Anime characters generated with neural networks

Nobody cares.

File: perfect-waifu_fb_7034507.jpg (24KB, 300x300px) Image search: [Google]

24KB, 300x300px

>>9124814
Using /sci/ence to create the perfect anime waifu.

File: BRILLIANT_.jpg (62KB, 795x595px) Image search: [Google]

62KB, 795x595px

>>9124814 >>9124820
Brilliant!

>>9124820
It seems to be rather advanced, it's just that Japanese researchers found fun to use a repository of anime style video game characters as a dataset

File: a.webm (271KB, 128x128px) Image search: [Google]

271KB, 128x128px

>>9124968
What else would you use?

>>9124819
I do.

Neural networks are awesome, and anime can look good.

>>9124814
They arr rook the same

File: IMG_0894.jpg (376KB, 1536x1082px) Image search: [Google]

376KB, 1536x1082px

>>9124814
>authors are all Chinese

they can't keep getting away with this

this is excellent, technology is well on its way to making anime real

>>9124814
Dont pretend you dont know how to spell that, you loser.

File: 1502674239135.jpg (43KB, 499x341px) Image search: [Google]

43KB, 499x341px

Can someone ELI5 how convolutional neural nets work and what makes them different from conventional multilayer perceptrons? I'm having a hard time wrapping my head around what the "convolution" part actually means. I know they're translation and scale invariant, but that's about it.

Looks like any other RNG generation

What is unique about this in comparison? It's just using preset "forms" right?

Is this any better than other anime generators?

File: giphy (1).gif (314KB, 512x384px) Image search: [Google]

314KB, 512x384px

>>9124814
Top tier scientific research right here.

>>9125219
>It's just using preset "forms" right?
I don't think so, it learned all patterns from images itself (i.e. it learned how to identify eyes, hair, mouth, nose, and how they might look, and that an anime character usually has hair, 2 eyes, 0-1 noses, 1 mouth...)

>>9125204
They group parts of the image together in the assumption that pixels next to each other are not totally independent, and calculate weights etc. for those "sub-images"

File: index.png (51KB, 128x128px) Image search: [Google]

51KB, 128x128px

fuck

File: A61tsQ0.png (538KB, 640x810px) Image search: [Google]

538KB, 640x810px

>>9124814
Do it work for anime traps?

>>9125219
I guess you can dumb it down like that. at a very basic level its just a very non linear function that takes properties (hair=brown, eyes=red) and returns anime pics which was fitted to pics of anime girls that are labeled with those properties.

>>9125096
>This paper is presented as a Doujinshi in Comiket 92
Kek

MAKE ANIME GREAT AGAIN

>>9125372
Draw a girl - call it a boy.

And sometimes the girls are not even flat.

File: 4f3fbb93f4000_208066b.jpg (63KB, 600x600px) Image search: [Google]

63KB, 600x600px

>>9126142
Kek

>>9124814
Been posted two thousand times.

>>9124814
That's cool, I GUESS. Now what I really want to know is how long until someone downloads all of sad panda and adversarially generates PORN

>>9126142
HAHA OH WOW. Damn, I have a new goal in life now.

>>>9124814 (OP) (You)
>Been posted two thousand times.
Not with this type of generator, it's far more powerful in a way since the input space can be continuous . This technology can be used with another dataset and tags since it's based on training

>>9127476
Most toolsets are open source, there are many tutorials on basic techniques and since computing power in this case is linked to GPU's power it's gonna be available one day for nearly everyone, with the help of Moore's law and vulgarisation.
The best way to speed up this for porn is to make good datasets aviable; well tagged hentai doujins collection anyone?

>>9127724
No, I meant that one. Though I see they did some improvements in the last weeks. There were +10 threads on /a/ in 24 hours about it and quite a few on /g/. Interesting technology indeed.

>>9127748
>>make good datasets aviable; well tagged hentai doujins collection anyone?
You'd be surprised how well hentai is tagged

>>9125096
>the communists will save anime
Marx is smiling in his grave

>>9128113
It isn't.
>still no racial tags, just interracial
>don't differentiate between male and female NTR
>there's no double creampie
>no tags about specific number of people for orgies greater 3, just gangbang

>>9126287
Actually it's more like...
Draw a boy - call it a cat
Funny to see that to this day wizards keep turning people into animals...

:3

>>9124814
>no training/genetic algorithm to tell the AI what I like/dislike about the waifu and let it generate me a better one
discarded

File: __emilia_re_zero_kara_hajimeru_isekai_seikatsu_drawn_by_silver_chenwen__d8d790e85b550ca8872010151b533801.jpg (1MB, 1750x1750px) Image search: [Google]

1MB, 1750x1750px

>>9124820
For me, it's Emilia-tan

>>9128168
You've never been to sadpanda. They are mostly extremely well tagged.

>>9128235
https://www.youtube.com/watch?v=H_NmRuW6848
is this you?

>>9128116
>"socialism with Chinese characteristics" is communism
Go home, Deng.

>>9124814
Fourth row left is nice

>>9128116
>>9128800
>l*ftism
Fuck off to >>>/r/eddit/

>>9125295
Shrenpai I wuv wroo...

File: 19761987_467152603648527_1087157721284739072_n.jpg (42KB, 480x599px) Image search: [Google]

42KB, 480x599px

>>9124820