Have I met you ?!
Social networks are great tools to rekindle with people from your past, aren’t they?
You can check what Suzan has been doing since elementary school, or where you last saw John.
Speaking of John, his last picture is with a bunch of friends you know, but you can’t seem to recognize one of them. You look at their profile picture, still totally clueless about that girl:
I know why though.
She’s not actually real.
Never stepped even a single foot on earth.
This image was computer generated, with the famous artificial intelligence thing you keep hearing about nowadays (GANs to be more precise).
Basically, portraits ALL have similar poses and similar features (having a human face is a good start!). That makes it SO much easier for computers to create this specific type of pictures, because they vary millions of time less that other type of pictures. Drawing a human face can get easier with practice, if I were then to ask you to draw the Chateau de Versailles well you’d struggle because it’s really different to draw. Hence if you let the program do its thing, focusing on less than 100 things to look at (color of eyes, distance between the eyebrows, the teeth-nose distance….) then it will get quite good quite fast. Especially as it is easy to feed it millions of portraits to draw inspiration from (pun intended).
Now to briefly explain how GANs work:
GAN stands for Generative Adversarial Network, and it’s essentially two people fighting each other in a competitive game of deception.
One of these guys is called the generator. So he’s the one actually generating the images we’re speaking of. But like in life in general, you need to find a way to become better at what you do. So how does that happen?
Well there’s another guy called the discriminator. He’s the one judging what is being generated by the first guy. The twist is that what’s being given to the discriminator is not only the stuff the generator is creating. That guy also received 100% real images and he needs to guess which ones are the fakes (the generator’s).
So it’s like a game of hide and seek where the generator is getting better at tricking the discriminator. At first he’s on such a bad losing streak but at the end of the day it’s the discriminator who rehashes all their life’s choices because well he’s getting tricked nearly all the time!
And it could be anything that can be generated: pictures, audio, fake ids… (don’t try this at home please).
Now look at what the GAN we’re focusing on today did!
Website is here! (just refresh the page for a new one)
And I hope you’re scared of Tinder bots now.
¯\_(ツ)_/¯
And meanwhile this looks fun, think of the damage that can be done with that sort of technology. But enough with depressing stuff, go to the bonus for something to put a smile on your face.
BONUS
The website also fakes HORSES (a bit glitchy to be honest), and CUTE CATS, how crazy is that?
The cool thing about this article, is that I didn’t even need to ask for John’s friend’s authorization to be on my blog!
(but I do wish to give credits to the website owner Phil Wang, you can help support his work here)
