This New Technology Uses Artificial Intelligence To Create Portraits Of People That Don't Exist
A while ago, we featured an amazing new technology that was able to remove grain and watermarks from pictures, created by Nvidia. This time the technology giant is back with something even more amazing - a technology that is able to generate portraits of people that don't actually exist using artificial intelligence.
A while ago, we featured an amazing new technology that was able to remove grain and watermarks from pictures, created by Nvidia. This time the technology giant is back with something even more amazing - a technology that is able to generate portraits of people that don't actually exist using artificial intelligence (AI).
Image credits: Tero Karras, Samuli Laine, Timo Aila
The first use of this AI tool, called generative adversarial network (GAN), dates back to 2014. Back then the generated faces were grainy and colorless but after four years we finally have portraits that resemble real photographs.
It took researchers 8 weeks to train the AI to learn the various 'styles' of faces and produce unique 'people'.
Here are some faces that the “Style-Based Generator Architecture for GANs” created
Turns out this technology is capable of creating more than just faces - it is even able to generate 'fake' interiors and vehicles.
"These datasets were trained using the setup for the duration of 70M images for 'bedroom' and 'cats', and 46M for 'cars'"
Image credits: Tero Karras, Samuli Laine, Timo Aila
The first use of this AI tool, called generative adversarial network (GAN), dates back to 2014. Back then the generated faces were grainy and colorless but after four years we finally have portraits that resemble real photographs.
It took researchers 8 weeks to train the AI to learn the various 'styles' of faces and produce unique 'people'.
Here are some faces that the “Style-Based Generator Architecture for GANs” created
Turns out this technology is capable of creating more than just faces - it is even able to generate 'fake' interiors and vehicles.
"These datasets were trained using the setup for the duration of 70M images for 'bedroom' and 'cats', and 46M for 'cars'"
See how this technology works in the video below
h/t: Bored PandaDE
Written by
Demilked Editorial
SaaS Review Specialist · Demilked
With 5+ years in the creator, entertainment, and publishing spaces, Demilked shortlists, reviews, and ranks leading tools that actually make your life easier.