Artists Use AI to Make AOT Characters Look Real

This post contains affiliate links. If you use these links to buy something, I may earn a small commission, without any extra cost to you. As an Amazon Associate, I earn from qualifying purchases. Thank you!

Seemingly impossible to imagine how Eren would look like a normal, real person, right?

Well, not anymore! The age of AI is already here.

The realistic portraits of Attack on Titan characters that can be found below were generated in Artbreeder. It’s basically an AI website that generates new unique art by cross-breeding all kinds of art previously uploaded into it. Sounds complicated, but in reality, it’s very simple.

Without further ado, let’s take a look at some of the best real-life portraits of Attack on Titan characters!

Check Out the Latest AOT Products on Amazon!

1. Levi Ackerman

  • Created by arminalert638

2. Mikasa Ackerman

  • created by letsycakes

3. Eren Jaeger

  • created by erenjaegerishot

4. Erwin Smith

  • created by roniyuliyanto

5. Armin Arlelt

  • created by puffpunks

6. Jean Kirstein

  • created by serpahic.creed

7. Sasha Blouse

  • created by snoopy

8. Connie Springer

  • created by maymtt2002

9. Krista Lenz

  • created by yennaerys

10. Ymir

  • created by sumerwithin2

11. Annie Leonhart

  • created by seraphic.creed

12. Reiner Braun

  • created by seraphic.creed
Share this:


Hi, my name is Mia and I'm the founder of 9 Tailed Kitsune. I love anime, games, and esports (I am also a bit weird). My passion for anime sparked when I was about 7 (Phantom Thief Jeanne, hello!) and has never stopped.

3 thoughts on “Artists Use AI to Make AOT Characters Look Real

  • December 16, 2020 at 12:08 am

    Levi looks distantly like Robert P. In all they’re very good AI art. Anymore realistic and they might as well be real people

    • December 18, 2020 at 12:29 pm

      Ikr? I was stunned that something like this exists!

  • June 5, 2021 at 11:10 pm

    Some of them arent tuned well, you cant rely on the AI alone to generate a good result on the first try without lots of adjustment, since the model has internal biases. If you feed it a photoreal portrait, it won’t look the same, so the control case fails.

    But you can keep doing mutations and parameter adjustments to semi-manually get good results.
    10 and 7 are examples of not enough “chaos” parameter, which converges onto similar looking faces.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.