Just another AI test. I gave Gencraft.com the names of some famous people. I took the better of the two pictures generated for each name. Then I matched them to actual pictures. I tried to avoid iconic real images, where possible.
In most cases, I reduce the resolution and color depth of the AI image to make it more closely match the real image. Turns out, the best clue for real-versus-fake is that the fake images — done with modern tech — have much better resolution that the real ones. That alone makes them easy to spot.
Aside from one complete fail, Gencraft managed to produce some reasonable approximation for every name I entered.
So, which is real, and which is AI? Note that I put no effort into getting AI to generate a realistic picture of the individuals in question.
Answers given below.
Answers below.
The real picture in each pair:
- Marilyn Monroe — right
- Indira Gandhi — left
- Margaret Thatcher — left
- John F Kennedy — right
- Lyndon Johnson — right
- Josef Stalin — right
- Albert Einstein — left
- Mohandas Gandhi — right
Addendum
And the one fail? Golda Meir. Gencraft.com apparently didn’t know who she was. Below is the better of the two images I got back. Gold, yes, Meir, no.
The final oddity is that biopics have already blurred the line between real and imaginary images of famous people. In picking the real image, I had to be sure to steer clear of actors portraying the individual.
In particular, Margaret Thatcher was tough, because Meryl Streep, in makeup, is a dead ringer for the real Margaret Thatcher. Below, Streep is left, Thatcher is right. I think.
This last point makes me wonder about the extent to which the AI is able to separate genuine images of famous people, from Hollywood images of the same individuals. And, in the longer run, the extent to which AI-generate images of the famous will eventually become self-referential. Will a newly-generated AI image of Thatcher be based on not just the real Thatcher, or the Hollywood Thatcher, but in addition, on other AI renderings of Thatcher.