#š„ļ½anime
1 messages Ā· Page 134 of 1
nah I meant the one I'm farting with now, but, honestly... anything. lol
LOL ok
Generic prompt
Model: Lets add Thick thighs
thank you model
yeah its unique
my favorite game series though so
i had to
also since it took a month to learn how to train it now i know how to do another one
when in doubt, CoffeeGirl: 1girl, (\:d:1.0) autumn leaves, bangs, black headwear, blurry background, blush, brown eyes, brown hair, brown scarf, brown skirt, cardigan, coffee, cowboy shot, cup, disposable cup, drink, falling leaves, beanie, holding, holding drink, leaf, long hair, long sleeves, looking at viewer, open cardigan, open mouth, plaid, plaid skirt, puffy long sleeves, red sweater, scarf, shirt, shirt tucked in, sidelocks, skirt, smile, solo, sweater, white shirt, extremely detailed, intricate, masterpiece, absurdres BREAK cute, heart-shaped pupils, atmospheric perspective, full body, mid shot
when in doubt thighs and black hole background
Send me prompts to try in my checkpoint
might as well check to see if it knows Touhou remilia scarlet, mob cap, bat wings, ascot, red eyes, superb, sitting on a rock beside a stream, nature, outdoors, masterpiece, best quality
on it
will be about 3mins Im doing 150steps and then 150 hires upscaling steps
you will only know when you see
CoffeeGirl is her name
She also has a CoffeeCannonā¢

eta 25sec
whats the model
they all look similar to me so i can never figure it out
LOL
KizukiAI
cute!
Im gonna give it 1 more shot with less denoising but here is the first one
Overall, that's pretty close. wings aren't quite right and she's missing her mobcap, but it's mostly there.
Yeah, I saw that after looking closer
forgot to remove black hair is this the character ?
kindof???
idk if my model knows it
it knows the hat but anything else ehhhh
I fixed the prompt a bit lets see
it was scuffed like my gens 
here's an example of how she should look
Next one is about 10% closer but I will still send it
the dress, ascot, and hat are 100% Touhou. everything else, not quite
bruh
Not every model has to be good at existing characters. Here's one that will test holding a staff/weapon and heterochromia: 1girl, ahoge, blue eyes, blush, dress, eyes visible through hair, flat chest, hair over eyes, (heterochromia:1.1), jewelry, necklace, pointy ears, short ponytail, solo, off shoulder, white dress, white hair, BREAK indoors, counter, adventurer's guild, (fantasy setting, priestess, adventurer:1.2), holding staff, hair flower, comfy, a sense of depth, spacious room, superb, extremely detailed, intricate, best quality masterpiece
not bad. the style is really nice and she's cute
default
pat pat dats a good model
I can't hires don't judge 
this is an old merge, god i need to fix the uncanny valley face. also i think a1111 didnt see a vae
its supposed to allow realistic faces but lke
idk if i need adetailer or something
adetailer would probably help a lot
its a weird mix of anime and dreamshaper, so probably
dreamshaper has always been uncanny
I feel like dreamshaper needs prompt
its hard to use
Do you guys wanna see my negatives,
ok
(easynegative), out of frame, lowres, text, error, cropped, worst quality, low quality, jpeg artifacts, duplicate, wrong proportions, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, username, watermark, signature, bad hands, (low quality, worst quality:1.3), backlit, corrupted image, letters, words, censored, multiple fingers, multiple legs, bad hands, blur, pixelated, bad anatomy:1.2, cropped hands, artist name, extra digit, loli, fewer digit, author name, names, autograph, hand writing, written text, five legs, five legs, 3 legs, three legs, five legs, 5 legs, bad hands, bad legs, bad body shape, multiple tails, two tails, deformated body, imperfect painting, blurry hands, blurry face, blurry body, blurry image, bad eyes, ugly face, ugly eyes, repetition, ugly body, child, kid, underage, shota
tbf i dont think ill use it actually, i just made it for a friend who had older aged characters and needed something done
just... use a TI
or 2 or 3
No need for all of that
here's another one for testing 1girl, (fennec:1.2)_ears, orange hair, brown eyes, hair bow, sidelocks, (messy hair:1.1), long hair, backpack, hiking, outdoors, nature, full body, mid shot
negative_hand-neg, (KHFB, AuroraNegative:1.1), (Worst Quality, Low Quality:1.4), border, grayscale, (watermark, multiple views:1.2)
all I use cough cough got from someone
apparently my blah checkpoint is a little stubborn with anything pertaining to items
honestly, the majority of this is not helping at all. images aren't tagged with things like cloned face or any of this five legs, five legs, 3 legs, three legs, five legs, 5 legs, bad hands, bad legs, bad body shape as a few examples. CLIP likely understands this negative as you really don't want legs and then it fights with itself because it knows that something is supposed to be where the legs should be.
yep, the staff could be a better, but at least she's basically holding it right.
This is a 20 step with 20 step hires
size doesn't matter
I did (holding staff:1.4), thought flowers where close enough
its flammable, BURN IT
improved
LMAO
its the stupid nose
@supple raptor Look at this POWERFUL fire staff!!
different sampling method changed something
candle-on-stick, truly the ultimate weapon
whats it made of?
dpm++ 3m SDE Karras
This is basically what i did 2/3 counterfeit, 1/3 anythingv5. and then I took those merged and threw AnyLoRA 1/2 way
after that I merged
counterfeit XL
and THEN
counterfeit is probably still my favorite model ngl
What I did is merge SDXL Yamer's Anime
2/3
so its jam packed
and the model I was using is before the XL merges. I merged XL counterfeit with SDXL Yamer's Anime and then merged the 2/3 counterfeit, 1/3 anythingv5 | AnyLoRA 1/2 to get Pentellion
20 step XL is better at actually following the promps
Is this the flat chest prompt ? 
yes
its probably the merges
Ill upload my sdxl to civitai tonight
but it finally finished a 150 step img
that counterfeit drawn style
That woods is noice
The XL can do things like this
send prompt
(Anime male: in suit), built, buff(sexy),(abs), drawn:pier:background, (Dark:Red:black:vibrant:lightning:background), japan, (dam:in background)
lol
It made an emoji
kek
Well that is one neat sword.

