Not generating your waifu with GANs

>not generating your waifu with GANs
make.girls.moe/technical_report.pdf

Other urls found in this thread:

make.girls.moe
github.com/makegirlsmoe
twitter.com/NSFWRedditVideo

They all look like the same person with dyed hair.

Is this how you make anime real?

make.girls.moe

>girl

I win

No, I win.

C-CUTE
Thanks, OP

What the fuck is wrong with you people

>Posting in a website that was primarily for anime
What did you expect

rip

Anime website.

Not bad

...

...

hnng

why do NN always produce such small images

Cute, although generic.

unless it's waifu2x

>They unironically used muh eroge shits for their databases

Why? There is a website name pixiv

...

AI will make anime real.

which is deviantart for weebs, what a great idea

As expected of the Chinese.

>Anime
Hi.
Go to hell.

I'm pretty sure they're still producing better artworks than current Jap '''illustrators''' who should be retired 10 years ago.

Now we need a version of this program that only generate smug anime girls.

Chrometards BTFO/Apple wins again.

thanks op
I found the one

I like cute anime girls

>First publised as a Doujinshi on Comiket 92, Summer 2017
Huh?

ying fucking tao

who dat boy

owo

Definition of doujinshi in Japanese is more than hentai manga; It means books published independently.

Thanks for the info, user!

This looks like cool technology... Can't wait until the backend is open source.

Isn't it already open-source?

github.com/makegirlsmoe

perfect

This repo contains the front-end part of make.girls.moe.

Model training scripts will be open-sourced soon.

Then make an issue on github

...

What?

I mean this

Blush: on

this

That's rude. They already told that they will open source it soon, why rush them?

Blush: off

Needs work for sure. That's not blush off. That's the ahegao.

for the lulz

That's due to different angles of images it was trained on.

Thank you for the explanation but I did not ask for explanation I just stated it clearly needs some work.

It also seems like Glasses: On makes the image a lot more distorted than if glasses were off.

I did not ask you what did you ask for.

Shouldn't have replied to me then.

Actually, YOU shouldn't have replied to me.

I made two posts without replying to anyone. You replied to one of those.

If you set enough parameters that it can't just take an image that already exists it starts looking like a bad copy paste of different images.

Well, next time add "no unsolicited replies" disclaimer, ok?

We should start using this as a benchmark to measure single thread performance. It takes 18 seconds to generate a new cute girl on my machine.

I'd rather correct you next time.

But you didn't correct me?

I'm sure it's done in javascript so it would depend on the browser, too.

I made a post stating that your response was uncalled for.

...

Yes, but that's not a correction.

Call it what you wish.

Am I kawai user?

...

>although generic.
What exactly is the problem with this? Who cares.

I'm just correcting you.

>Touhou in-game artwork.

I still consider myself to be correct.

that would be dumb, character sprites are definitely a much better source to train it on.

90% of art on pixiv is garbage.

GOD LEFT ME UNFINISHED

user, have you ZUN's drawing ability?

No, to be correct you should add a disclaimer that you want only strictly related replies next time.

It's always implied.

...and 10% of art on pixiv is great.

It's not implied anywhere. Nature of this website make replies extremely chaotic.

tout moi

Fuck off with this reddit shit.

You're right. I'll stop.

Why is that?

>Even a neural network can draw better than ZUN

>I'll stop.
don't, your autism is fucking funny

Here's one put through waifu2x.

He has improved slightly over the last decade though.

...

that's ableist

no one cares, FAGGOT

...

...

Original.
w2x y/n?

naw nigga

Alright, I was thinking the same anyway.
Wish the site could generate larger images.

w2x is pretty shit when it has little to work with

Silver hair and red eyes is a good mix

That's due to different angles of images it was trained on.

...