#🍥|anime
1 messages · Page 225 of 1
this is all only for you!

No
Not bikini armor so can't be good armor
Anyone currently have novel ai?
I wanna know if it can do complex concepts or not say
say somenody holding a knife like just stabbed somebody
autism can do it, so i'd assume nai can
yeah currently the finetunes of sdxl are usually
either u get the styles, but 0 ability to create something that's not in the dataset, or you don't get the styles
i mean, you always have 0 ability to create something not in the dataset
but, that's what loras are for, they extend the dataset
captain, got a dataset, trying to create a lora, hmm always stuck oh so slow(for 1step it gets like 1.5-2mintues), can you try it with your own parameters?
now I cant figure out what settings should I input
Technically if that was the case you will be able to create nothing
Guess what i mean is like the ability to interpret a bit of new things
a model can't make new things, it can only piece together what it knows
if a model doesn't know what a knife is, or what it looks like when someone is holding it, that model will never produce an image with that
training a model is basically feeding it information so it knows what things are, but they can only 'know' a certain amount
so, you train a lora for a specific set of knowledge and apply it to a model that doesn't know (or knows very little)
it's also really bad at context, because it can't think for itself, so it might know what a knife looks like but it doesn't know it's called a knife so you can't tell it to show one
or, it knows a different name
ye
if you still can't find a model or model + lora combo that does what you want out of the hundreds available, then you have to train one
i either never wrote down my training settings or i lost the file i put it in, i'd have to figure it out from scratch
which, tbh, needs to be done for basically every lora/model anyway, there's no magic numbers that always work
of the settings i saw you post before you deleted it, i think your repeats are too high, but i'd have to look it up again
nah I just need to check its my vram or kohya ss problem
I trained a lora yesterday
it was terrible lora but trained so now its like always stuck
yesh those settings were my last training session
What lora u making? Character?
yesh ofc
I thought it was a vram problem so I used batch size 1, it didnt work either, so I have no clue now what to do
what gpu?
prev lora trained with batch size 3 
4070
sdxl model?
well, last i was training anything you couldn't do sdxl on a 16gb card, so i dunno
it was using like 23gb of my vram
tf, but I did it prev, deym I have no idea wots going on
first time i come into this chat again and people are actually typing lol
did you change the dataset or anything?
it more common than anime images tbh
nope didnt touch the dataset
That is not the plan, I cluster images by style attributes (which is less precise than specific artist but more precise than 'sketch with color highlights' or '80s anime') and then tag images belongin to same cluster as 'style_42'.
@wooden coral @dull jackal #🍥|anime message
Are you sure you...didn't want that gift card!? XD
Oh no I missed 50 Dollars again 
ok so I've been trying to do two mules side by side for someone and well I gotta share this. I think i've seen pigmy mules in an anime.. not sure but here 😄

🤣
6 wheeled skate board with cute stuff
👍
How can you rescale several images at the same time in the stable? it makes me very slow
@vital raptor

I'm always lurking
Just ate food so im lazy looking at new discord messages 
😭
😭 ❤️
I meant you can get good quality images but you need to put more emphasis on the texture/ style. Otherwise it's feel like the quality is bad
only 15 or 20
😭
i wish i can 😭
nice
that's me 😄
Many smug :3 pirates
also witches
another test, cute
tomorrow i try with pony
after all that was the model the workflow was supposed to use
the prompt is pretty simple and touching the latent is a big headache
for now i fixed some settings on the steps and depth side
when i test pony and works as i expect, i work on the latent to improve it a little, if i can
chyler leigh as a Keeshond girl wearing letter jacket
hands noooo 😢
that's a little weird
makes me unconfortable the pose of the legs
soma, i would delete that, for the staff, may be a little risque
everything is lol
it's a little too close in being an up-skirt
what will naked bears do together in jungle ?🤔 🫣
normal bears dont' wear clothes , silly
and expecially on a looking underage character, isn't the best choice
yeah thats what i ask 😄 what will 2 normal naked bears do together in jungle ?🤔 🫣
oh im sorry that was unintentional didn`t mean it
@bitter latch 
@steady grail is this ok ?
bearly anything
yep super charming
i don't like to make people uncomfortable, but let's try to self moderate, without the intervention of the staff. out there is full of people ready to point the finger and make trouble
lol
after watching saw, honestly i'm desensitized from anything XD
what is your most favourite horror ?
ok this is sorta anime and it is made by AI so here... https://suno.com/song/8ff34842-b769-405d-b64a-ea7377b31715
twangy country song. Listen and make your own with Suno.
hard question. horror is a genre hard to grasp
ya know to totally drail things
can AI make me to an anime ?
more than likely
really ?
just in digital form only
https://www.youtube.com/shorts/UpttYjF9kcw make me to anime please me not the girl who is next to me im the one on the left just me plz
#viral #nghile #nghile30_41_5 #vanphonggenz #dilamcogivui #gogovnz
when its gonna be done ?
oh u were talkin about sleepin
why go to bed so early
early for you might be late for them
and some of us are old fats who should goto bed early
i have a turn is some hours XD
but I don't. been sleeping in 2 or 3 our periods latey, , damn spring
👋

