We’ve already seen how AI can give Street Fighter characters human faces. Now, let’s see what it does to manga ones.
YouTuber AImikan took images of manga characters and feed them through the artificial intelligence composite image site Artbreeder.
But Photoshop, this ain’t.
For example, it’s not possible to simply add a mole or a scare here or there. The process is not passive, but rather, it can be time-consuming and taxing. AImikan can spend as little as thirty minutes on a single character and as much as five hours.
The site allows users to upload images, composite them with AI images, and then tweak their creations. In the portrait mode, there are different categories, such as “Age,” “Eyes Open”, “Happy,” and more than can be increased or decreasing, changing the appearance.
This allows users to create realistic — or sometimes, vaguely realistic — versions of the characters.
On AImikan’s channel, the YouTuber explains the goal is challenging whether it’s possible to recreate manga anime characters with machine learning.
“Since it will be reproduced by machine learning, there will be ‘characters that are not quite similar,’” wrote On AImikan, asking viewers to be forgiving about certain aspects.
What makes its use with manga characters so fascinating is, like with many video game characters, they are highly stylised depictions of people. However, one key difference is that manga characters are 2 dimensional drawings. In comparison to many 3D video game characters, the game between manga characters and real-humans is even greater.
Moreover, with so many manga getting live-action movie updates, the AI versions provide a fascinating look at fantasy casting choices. These are not real people! But if they were, some of them would be perfect.
But not all of them, because using AI does have certain limitations. Unlike Photoshop, where your imagination can run wild, Artbreeder operates within the parameters of machine learning. This, however, is what makes the results interesting.
Before to subscribe to AImikan’s channel for future videos.