Artificial Intelligence Creates Portraits Of People Who Don’t Exist And It’s Creepy

Who said robots couldn’t be creative? artificial intelligence (AI) is no longer the stuff of science fiction films and now this technology is tapping into the art world. In a new paper published by NVIDIA researchers, they released portraits completely generated by artificial intelligence (GAN) but with a twist – none of the people in the images actually exist.

Image credits: Tero Karras, Samuli Laine, Timo Aila

According to The Verge GAN was introduced in 2014. The portraits at the time were semi life-like, grainy and in black and white. Jump forward to the portraits today and the resemblance to a real person is so accurate that most people can’t tell the difference.

The current researchers were able to use their AI to copy the ‘styles’ of various face sources then create blends of these originals to create completely new “people.”It took 8 weeks for the researchers to train the AI, using powerful GPUs.

Faces generated by “A Style-Based Generator Architecture for GANs”

But this revolutionary technology doesn’t stop at faces, in 2014 75% of IKEA’s catalogue photos were computer generated. Just like with any technological advance there could be negative repercussion. With the ability to create faces, furniture, cars and real estate it calls into question the future for stock photographers and where their place will be in this new digitized world.

“These datasets were trained using the setup for the duration of 70M images for ‘bedroom’ and ‘cats’, and 46M for ‘cars’”

Watch how AI has progressed in this 6- minute video

Share

Leave a Reply

Your email address will not be published. Required fields are marked *

By continuing to use the site, you agree to the use of cookies. more information

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.

Close