My name is Mike Solomon, and I learned a thing or two about AI-generated faces when I built the web app “Judge Fake People” last year. Here’s a little guide on how to spot fake photos of fake people online.
All the people shown below are fake.
Weird backgrounds are a dead giveaway. Some are easier to spot than others.
Look out for a “uni-tooth”. It’s like a unibrow but as a front tooth. Some people have them in real life — Tom Cruise used to before he got braces — but they’re rare.
Mismatched, weird earrings are another dead giveaway (unless they become a thing).
Also, look for weird visual artifacts. The neural networks that make these images are often trained on photos/videos of people giving talks and speeches, so look for artifacts near where microphones would be.
Back to the earrings; the mismatched earrings are caused by the AI’s inability to recognize the importance of symmetry. So look for other obvious signs of unsymmetrical things — usually they’re easiest to spot in earlobes and glasses. (The latest AI’s have probably solved this by now, so your mileage may vary).
Finally, these next ones are pretty tricky to identify as fake. Be aware of photos where you only see 1 or 0 ears. But here’s the thing — as of now (2020) most fake people will all be positioned similarly in the frame; square in the center.
So why did I write this article? I got followed by this account and I genuinely can’t tell if it’s a real person or not:
What do you think?
In the arms race for public opinion and attention, hopefully we can be well armed.
About the author: Mike Solomon works on product strategy at Hearst Magazines and is the creator of Judge Fake People, a web app for ranking the attractiveness of AI-generated faces. The opinions expressed in this article are solely those of the author. You can find more of Solomon’s work and writing on his website and Twitter.
#abstract #art #abstractart #abstracters_anonymous #abstract_buff #abstraction #instagood #creative #artsy #beautiful #photooftheday #abstracto #stayabstract #instaabstract
https://ift.tt/32ySUso
0 Comments