hmm odd. hit generate and 0.9s ecs here's ya image, nothing.. wtf
and thats her dress..so dont' even !
where can i go and generate things ?
me 
what a doll.. literally hehe
@lyric zealot @green temple smol smol
do you have a decent gpu
indeed
wth you want to become mikasa 🤣🤣🤣
haha strawberries on head
yeah i meanhttps://www.youtube.com/shorts/ccE3geiSV6Q me on the left only not the girl who is next to me im the one who is wearing a tie 😄 i'm better fit with Mikasa way more than she is 😄
#viral #xuhuong #trucxinh #vanphong #dilamcogivui #nghile30_41_5 #genz #gogovnz
"ice cream seller "
nah i just have like my phone and my longtime laptop
I scream you scream, the toaster screams, I shoot the toaster
sure !
not a very good gen but kinda funny cx
I like it
so basically just generate images not videos 😦
Sora making videos but like i still can't download it
but not hmm anime tho
@brittle rampart its you!
i cant find the generator
what are you trying to find?
generate myself dancing become an anime character
probably quite difficult if you don't know what you are doing
if you already have the dance video, then you will probably wanna do ssomething like img2img
remaker ?
?
Img2Img employs an advanced diffusion-denoising mechanism to seamlessly translate images according to text inputs. Now, harness the power of Stable Diffusion img2img to unleash the boundless possibilities of your imagination.
no ?
well, you can use different methods, not sure about that link
since my anti virus auto flag up site with script
most people here use their own local pc to do stuff like img2img
but there's definitely cloud based one, just have to find a trusted one
and img2img is one image
for your case, you will wanna do the whole video
so it's probably gonna be some script built upon img2img
but not the girl who next to me tho only me
which i can't remember the name of
á đù vn
sao biết vậy hiyhi
"drinking water" lol
nyoooo... she a cat....
thx for the lighting lora tho !
now your gen on different height level! congratulation!
they are all so cute!
try
not perfect but shrugs
haha love it
so tired
ya know my brains on sleep when I do these
more accurate
tho that arm. hmm
this one went weird but interesting
like they are dreaming
we need call exorcist!
i mean i letrally typed it in prompt lol
this is a good! this a very very good!
😍

Sooma going insane when seeing smols
Yes!!! Yes, you are on top today!
i think she will become prettier without glasses 👍 🤣
indeed 🥹
glasses are a divine item that increases the degree of Kawaii level!
ohhhh ok 🤣
what are you doing to me today?!
You have glasses in your picture
You have that level too
you a bad bad boy...😂
sky casle?
That's a good idea
Waiting for you @brittle rampart x3
It's so far lmao
The castle
We're close!
oh yeeeeiiiii!!! 
"serving cake, party, "
If you want to support the channel please visit our Patreon page : https://www.patreon.com/mysticplugrecords
You can find more available membership plans/promotion options on https://slvg420.beatstars.com/
We also offer a vast variety of instrumentals, sound engineering (mixing & mastering), album cover Pink Floyd -- Another Brick In ...
thx 
can ask about model ?
this is SDXL or ....?
guys can we make normal videos into anime with stable diffusion or it works in stable warpfusion?
it's merged model based awpainting_v13 and SD 1.5
Bunny is a good teacher 
I have no idea how you would even do that, but I'm sure it's possible
guys how many minutes it take to convert 1080p@60fps video 1 minute video with rtx 4080?
I'm here in this chat today as the research officer, I'll need this in full to make sure everything is okay
i m thinking of buy gpu for stable diffusion i have 4 cores laptop with igpu lol
so thinking of buying 4080 or 7900xtx c
cause 7900xtx have more VRAM
and 4080 have more compute units
confused between which to buy
AMD doesn't optimize stuff well for AI on consumer cards
I wouldn't buy one if you plan on using for AI
which gpu u have?
The worst one