Something weird with that ferris wheel
didnt see it ontill you pointed it out
Speaking of which, I didn't even notice the water until now
I'm really starting to think my model is overtuned... "Teenagers"
That said, I do have something really cool to show
I saw these images and thought
Maybe I should try something
Seems a bit off but getting something
It's nice that it'll actually draw men well
I've noticed many anime models make them super feminine or very samey
The magic of prompt
And yes I'm glad
I actually have men
I kept getting super boyish/girlish characters
I was kind of getting Supernatural vibes from this
But wait
You want to see something super epic?
Might be worth upscaling
The sun and moon are impossible in this image
I love everything about this but his hands
How i can edit my photo
Time to get some prostheses
Ty my friend your kind words always bring me joy
Ok, well, I did end up creating a new program.
SD Quick View
ill get it up to github soon, it's almost ready
But, I need an icon. š Anyone wanna do an icon contest? š
what kind of icon
Just an icon for the program. This is my leading contender.
Needs more thighs

it's gonna be a tiny icon š
this
this will do
not that
yes that
fine wb this
you tryna make her cry? 
oh no you gave her cancer
Lmao
someone grab the plunger

Yooooo, better rename those models to baldnet

Nothing wrong with balding. Keep your head up
very cute
could i have the parameters for that plz?
now you guys made her self conscious and she wont take her hood off 
Bobs 
Thighs 
cat refuses to acknowledge my bald girl 
not again lmao
spoopy
I lovew taht hair tyyle
I gotta do this now
This day is for the birds. š®
still need an icon....
Tell her I said sorry
its like i can still hear them
whats the prompts cough
I did not prompt for anything that you are paying attention to in that image
I'm trying to get a certain style
not working well
I just woke up 
the RNG angle š
@lyric zealot 
Soosi with double pfps? 
I guess thatās a bug?
good morning! or afternoon idk lol
Seems like it 
colors like this, also @lyric zealot Do you have any idea of a lora that can do it
Itās morning 
for me its 9 pm rn
Same for me 
I donāt think so
Might have to look for one
hmm what country are you from?
O it happening again 
Argentina :o
yooo im from brazil :o
Is it because Iām on mobile
Not sure, now its the normal soosi pfp, the red oen 
O I think I remember you told, I think :O
Sus is normal for me
Yis 
funny discord moment
True
cute
4'11
I think thats something like 1.50 
i dont remember but maybe, i once was kinda active here, i dont think i ever posted any of my generations here though
you mean a normal way of measuring stuff
90cm
Exactly 

