well-known
Whaaaaa
is that dva 
Its not wuwa
?
Work on it
im too lazy 

smol is working another way on it

Stop being freaky
is she a voice actor ?
from overwatch
idol that works as a mech pilot i see her a lot on video sites 

damn sd3, what are you up to?
cant do that with sd3


edge E
she will smak u upside da head fool
Very very cute anime.💖💖💖
wow, region of kritra is pretty good
thx
sd3?
no 1.5
sd3
the 2° is a little too skimpy
the result is better than what i heard, it's an img2img?
it's pure prompting, just my eyes looking at the image and converting to prompt by brain
are you a bot? XD
lol
no, just had to many hours trial and error
and Im must admit sd3 does help, its snappy with prompts
in undersanding what's the most glaring difference between 3 and 1.5?
i'm too lazy by trying it myself lol
better at higher resolutions, and more snappy to prompt. less good at nsfw, less variation in poses. easier in different styles (without loras). seems to need lees steps for good result.
they say there will be an update in 2 weeks
and the license is now more relax
so, im expecting people to wait a little more and then start training finetunes and loras and all
i'm starting to think that backlash was fabricated
that would be a good marketing trick
i repeat, the results looks pretty much how it was 1.5 at first
yes
I had a deja vu sometimes. generating stuff, thinking i was on sd15. checked the settings and it was sd3
eheheh
for comparison, a result from my custom merged sdxl/pony model (only +/-40% pony)
i prefer the sd3 lol
i can see toward where this model is biased -.-'
i really like that one
Streaming available SEPT 5!: https://linkco.re/1UbnMv9z
Smartphone "Sony Xperia 5 V" Preinstalled Track
スマートフォン「Sony Xperia 5 V」プリインストール楽曲
Mili on Patreon: https://www.patreon.com/project_mili
Mili Merch & Goods: https://projectmili.store/
Follow Mili
Spotify: https://open.spotify.com/artist/0K05TDnN7xPwIHDOwD2YYs
Apple Music: https://itu...
song of the day
(2girls, DUO younger and older)
tho it's the 2girls and DUO that matter
oh, never did it
only learned about it recently myself
it's cute to experiment with it
i wounder what would happen by mixing it with region
shrugs
hmm I wonder if more number would work with like triple , quadruple, or many...
tests
Anyone need an umbrella?
haha: (anthromorphic dog girls, many, younger and older)
phoenix wright, doggies for all
These are actually multiple models, but all of them are basically Frankenstein's Monster: An amalgam of ripped models and merged with other ripped up models
the style is very cute
Thanks lol, I've actually stopped my experiments for quite a while but... decided to start it up again yesterday
No joke though, the current merged checkpoint is a bit insane
my mixes are always horrible and always only generate blobs lol
I merged my NOMD model with a Kawaicute mix 😄
and to keep from getting asian I had to put (asian:1.5) in neg 😄
oh and upon checking my ac.. no it wasnt' struggling, it was set at 88.. DOH!
but usually 1.5 is very cancerous XD
that doggo is a true chad
big head
ghibli is pretty hard to balance
the bg are fantastic but the details are a bit rough
hmmmm... 3girls, triple
attempting to fix it, gave me horrible mutations
I haven't had any issues with Ghibli funny enough
as i said, i'm terrible in merging lol
that way you only do 2 instead of 3 models
a and B witht he B being the important for config
faints
Put a bigger emphasis on Ghibli here
haha, ironic , song playing "Balls to the walls!"
till now the only somewhat good result was mixing aniverse to focus more on the anime style and less on 3d
guntz
so 3 girls, triple. works
one is showing too much her back assets and the 2° is literally half naked XD
good point
tries 4girls, quadruple
cute
tries other things.. i've run that into the ground
I really like the colors and the clean lineart
reminds me so much denno coil
Added to my watchlist
What do you know, it's already in my library... Wonder why I never seen it before
it's very very old
there are so many lost treasures between 2000 and 2010
why they are so creepy?
Miku and Yuki
shrugs
they must be little demons in the skin of cute angels >.<
are this horns between your hairs? squints his eyes
damn, i should stop drinking till late -.-'
oh boy lovecraftian gives some very creepy results
umm....
waits "when you see it"
ok not so creepy .. almost
tries to stay awake to get sleep schedule back on track
The heck?
Wires?
shrugs
oooh
ummm .. wtf
hmm ok umm.. creepy much? .. yes
is fading
takes out lovecraftian and puts in art by E.H. Shepard
ah ha
one more and bed
tries different model
I like how this one came out
You can really tell a good story with a good AI gen pic
Why he looking at him like that
"Did I forget the pizza on the oven?"
LMAO
"You know, you walk out the door. You see someone that you know, and they ask you how you are, and you just have to say that you're fine, when you're not really fine, but you just can't get into it because they would never understand."
I need to know whatever the heck that guy is
beard man

hes just a beard in human form
Giving me Legend of Zelda vibes
One of the funniest things I generated
A cute boy behind him will be the 3D sereenshot Image of দূর্গা পূজা, the name will be “দূগা পূজা” he will be sitting on a 3D “Porosh” name plate logo and in form of him will be a 3D “Porosh” logo, Happy New year style , 3D” দূর্গা ঠাকুর” with many lighting, green, Yellow, red , orange, sky blue, 3d gift boxes with the name”Porosh Dab nat
ugh, it's too hot here today
I think I got the soft color 2.5D flat anime look down 😄
realistic anime(kinda)
yawns and ponders going back to bed
goes back to bed.. .thud
How would you guys feel about me writing up a guide on my whole process?
FINALLY YOU MADE A BOXING GEN AGAIN
Do yk how long i waited for you to make one again
4 months it think
Sorry for the wait. 
I do not think her body can support those breasts

but ya know, santa is magick!