i need gpu for video conversion into anime
cant be worsed than mine i have intel igpu quicksync
wiht 4 core laptop lol
Nai v3 hits different
This is NAIv3, so you don't really need a good GPU to generate images with it
Already spent like 5k anlas and it's only been a week since I've bought it 😭
sorry i m nub can u explain me what it is is like a diiferent sofwatre than stable diffusion or like a server?
And then use the 10k anlas the next month lol
Not sure how to explain it in the best way but I think it's like midjourney where you have to pay in order to generate images with it
ooh cool
which cpu u have
I probably will honestly
i mean cpu gpu
can midjourney or is there any AI which can convert ur videos into anime
I got a ryzen 5 3600 and a rtx3050
You should probably be asking that question in a different channel, for example the tech support one
okays thx
does midjourney run smooth on it?
have u tried comparing mid journey with stable diffusion?
Never used midjourney, so I'm not sure
Midjourney can't be installed locally
Midjourney, just like NovelAI, is a subscription service, so you'll have to pay for it in order to use it, and you can't install it locally
And you also don't need to have a good GPU for these, because the images are done on their own GPUs
@ivory pebble does NAI have the character ouyang xueli, I was wondering because she doesnt appear in danbooru or pixiv
Nope, I've tried generating her. If she doesn't have any images on danbooru then NAI can't generate her.
deym
However you could probably generate her using the vibe transfer feature on NAI. Not sure if it would be good but it's still an option
its like reference of smthing?
Don't really know how to use it but once I figure it out I'll see if it's possible
kay...
I think it's kinda like their version of IPAdapter, so yeah pretty much
wooow lightning
Ugly 
Lmao my prompt is stupid
🤣🤣 np
"Cute modern anime style, cute and unhappy small girl, she's wearing a fluffy pink dress, close up portrait, she's angry that her father didn't buy the pair of shoes she wanted, well lit living room, window, lens flare, her face is a bit round, she has a bow on her hair and her arms are crossed"
Kek
Can it do fluffy pajamas
Extremely smolll
2 ears noooooo i think "fake ears " neg will solve the problem
Das a fluffy worm
It's a dragon
@zinc flax extremely angy fluffy pajamas wearer
Snek
Cuuute
Super cute
yaaaaay
Do striped fluffy pajamas
agree
cuuuute 😭
A bit of trouble to get it right
do we have a danbooru tag for that dark area of face?
Smol
is this cutemix model ?
Shadowed face
I think
sweetfruit_coco
👍
Don't ask where I got it since those are old images
Shaded face

Went to check
yesh it works
i never used stable so u guys are producing these images with text to image?
also how long it takes to produce one image depending on graphic card
KAWAIII ELF CUTE
OMG i want her so bad she is so maou maou cute
i want to pet her and pull her cheecks so bnad
bad
omg heartattack
stap it ! stap it awwwwwww
vil which pc u have?
gpu?
also how many minutes or seconds for 1 image?
gtx1070 but those images I sent were done with Nai
what is nai?
NovelAI
like midjourney?>
Yes
But Nai is just best
Well Nai is quite fast but more like their model is one of the best there
ya so price monthly?
but they must be using some sorta gpu right to produce their images
These or 25$ monthly sub for infinite gens at 1024x1024 resolution
Shrug
They have something better than consumer gpu
is there 1080 resolution?
1920x1080 for youtbe?
thumbnails
Yes but it's quite expensive
also stable has 1080p 1920x1080p for youtube thumbnails?
gotte be careful with this "elf" it may bite
AWWWWW i m going otherweise ill die
so cute
so cute omg
NaiV3 is finetuned SDXL model
@nova remnant can`t get it but got this
that elf can bite me
cute neko ears
ooh
indeed
stable diffusion is free right?
Well you can gen on your gpu if you want
Don't yank the hair
this smile? what do we call it?
Smile
No need to thank me like and subscribe
👍
yandere smile? weird smile? evil smile?
souka
guess I'll just caption evil smile
Light smile probably
yeah PROBABLY
nah but shes so good some times

I'm hungry
I'm on my way!
you are a cunning one, brother rabbit! i cant fly!

