A guy named Deepfaces
It all started last November on the Reddit forum with a man named Deepfakes. A guy among many who likes to watch porn, but his creative side pushes him a little further. He designs porn montages with celebrities.
So far nothing new, these photoshop montages porn actresses replaced by faces of celebrities exist for years and have almost no interest at the time of web 2.0.
“I just found a clever way to do face-swap.”
The novelty is that here we are talking about video and the result is simply amazing.
Through the deep learning technique, he manages to transfer the face of Gal Gadot or Taylor Swift to the porn scene of his choice.
And for the past few weeks this technique has been exploding.
Deep reality
Deep learning is still in its beginnings, but innovations in the field are impressive and even disturbing.
We remember google's DeepDream program, which uses a network of artificial neurons to find and reinforce structures in images using recognizable shapes.
Simply put, it is an algorithm inspired by the visual cortex of animals whose vision is a mixture of what the eye sees and what the brain interprets through the images it has recorded.
It was originally used to make recommendations through image and video recognition
But the scariest part is yet to come.
A few months ago a team of researchers at the University of Washington was able to generate a perfect fake Obama.
Their challenge was to faithfully reproduce the movements of a subject's lips from a simple sound file.
The researchers relied on 14 hours of videos of speeches to form a neural network, which was used so that this mouth could mimic a sound.
Research of this type is flourishing all over the world and many examples can be found on the web, sometimes created from a simple webcam.
But the most impressive research is undoubtedly coming from Nvidia, who opened a path to the end of reality as we know it.
After being able to generate false faces from celebrities.
Nvidia goes even further, creating artificial intelligence that can simply recreate reality.
A tool that acts as a sort of filter by being able to transform
for example a landscape from day to night.
Impressive, isn't it?
Well, that's the type of algorithm Deepfakes uses.
In this video, Nvidia's algorithm transforms the content of the image, transforming a snowy landscape into a summer day.
Say hello to Fakeapp
“With hundreds of face images, I can easily generate millions of distorted images to train the network,” he said. “After that if I feed the network someone else's face, the network will think it's just another distorted image and try to make it look like the training face.
We were all more or less impressed by the incrustations of old characters in the new star wars. Incrustations that cost millions of dollars, and thousands of hours of work.
Here's what a 20min Reddit version looks like
The deepfake community is growing day by day, as well as initiatives to make the creation of this type of content more accessible.
As with the creation of the Fake application, a simple program allowing the creation of deepfakes in an automated way.
Today anyone can easily create their own Deepfake or Fakeporn.
All you need is a computer with a good quality graphics card and lots of face photos.
And here again, there is no need to look far away.
Photo banks are made available to the Redditor,
Scarlett Johannsson and Nicolas cage being in the lead.
Although the results are not always convincing, users remain very active and do not hesitate to help each other, and I invite you to watch if you get the urge.
Black Mirror
Video is now at the heart of the internet, and many major applications and sites have understood this well by directing their platform towards this format. This is the case of Facebook, instituted or Snapchat where videos flood news feeds.
The flow of information has never been so rapid and numerous, making it increasingly difficult to verify information.
We live it every day with the fakes news that aim to manipulate public opinion for political purposes. And even if this information can be easily demonstrated a few days later, it is often too late and the accumulation of this information can easily become truths.
Fake apps then pose a real problem.
A problem that is not new, false images have always circulated,
whether it is in painting to tell the hardly exaggerated exploits of some kings, or the censorship of official photographs in the Soviet Union.
But where we can doubt an image, can we doubt a video? Especially when it was filmed with a simple phone?
This is my first post, don't hesitate to follow me and comment.
A big thank you to 
.
.