Getting smoller 
HECK YES... I just solved a memory leak that took me like 2 hours
Fair, and I think you may have said it :o
Small enough to be a chibi, so we can boil them 
oh my gosh what a thrill... heart is pounding
10cm
ty!!
Download link soonā¢ļø

The unprompted twin 
Bobbers
Lewd
I'm so tempted to get a quest 3
@nova remnant Ever did more women in bikinis?
here metadata
tysm!!!!
V1.0 now available
Open source and binaries available: https://github.com/GarlicCookie/SD-Quick-View
Mini Milkā¢ļø
prompt šš
A little 
lol
I think a shrimp girl would look a little scary
Or maybe cute
do you wanna try
Too busy making mini milk 
We're getting FASTER (Faster) HARDER (Harder)
[now your turn]
almost
Mini milk 
O, I thought you were referencing Daft Punk, but its... Scooter?
correct
O, wao, I didn't know about that song 
Ok, how about this, then?
the entire music video is just them driving dune buggies
I think they just wanted to drive dune buggies
so they made a song
what's wrong with using feet to measure? I have 2 of them with me all the time and naturally we'll divide them into 12 segments for measuring smaller things, we'll call those inches naturally. oh and for really long distances we'll just use a nice, even 5280 feet and we'll call that a mile. I can't think of any other possible system that would be easier to remember.
I mean, fair, that is a understandable thing to want to do XD
And how do you measure... an inch?

it's one 12th of a foot
And how long is a 10th of that 12th of a foot 
XD
I think Smile means feet aren't created on an assembly
different sizes
yes, I thought the sarcasm dripping off my explanation was obvious enough but I probably should have added an /s just to be safe. good to have you back btw
thank you, hopefully this is the last time
I mean, I got the sarcasm, I wanted to push it tho xD
So apparently the British are to blame for the asinine 5280 ft in a mile
The basic concept of the mile originated in Roman times. The Romans used a unit of distance called the mille passum, which literally translated into "a thousand paces."Ā Since each pace was considered to be five Roman feetāwhich were a bit shorter than our modern feetāthe mile ended up being 5,000 Roman feet, or roughly 4,850 of our modern feet.
If the mile originated with 5,000 Roman feet, how did we end up with a mile that is 5,280 feet? Blame the furlong. The furlong wasn't always just an arcane unit of measure that horseracing fans gabbed about; it once had significance as the length of the furrow a team of oxen could plow in a day. In 1592, Parliament set about determining the length of the mile and decided that each one should be made up of eight furlongs. Since a furlong was 660 feet, we ended up with a 5,280-foot mile.
Rome also had their army slaves lay the brickwork for roads as they travelled to war so it would be easier for the next legion
You think that's bad, they used to use several other units of measure that somehow made even less sense
The definition of the Mile was changed around 1593 during the reign of Queen Elizabeth I of England. The definition of the Furlong was changed from 600 feet to 660 feet. Since that happened, the Definition of the Mile has to change since a Furlong is 1/8 Mile, The defined was increased to 5280 feet, 320 Rods or 1760 Yards, or 80 chains.
a rod is 16.5ft
wtf is this shit?
oh god you just made me remember no wait I forgot again
Mini Milkā¢ļø 
Sir, I'm sorry, but we're speaking in imperial measurements here. I'm going to need you to convert those decimals to fractions


We need a board measured at 37x23 penises
just cause ancient rome
they had a phallic obsession
non sexual in my scenario
True
I mean, ancient romans were the real men
No women in the love making, only men

gygakad
well, there wasn't any concept of homosexuality in ancient rome
I just imagine that south park episode with Mr Garrisons...on a mouse
running around singing
like that happening in ancient rome
@magic wave 
I don't watch South Park, maybe I should 
You made your own milk? 
South Park is the best
I should put animal ears in the negative, this model loves them
old South Park was pretty good. later seasons lost the magic
old nubby
š“
nah I think seasons are a bit hit or miss, it's just more noticable when a season only has 6 episodes
as opposed to 12+
What you mean? 
Get new GPU 

I'll ask in the streets money for one to make cowgirls
With a humble purpose like this who wouldnt help