Oh no
such a terrible thing
I'll support them for her
👍
It's Santa Tifa!
😱
I canceled this one half way but just had to save it anyway because of how cute it is
The big dilemma, study or have fun
Damn it Apple!
cute
Very
spell casting practice
LMAO

Spell casting practice completed. That was some serious powerful time shifting spell...
new lora I trained today, still trying to see if it's garbage entirely or what
Model for landscape recommendations?prefer simpler 2d style
for something like this i used kakigori

her bag!!! 

the prompting is too 
tbh, i don't even know why i paid for it

i watched a video a few weeks ago about how jpg is perceptually indistinguishable from lossless png, but like 10x smaller has saved me so much hdd space
i set my sd-ui to save as 90% quality jpg instead of png and images that would be like 3-4mb are like 300kb
it is perceivable but agreed png not worth the space
I use webp on lossy 100% compression
because I found them better than jpegs
i did a few tests and i could see differences at like below 85% quality
But png can be made transparent background and not jpg right? But I guess that's specific cases not general
I don't think SD can generate in transparent background
correct in general though
webp can do transparency
I think there are special add ons for transparency.
But I guess you could just convert the files you need if you wanted the transparency when needed
Is there any point in using 1.5 models nowadays?
i mean, some of them are still good, and also lower vram usage?
does resize from seed aspect ratio have to match the og image
I am looking at mistoon_anime and I really like the aesthetic
i have never ever used that before, so no idea what that does
definitely looks really clean
download and try it 
I have been using the 1.0 for over a year, it's great. 2.0 was too anime, I just got 3.0 now which is supposedly a better version of 1.0 while still being between anime and toon

interesting, haven't been playing with SD1.5 much, but i downloaded that, gonna try it one day

im not super sure myself but i have been assuming it uses the seed from a larger resolution and shrinks it down to fit the input dimensions
like as if u were genning a 2048* whatever image but resized to tiny so u can get more details im only really guessing though
I'm still using 1.5... I think? Honestly not sure what Pony is, but I think I have some of it in my current checkpoint merge
Wow old school kimba
Ayyyy
can xl be run on 6gb of vram
I'm running xl models on 8gb vram
using old forge
so you might be able to do it as well
Tried 1.5 again, forgot how much faster it is. I like seeing the pictures appear much quicker
3060, has 12gb vram
That doesn't even make sense... My machine can't handle XL, though I'm willing to bet that some pony checkpoints aren't that heavy
Holy crap how
And what's Forge? Is that another UI?
I've only ever known automatic
I do want to look at Comfy, buuuuuut
I already have a decent set up
ive been liking forge its like a1 but different
you can't merge sdxl models into 1.5
what
I kinda miss some of my sd1.5 gens, but I don't miss the horribly deformed hands
Then I have no idea what the heck I'm doing. I could swear that I'm using 1.5 and Pony merge
maybe if it's as Sus says, a 1.5 version of Pony
Just had to double back and check what I merged (lucky I remember this time)
SDVN Anime (the Ghibli checkpoint I merged with) is in fact SD1.5. I did recall having some errors when trying out a few other checkpoint merges which I did realize failed because of low VRAM
Need money, need VRAM
if do you like an easy but very flexible set up, use krita diffusion
I love Krita, what is Krita Diffusion lol
can't push weird stuff like with comfy, but i really like the regional of krita
Honestly, my current setup is great.... except for hands
a script of krita, look on google
So they made a script huh? Awesome
yup
works very well, but i miss al the funtion of krita
but krita can do other stuff, like regional very easily, gen high res images very wast without weird mutations
inface and everything else you expect from a normal set up

but still, for me is very hard to leave behind my setup of comfy
i can't do regional, but prompting is much easier
Well, like I said, I like my current setup
I plan on getting a better computer at some point, from which I can expand and spread out with
my boss wants me to buy photoshop but i feel like gimp and inkscape can do everything it can do
photoshop is not that app that changed policy asking for a part of copyright on everything you do with their app?
If you need PS but it's too costly, feel free to DM me 😉
How do you get landscape to be shown directly side - on and at ground level? Like a side scroller game. I put in "profile view, side view, ground level" but it's not what im looking for
bottom prospective?
maybe cross section or orthographic
maybe you can find a sidescrolling game lora
or something that can generate texture tiles
I'll try ty
I couldn't find any on civitai surprisingly
what you might be able to do is draw like, a blue sky bg rectangle, and a brown earth rectangle, then i2i it with some 8bit texture lora?
1.5 hands are okay though
Well, model dependent I suppose lol
it's a lot more luck
it's free if you know where to look
i might do that cuz i dont wanna spend a hundred bucks on a subscription
If you don't hack me you can borrow mine
i have old versions somewhere i think
I dont use it much anymore
But in still paying
Also I have a crack link for it somewhere
i remember seeing psd capability for either blender or gimp
I hate Adobe but it's the best software
or whatever the photoshop format is
Psd yeah
Forge is basically a fork of auto1111. It was made with the intention of optimising vram usage
Yup, there’s a 1.5 version of pony
Wait deadass?
Kinda, I didn’t really update, so it’s still good for me
Oh nice
I never update my stuff just for shit like that
Who knows when the next update will be broken and who knows if anyone will bother to fix it
yup, but it's not that good from what i heard
the funny thing is that someone already make a fork of forge, and implemented most new updates from auto1111 to forge
same thing gonna happen when everyone jumps over to SD3 (ps: if sd3 ever gets better)
I love people who find broken/obsolete software and just "fork" it
Sd3 looks meh to me tbh
the nice part is people are collecting datasets