Bunny hoodie girl selfies
Sd3 is good but it's not finetuned
it's a really general model
so it's kind of iffy with anime in my opinion
You're iffy

SD3 is really good
Bnnuy
@brittle rampart you!
=^.^=
ho! this is taxi for me?
sorry i didnt understand can u explain ?is it free to convert images with AI in generative fill in 1920x1080p and convert videos with AI into anime?
You can take any image as reference and generate anime version of it with AI.
And as for free you can do that with your gpu locally
SD is open source you can use free locally and I'm not sure if any cloud lets you use ai gen free
ooh so NOVEL ai is cloud based versionn of SD?
Yes
Which is faster cloud based or gpu?
locally
locally also it converts text to images?
Depends how fast your internet and what gpu you have
omg so cute that ears
and her hair colour so sweet
Well you can do this but I haven't done it rly since if you want frames consistent it requires a lot of work
okays thx ill try that
wait so i m a bit confused
SD locally with gpu converts text to images right?
for free?
Yes
But "free" is based on your electricity bill
Web based services have their own gpu that's most likely better than you have
ooh so they might have 4090
That's still low end card on this field
electricity bill as in gpu usgae?
Yeah
by a lot, unless your power is extremely costly where you live
i live in india
Also one reason people use web based services is since some have their own model just like Nai which is one of the best model at the moment
my computer with a 4090 and a 1200w power supply could cost a lot if it was maxed out 24/7, somewhere around $120usd/month with local power cost, but in reality it uses more like $6-20/month depending on how much i'm generating and gaming, and even that it probably over estimating by a lot
versus some cloud service that would cost around $25/month even if you didnt use it, and is limited in how many images you can generate
but if u include what that computer cost, it would be quite a lot i would assume
and some days i'm generating over 1000 images in a few hours, so there's no way a cloud service is viable for me
sure, just my gpu was $2000, but i didnt buy it for AI generation
i bought it for gaming and 3D modeling work, it's something i had or was going to have anyways
Doesn't rly matter for casual user
i also have my old 3080TI in my home server just because
perhaps
I have only 1070
now, if I was someone that was not already going to buy expensive PC parts and had no reason to own them besides AI generation, i'd absolutely find some kind of cloud service that works for me
I imagine having $2000 for nai instead of a 4090 would last you quite some time
in fact, i debated a long time even getting the 4090 to begin with because i didnt need to upgrade from the 3080TI and something like NAI probably would have been more cost effective, but ultimately my freelance work paid for the card and it made my AI generations take less time so it's been worth it for me in the long term
so you got two of them plugged in? or 3080ti benched?
the 3080TI is in my server, I'm planning on using it for AI facial and voice recognition for my home automation stuff, potentially using a llm for a 'virtual assistant' type of thing
but otherwise it's just sitting there idle most of the time
Nai is very much worth it if you don't need complete control or privacy
It's amazing for what it does
it genuinely shits on everything anime related we have rn
that's subjective at best, but if you like it, great
ye i put nai on top, beating everything by a landslide
which makes it more depressing that its not free
Where she going to 
Smolll 
Big
👍
it's near the bottom of the list for me because it's not free, for me it's more worth it to spend the time to find a model and settings that produce the qualities i'm looking for (or train something if i have to) instead of just paying for a monthly service for no real benefit
on the other hand, if i wanted to generate specific characters in that particular style, it might be higher on my list
Marina 
That looks sick :o