Look at png info 
I should've tried that
š

Yis :o
the project ai discord tends to be more active
But I like the people here 
stop with those weird fetishes
š«
heres a girl,enjoy š


evenin'
In case you didn't see my new tool. #š„ļ½anime message
Said I'd do it and I did. š
ooo I got 25 new dalle3 tokens or whatever...
now make him pregnant š«
Anyone try sdxl with anime yet
Model? This has great coherence
dalle3 š¦
my three fingers
Figured 
I need it local š
Probably would need 48gb vram
Amount of parameters must be crazy
elevated pillow
Fixed the hands, kinda
evenin' @past junco
Good afternoon
š
(view from behind, in my dms:1.5)

my honest reaction
fake sky 
smoking is bad
twins as an apology šāāļø
whyzat?
coding stuff 
I'll go insane with her
but what about me
"Shopping cart with a orbital laser weapon attached to it, destroying a tank"
Bro dalle3 can't aim
lol
š
Dunno her name, but she's hot. š
ii cant find the upscaler you use
well heres two of them
you mean 4 of them 
I wish dalle weren't such a prude
I wish I could get more credits
it's worth using a token just to get 1 of these
now try to do outpaiting to see if sd can replicate those screens
I actually don't know how to outpaint
I dig the measurements of both women.
I can do that in paint, just copy and paste them, why would it be hard?
š
heres my special tool (idk whats she holding
)
Extended pro version of a wine bottle opener

@native halo let's watch the sunset together ...

"Um... there's something behind you"
I know lmao
@magic wave
Probably more than that. It seems to be running the prompt you input through an LLM before running that through the text encoder.
Good
actualy
Man just when I least needed it I got woken up by bombings
@magic wave better version
idk what happened here
iboy became very generic
reminds me back in July I was trying to gen an OC
I think the one I'm generating now looks a bit like a blonde version of Staz
This guy (not generated)
Man I can literally feel the impacts of ones exploding nearby
stay safe, best wishes 
Thx
Woah what the, hope for the best
I think it also goes by x4 Ultra or something like that 
let us pray
Updated my new viewer with some quality of life features: https://github.com/GarlicCookie/SD-Quick-View
v1.1 available here: https://github.com/GarlicCookie/SD-Quick-View/releases/tag/v1.1
Cow
You don't think it's still possible for me to train a Lora if I use 1:1 ratio images?
why not?
I'll try it, though I'll do it at a later time

prayin for i have sinned š
You mean all the images are 1:1 ar?
That would work but if you were to use that lora on an image with another ar, it will be less effective
Tho it will work yeaj
kinda sussy ngl
It's a smol sus
Pregananant
yikes
real
got me and my waifu by mistake
was supposed to be "bad cosplay man as anime girl" but instead she went with him to the convention

long time no see 
hope you are doing well 
yep, I got eaten by life for a while
a little better yeah, and you, going all right ?
always the nicest place to come back to at least, it seems to be calm around here but the pillars of the community are still around š
well, people leaves for awhile, then they will be back, that's how it is
i mean, i also left to game for awhile, before coming back 
same, some good old gaming did lots of good
what did you play recently ?
I burned time on BG3 for a good while :p
such a good game though. and voice acted to the perfection, letting you play even when super tired
I loved all of it
true true
i still need to complete it tho, i'm nearing the end
but like suddenly lost interest in gaming totally
so back to SD 
it was worth it to the end for me yeah, I'm giving it a break now, and I'm sure I'll do another playthrough one day
I came back with a project for SD too š
I'm going to try and train a good model based on my late mother's art. She had lots of things done, and also lots of things unfinished. I want to see if I could train a good model on it, and complete some of her pieces she never finished
so no sneak peak for now
ahhh
that's good to hear, i hope you succeed 
I still need to train my first ever lora/model
me too ^^ even if I'm not sure how far I want to push the "experiment" ethically
complicated
like, should I paint "superman" in her style ?
that will be up to you to decide i guess
possibilities exist and are strange to consider
well, it will be up to you to consider what you really want
nonetheless, im sure you will acheive your goals

thanks ā¤ļø and yeah, I believe too !
plus she had enough to train correctly from what I understood in my lots of models
any specific goal in mind or just wanting to try your hand at it ?
currently nothing in mind tbh
but im still really limited in my vram
and don't want training to take a long time
as such, i'm mostly waiting to upgrade first before training

yeah, logical
I still need to check what's new, I just came back yesterday
I haven't tried XL since it came out
I suppose training must have evolved quite a bit too
i think the big thing would be sdxl, not sure what other major stuff tho
not sure about training wise, might have to ask nubby 
also an idea that sprout into my mind those last few months
imagine a new "pokemon" game

