Among all the real-life studies on neural networks, machine learning, and artificial intelligence being conducted, there are many experiments… that lie on the thin line between fascinating and terrifying.
Image processing has recently emerged as a fantastic tool of artificial neural networks, partly due to decades of people sharing photos and selfies online. As a result, we have a massive repository of face images to “harvest” and utilize, training AI to do everything from simulating the aging process of users in mobile apps to creating surreal collections of faces of people who don’t even exist.
The stock photo industry will never be the same again, but Mario Klingemann wondered what would happen if he asked those artificial neural networks to generate virtual face photographs synchronized with music – and as you can see in the video below, we have some truly impressive faces created as the music thumps!
Klingemann used the StyleGAN2 generative adversarial network developed by Nvidia, which was released as open source over a year ago. He didn’t conduct the custom training process himself but instead fine-tuned the GAN to vary the generated results based on the audio spectrum of an input sound file – in this case, the song “Triggernometry” by Kraftamt.
Some of Klingemann’s Twitter followers suggested that he should slow down the GAN-generated video a bit to see the horrifying things hidden within. You can see the images below and immediately feel the horror. Be warned not to scroll down to see more if you have a history of heart conditions or are reading this at midnight!
*** Final warning! ***
Source: Gizmodo