This AI Makes Celebrities Old…For a Price! 👵

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today we are going to take a bunch of celebrities,   and imagine what they looked as tiny little  babies. And then, we will also make them old. And   at the end of this video, I’ll also step  up to the plate, and become a baby myself. So, what is this black magic here? Well, what  you see here is a bunch of synthetic humans,   created by a learning-based technique called  StyleGAN3, which appeared this year, in June   2021. It is a neural network-based learning  algorithm that is capable of synthesizing these   eye-poppingly detailed images of human beings  that don’t even exist, and even animate them. Now, how does it do all this black magic? Well,  it takes walks in a latent space. What is that?    A latent space is a made-up place  where we are trying to organize data   in a way that similar things are close to each  other. In our earlier work, we were looking to   generate hundreds of variants of a material model  to populate this scene. In this latent space,   we can concoct all of these really  cool digital material models.    A link to this work is available  in the video description.    StyleGAN uses walks in a similar latent space  to create these human faces and animate them. And now, hold on to your papers, because a  latent space can represent not only materials   or the head movement and smiles for people, but  even better, age too. You remember these amazing   transformations from the intro. So, how does it  do that? Well, similarly to the font and material   examples, we can embed the source image into a  latent space, and take a path therein. It looks   like this. Please remember this embedding step,  because we are going to refer to it in a moment. And now comes the twist, the latent space for  this new method is built such that when we   take these walks, it disentangles age from other  attributes. This means that only the age changes,   and nothing else changes. This is very  challenging to pull off, because normally,   when we change our location in the latent space,   not just one thing changes, everything  changes. This was the case with the materials. But not with this method, which can  take photos of well-known celebrities,   and make them look younger or older. I  kind of want to do this myself too. So,   you know what? Now, it’s my turn. This is what  baby Károly might look like after reading baby   papers. And this is old man Károly, complaining  that papers were way better back in his day. And this is supposedly baby Károly from a talk   a NATO conference. Look, apparently  they let anybody in these days! Now, this is all well and good, but there  is a price to be paid for all this. So   what is the price? Let’s find out together  what that is. Here is the reference image of   me. And here is how the transformations came  out. Did you find the issue? Well, the issue   is that I don’t really look like this. Not only  because this beard was synthesized onto my face   by an earlier AI, but really, I can’t really find  my exact image in this. Take another look. This   is what the input image looked like. Can you  find it in the outputs somewhere? Not really. Same with the conference image, this  is the actual original image of me,   and this is the output of the AI.   So, I can’t find myself. Why is that? Now, you remember I mentioned earlier that we  embed the source image into the latent space.    And this step, is, unfortunately, imperfect.   We start out from not exactly the same image,   but only something similar to it. This is the  price to be paid for these amazing results,   and with that, please remember to  invoke the First Law of Papers.    Which says, do not look at where we are, look at  where we will be two more papers down the line. Now, even better news! As of the writing  of this episode, you can try it yourself!    Now, be warned that our club  of Fellow Scholars is growing   rapidly and you all are always so curious that  we usually go over and crash these websites   upon the publishing of these episodes.   If that happens, please be patient! Otherwise, if you tried it, please let  me know in the comments how it went   or just tweet at me. I’d love to see some  more baby scholars. What a time to be alive!

Thanks for watching and for your generous  support, and I'll see you next time!