but all monsters are evolved using img2img. by hand of course, not during the gameplay or whatever, done before that during development
but you may build a mega roaster of monsters with great graphics, relating to each others, and with a lot more possibilities. I was thinking in particular, all pokemons coming from a single same pokemon, like a slime or something
the morphing abilities of SD seem too perfect for things like that
well it's not a concept, just an evolution on a concept already existing, but I'm mostly taking the "evolution" and capture part of pokemon. gameplay doesn't need to relate at all.
I was exploring this idea, using a pirate theme, where all monsters are your ship's crew
well lots of concept on paper for now, and I'm trying to test if that idea of evolving everything from a single slime could work

might work, need to set up denoising strength to be not too high,
and maybe after multi img2img, you can get something totally new that dont even resemble the first creature
yeah, and masks to evolve only parts of the body too. I was also looking through my collection of models to decide what to use
given the tests I did on loopback i2i back in the day, I think it really needs to be driven carefully. going too much into it brings lots of magenta and artefacts.
Using some controlnet, layers to split the creature in parts and evolve each part, ... those seem possible
I'm trying to find a way that is still quite fast to produce those, to keep quality and quantity high, but I think I'm aiming too high, and should limit more my scope for a start
yup, sometime like a feedback loop?
im almost certain something like that came out recently
I'm not sure it's possible to put out any game made mostly on AI art currently though. from what I saw, Steam seemed to be refusing those
yeah, it's what I'm thinking too. plus going photoshop on all creations could also be sufficient
it's really complicated, I'm not sure. it needs to evolve in a clearer way for sure
unless your game struck gold and becomes really huge
yeah, but even then, look at those "assets stolen and pasted" games like "go up", going on steam frontpage for a few days, getting delisted after
people try and do it it seems x)
never got the concept of those rage games. it makes me rage quit fast, only thing it got going for it x)
well, i think most of the assets are just on free on unity/unreal engine(not sure which), and people can just grab it
most are yeah. some were problematic and brang the delisting, but most was clean
I think it got solved since, removing the problematic assets
writting on a keyboard.
fun interpretation xD
that reminds me, I need to get back into LLMs, and see how long term memory has improved
I'm sure people have been working on things like that, this was too fast evolving when I left
what's your bot about ? (if you can talk about it)
haven't really pay much attention to LLMs, but chatgpt have eyes now 
well, it's a super simple discord bot that uses automatic1111 api to push for generation on my local system
but i can't get it to work concurrently

damn
I like bot coding but I'm not a fan of discord itself x) I did bot things on IRC long time ago, but my experience boting on discord has been a lot less fun
I did try to deploy a bot that was doing just what you described, not sure I remember the name, it was an open source project
I think it was this one
yup i saw that, and was trying to understand what's going on with it
so that i can implement the queue system to my own version
but it's confusing
and i think discord api updated
so more confusing
concurent user is always a complicated part. at least you have no "latency/user synchronisation" to handle :p
I used this one for a while and did a little tweaking of the code locally. I should have made my own fork but never did. It is primarily a chatbot, but has A1111 capabilities for image gen too. https://github.com/mercm8/chat-llama-discord-bot
smart-dog is Veggie's project. It runs on GPT 3.5 Turbo. Mine were only ever online for short periods since they ran locally. I had Makima, 2b, Yui (yandere) , and Kyoko (tsundere).
I really need to catch up on the state of the LLMs
Same. I haven't really looked around probably since you took your break.
I haven't even tried Llama2 yet.
it must have moved, it was going full on "automatic1111" with plugins from what I remember
I just pulled up the Ooba repo and I think every folder/file in the root directory has seen an update since I last ran it. XD
Wow, I don't know how you managed to have each character clearly defined, but it's very well done.
MultiDiffusion or something like that on Comfy?
Hey @wind girder , how are you?
hey there Seb ! I'm ok, back for a little more fun around here š
haven't used sd in months
why is this channel so inactive now?
i'm trying to perform better with comfyui and this is taking more time than expected
well, people come and go
most people got really excited at the start of AI art generation, and well people lost interest as well

but those that stay, will probably be staying for a long time
and most would come and visit once in awhile
at least the time generation decreased right?
upscaling are way more faster
512x512 are pretty much the same
on auto 2048 / 2048 was way out of my league
i can at least gen at decent speed for sdxl on comfyui
but sdxl model isn't enough for me to make the switch yet
I cheated - Dalle3. š
I can get similar definition in SD but not with the coherence.
bruh wtf happend with sd... i didnt even put like.. any work into this
No good sdxl anime model yet š¢

