NVIDIA’s researchers came up with an impressive algorithm that’s able to generate realistic faces. Some of them are so realistic that you may have a hard time figuring out that they were computer-generated. If you’re up for a challenge, there’s now a website where you can test how many fake faces you can distinguish from real ones. It can get more difficult than you may think.
The website Which Face Is Real is has been developed by Jevin West and Carl Bergstrom at the University of Washington. It was made as a part of the Calling Bulls*it project, which helps people to “identify bulls*it, see through it, and combat it with effective analysis and argument.”
Which Face Is Real uses the images based by The StyleGAN developed by NVIDIA. It’s simple to use: when you go to the front page, you’ll see a pair of portraits. All you need to do is click on the face that you think is real. After making a choice, the real face will be framed green.
Some of the AI-generated faces are pretty easy to spot, as there are sometimes fairly obvious glitches. However, there are some faces that look so realistic, that you will take ages to make the right guess. Or you won’t even make it at all. There’s no scoreboard or anything like it, you can play as much as you like and you can keep the score for yourself. I tried 20 pairs, and I made two mistakes. For five or six pairs I really had to focus to choose a real face.
Making computer-generated faces this realistic could pose some problems. Some of them that pop to mind are: would this make fake news more believable? Also, would this make it easier to create fake identities and misuse them in various ways? And finally, would fake portraits mean less work for stock models?
What do you think of all this? And how many fake faces were you able to guess?
FIND THIS INTERESTING? SHARE IT WITH YOUR FRIENDS!