Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Friday August 18 2017, @09:43PM   Printer-friendly
from the anime++ dept.

Just when you thought that anime character designs couldn't get any more generic, machine learning comes to the rescue:

A collaborative team from Fudan University, Carnegie Mellon University, Tongji University and Stony Brook University have created a generative adversarial network (GAN) that can generate high-quality anime character drawings with just minimal input from humans. You can check the demonstration website and make your own anime renders by inputting some basic preferences.

You can choose hair and eye color and then decide on what accessories you might like such as glasses, hats or ribbons. The site will then generate a character for you based on your inputs. There are some things that could be improved for sure, but overall, it's a really fun application.

Researchers used a technique called DRAGAN to train the AI. The demo site was built on ReactJS. The scientists have summed up the full process in more detail in this technical report. The source code is also available from here.

Of course, designing a character is only the first step. How about a custom Gatebox holographic waifu? Or import your character into an advanced VR fantasy world powered by petaflops GPUs. Which is just a stopgap measure until you jack into the Matrix directly, allowing you to caress your loved one (as far as your brain is concerned).

Is this what happens when you poach Carnegie Mellon University's top scientists?

"Moe" (pronounced "mo-ayyy") refers to "kawaii"/cute characters.


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by takyon on Saturday August 19 2017, @12:49PM

    by takyon (881) <reversethis-{gro ... s} {ta} {noykat}> on Saturday August 19 2017, @12:49PM (#556336) Journal

    There might be a hand-waving 👋 solution to that. For example, just declare VR to be 2D because you are ultimately looking at flat smartphone-sized screen(s). For the neural VR, just say it is moe's graduation to 3D. It could also be possible to simulate an anime style where every frame perceived by the eyes still looks kind of flat. Not sure if that would make the user puke, but that's why we have early adopters.

    --
    [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2