Well, I said I'm a Game Developer if that's what you're referring to
Yeah. š It inspired me to finally make another helper program for SD images that I'd been talking about.
so, if you're interested: https://github.com/GarlicCookie/SD-Quick-View
Oh? What's that?
That's rather neat actually
I would have thought one of the extensions did that
But yeah this is really nice
I'll try it
I wanted a native, non-web
Does it read the generated .txt files?
No, only metadata built-in to the generations.
This is the one I did a while back for easier curation. I use this after generating large batches to curate. https://github.com/GarlicCookie/PNG-SD-Info-Viewer
I wasn't sure the files held that metadata
Yeah the metadata can be stored in the file.
One of the things I was trying to do was ID all my images and give them tags and put them into seperate folders, and then archive the whole set periodically
I should specify that these work with Auto1111. Not sure how well with other tools, though they might.
How do I check the metadata? Hold on
Automatic1111
oh yeah you should be good
in the interface, go to the PNG Info tab
load in one of your generations and see
In Automatic1111?
I need some good keywords to stop the "dress in hand" thing. This one came out well, but usually annoys me.
yep!
I was trying to check the files themselves
they store it in a weird way
it's in like a custom exif field that doesn't easily show up
Give me a sec
I didn't even realize that this was a thing in Automatic1111
Looks like she is holding a rubber chicken.
I thought there might have been metadata and stuff baked in the files but I wasn't sure
Yeah, most of the tools embed the metadata now.
It's really easy to scan all your saved images and pull out all the prompts for a big list. š I do that sometimes and then run a big list against a new model I'm checking out to see how it handles them.
exiftool can do it
exiftool(-k).exe -ext png -p "$parameters" -r "c:\path\to\images" > outputfull.txt
Handy to have.
I wonder... I could write some programs that can help do a couple things to aid the process. I don't really like Python but I could try something
And I see
That's a neat idea
wow Ive posted a lot of images this morning
ran a big batch last night, so ....
That Muk Monsieur? 
a hovering cigarette, a woman struggling to get out of her slave collar, a balloon resting on a needle inside a glass of cornflakes
what the green lady and the pokemonish waiter are...I dunno
a balloon pokemon waiter
and Pikachu cursed to sit on a needle with no face detail
the dude on the left really gets me
this is cool
Hands are broken... not even inpaint could fix them
She's very beautiful as usual.
@lyric zealot How does she keep on expressing herself so well, to say the least?
Hands are still messed up
this character comes out well I think
whatever a 'rwby / misralrose' is... no clue
wish I could put in some sort of negative prompt to STOP THIS though...
A beautifully expressive girl in this pic, preparing for the winter.
the coffee girl 
Inpaint harder 
I am so bad at inpainting ong fr
šthick
big meow 
no idea honestly
but something feels off when prompting
š
omg guizmus
cute
the one and only
omg yes š coucou Kazoo š
what's this model?
is that even anime tho
not sure, I was going through my best of folder. I believe it was this one
how is that a question
?
I'll auto ban myself as punishment
man I wanna see a water color anime
tho it would take them halff a year to make like 1 episode
frfr
Is there a channel for watercolor stuff >.>
yeah, maybe one day with AI's help though, could go a lot faster
hehe
we have lots of different channels by theme, but none specificaly watercolor. too many channels mean they are all dead, so for things used rarely, we merge channel. closest would be in #1070829540772937728
does bad cosplay of anime belong in anime channel ?
yes
oh god
lol
lol
Fair, style looks like Muk Monsieur 
bad cosplay man as ryuk
Is that Saitama? xD
yep x)
Not something I expected to see today

XD
I typed "bad cosplay" on google, took the best ones, cropped, trained 20 minutes on captions such as "bad cosplay man as my little poney" and tadaa
some booba chat, clean your eyes from that
I haven't heard of Muk Monsieur until now
just googled
I do recognize some of their works tho
hehe
i only find the pokemon lol
but honestly, this model is weird, everything looks like out of focus, i don't like it
probably cuz of training data
I see many ||questionable|| things 
googly eyes
i just try with another one, i downloaded 80g ofrandom models
@wind girder how many bad cosplay pictures did that take
i'm looking for a gummy and colorful model
42
full dataset captionned here

(I put the dataset with my models too, it's on the link I provided almost all the time)
bad cosplay man as pikachu
OH GOD
the hand
the daughter of et