It me :O
I've finished my training of all Funger 2 girls cx
(the main ones at least)
I wouldn't really say it's better than everything by a landslide. It's got a really good base style and the fact that it has so many innate artists is really good, however models such as Pony or AutismMix are still better with overall prompt adherence and with certain concepts.
If anything a mix between NAI and Pony would be the perfect model in my opinion
I agree too, I don't like prompting in NAI, it feels more stiff than Pony prompting, and the innate artists are trash, my LoRAs are way better
XD
Its good at following prompts, but it doesnt like, stretch as much with natural prompting I'd say :o
Big Loras for the win 
At least in my experience
Loras are definitely another big reason why I prefer making certain images with Pony rather than NAI
Yeee, I mean I mostly do use my own styles, and by mostly I mean always, cause I usually train like smaller/more obscure artists, and specially like twitter artists or stuff like that, that you wont find on NAI
It's good until you try doing something a bit more complex, it completely ignores prompts even after adding weights to them
Thats true too
@magic wave this is you
I think their limiting factor is the fact that they only use danbooru images for their dataset, which is good for anime don't get me wrong but a model like Pony, which has a mixed dataset, does better with certain concepts like I said
O, fair :O
That too, and theres also huge like very niche but strong benefits to local, like SVD merge LoRAs that NAI probably will never have cause its something very manual to do
So basically
NAI = Apple
A1111/Forge SD = Android
Comfy SD = Linux?
I mean who knows, NAI plan on making their own base model soon, they got the money for all of these tools to be added. It just depends on if they want to do it or not I guess
I'd say if you put pony coherent at like 9 then nai is about like 7/10
but style a 10
So i still rate it higher
I mean I doubt they do complex stuff like that, maybe their base model will be really good, but won't have the same freedom
The style is mainly why I'm using it currently, but yeah like I said it struggles with certain things that are quite easy to do on Pony
No base? Damn
I mean, personally I think I like the coherency of NAI but not the style
Yeah i havent gotten around to test pony for coherence
but i still put nai on top because of the style and coherence wise its more usable than average anime sdxl
To put it simply NAI is the best SFW anime model, Pony excels at NSFW stuff but it doesn't have a good base style for anime
Oh hey another style thief just like me :DDDDD
This is pretty much what most people will tell you
These are some pics I've generated with styles I've trained, and I prefer my stuff over anything NAI has xD
XD i also steal styles
I have like 3 loras so far i think
I don't know how many I have, for SD 1.5 I probs have like
Too lazy i just like merge at random weights see what stuff does
60, and for SDXL so far probably like 30
It made pics i liked thats 4 sure
Ye man i cant seem to find some pics it made online
I tried the lora block weight extension, and I didn't like it too much, didn't find much use for the INs or the latter part of the OUTs
Dunno how merging goes, but my experience with the dynamic weights was kinda mid
like people like it that much im expecting awesome shit
At the end of the day all of this is subjective, so it really doesn't matter whether you use NAI or Pony, it just comes down to which one you like more, both are really good models
Lol i made fire stuff with merging loras
True too :o
What do you mean?
Fair fair :o
ok maybe not THAT good but like 3 loras and i got a bunch of styles from switching weights
lemme if i can show any of em rn
Weight switch boom styles @magic wave
I want to see examples (ofc i know those pics can't be here but i wanna know how good it really is)
Yeh, so its kinda the same as the lora block weight extension, very minor diferences :o
{{{{{nai}}}}}
i don't do alot of training now because well im lazy

