[Drako]

Yeah I know what you're thinking. "You can make anime girls with AI, how the hell"? As crazy and amazing as it sounds, it's possible. And no, they don't even come close to looking like a 5 year old drew it, they all look professional!

But how is this possible? It was all possible thanks to a group of University students from Fudan University, Carnegie Mellon University, Tongji University and Stony Brook University who teamed up to make this project. But how are the characters actually generated? This project uses a Generative Adversarial Network (GAN). And the AI is then trained using a method called DRAGAN.

And with all of this hard work, they're still dedicated and keeping updates on the project. So, want to check them and the project out? Just check out these links below.

Official site - https://make.girls.moe
Their GitHub - https://github.com/makegirlsmoe

[Blink]

I'm pretty sure I posted this a while ago. Anyway, since then I've discovered some more things:

MGM accomplished that much by making the problem easier: they added some light supervision in the form of a crude tag embedding5, and then simplifying the problem drastically to n=42k faces cropped from professional video game character artwork, which I regarded as not an acceptable solution—the faces were small & boring, and it was unclear if this data-cleaning approach could scale to anime faces in general, much less anime images in general. They are recognizably anime faces but the resolution is low and the quality is not great

And so now we arrive at "This Waifu Does Not Exist" and "These Waifus Do Not Exist" (the latter version is nicer; also see https://www.gwern.net/Faces and https://www.gwern.net/TWDNE for more information).

What's also interesting though, is similar projects such as "This Person Does Not Exist", and "This Cat Does Not Exist", which both give results of good quality.