(((((nai)))))
It was pretty easy to keep track of the good pics people were generating with it on Civitai when Pony first released but nowadays there's thousands of images to scroll through, so it's hard to find those exact images you're looking for
I can recommend you some apps to make that easier :o
I mean, I feel like good pics are pretty easy to gen :o
I use autism 24/7

Got it built-in
Well yes but if you're looking to find some online it's probably gonna be a bit hard to do so. Generating images yourself is much better

Cool thing before i go for a moment, so this x lora has very little amount of images, and poor settings (i dont know wtf options mean), so images are BROKEN. But when i use it on top of this y lora on a low weight, it "stabilizes" the x lora
I'm not talking autism like @lyric zealot
True too :o

And I get better shit
I mean, that just means one of em is bad :o
hmm? which one?
Yeah i have no pc to keep retrying, and the style fidelity is close enough
2nd one :O
Like it helps fixes the X lora
Fair, I mean if you want I can send you my settings for styles :o

Turkey
My fucking phone can’t type
XD

Kek

I wasn't satisfied with the current Cirno LoRAs out there, so I decided to create my own. I have chosen to use 1024 dim to capture every detail per...
That is a Cirno :o
black bars
Sitting Samarie
Uhm actually, anime is not allowed in the anime channel, thank you very much 🤓☝️
Now make her do a 360 backflip
She tried that but hit the 4th wall
Uglee, turn her into shrek
Spooki Smarie
It just broke lmao
Pony doesn't know Shrek 
depressed
Wtf, PonyV7 where are you 😠
True
eh!? released?
O thats pretty :o
I wish

Realistically it's gonna take around 3-4 months before it's released
Hands a bit scuffed but this is pretty neat :o
Big gun :o
@gloomy scroll 
I will refrain from making some inappropriate joke about this
for now, astra is still collecting data till sd3 is pubrically released
cute
You mean like, images for the dataset? Is there gonna be more than 10M?
W rizz
no idea how much he collected. but i know he is still collecting
yesh world is done
anyway sd3 is still not public, so, he have still time
nai deleted forever
I think I saw he mentioned it the dataset is currently at 10M
love that effect
no open source 
You're just a cat what do you know
a day ago he was looking for characters to train the model with

I mean its inside
More anime is always a good choice in my honest opinion
Cx
totally agree
yesh
More anime = more anime booba = profit

Shut
I did another train picture
this time is different
the catastrophe is...
Man mixing styles on NAI is so fun but also very irritating cause I can never decide which style looks the best
me wondering will we able to run sd3
This is so cursed oh my lord 😭
Probably, I mean maybe not the biggest model but the 2B one should be fine
ahhh big oneeeee
I have no clue honestly, guess we'll just have to wait and see
May 6th is the release date
Source: Me
hmm sadge if we cant run big one
Even the weaker ones should be better than SDXL, that's all that matter to me honestly
I don't think it's a huge improvement compared to SDXL but it's still a step in the right direction
big = more animes
for suure
wots dat? hiphop cap?
it was just a gen
some weird hat
I put "teacher hat"
Was trying to use pony to generate a girl holding a knife
Did not go as well as expected that's for sure
How do i prompt this shit???
(it doesn't know how holding a knife works)
I just do "holding knife"
1y old model > Pony
it just doesn't work
Real
Maybe sd just not great at doing that stuff lemme see if there's any thing at all that generates this well
Eh nothing really tbh
So i guess controlnet the way
Weird :o
I was able to generate images of a girl holding a knife with autismmix pretty easily
I didn't keep or post any because they were a bit bloody, but it's really not that hard to do
Idk man i just cant do it
🤷♂️
literally just added 'holding knife' to my prompt
hands are rarely done 100% properly to begin with, so some inpainting might be desired, but generating the basics is simple enough
not really sure what you're doing to test things out, but you were going to 'see if there's anything that generates this well' and then less than 10 mintues later somehow came to the conclusion that nothing can do it
there are thousands on different models and loras out there, and many different ways of prompting them, no single thing will produce what you want every time
and obviously, the same model i told you yesterday can do it can do it

he was attempting to stab you with a waifu handling a knife, but you are still there, soo... indead, he failed <.<
well, it's gonna take a bit more than a waifu to stab me 😄
noise like what?
do you want to use this image as noise?
no no noo
no expert in training, but i think depend from what do you want the base resolution to be
if most of the img are 512x512, you will gen better 512x512 images
I would try to avoid them as much as possible, but if it's just a small portion of the dataset then it probably won't be too big of a deal. Also, if it's all you have to train with then you might as well try. In many cases, the VAE encode pass that occurs before the model ever sees the image during training will help to obscure problems like that since the data is getting crushed down into 8x8 latents anyways.
okay thanks, and Im doing a 1.5 lora with 576x576
I got images with
524x766
645x645 res
should I turn on upscale with bucket or not
nubby is the expert
nubby what do you suggest to train a cursed lora that kills the utilizer like the film the ring? XD
Bucketing is basically always beneficial. Turn that on for sure since you have differently shaped images.
Upscaling won't be beneficial and could be harmful.
576x576 = 331,776 pixels
521x766 = 399,086 pixels
645x645 = 416,025 pixels
so both image sizes will already be downscaled a bit anyways.
woow did math also, deym thanks for that
Any lora or lora variant trained on Shrek images seems like it would probably be pretty effective at that. 








