This Face Does Not Exist

Every refresh gives you a different AI-Generated face.

This isn’t scary at all. It strikes me, though, that they still can’t do realistic artificial voices, so at least voice actors will still have jobs. I also wonder how well they can do facial expressions.

BTW, I was at the Space Transportation Conference this week in DC, and didn’t have much time to blog. Just got home last night, in time for Valentine’s Day dinner (which we iron-chefed from stuff in the freezer and fridge: Steak and shrimp, with a nice merlot, and some Godiva that I bought at DFW).

[Update a few minutes later]

[Update a while later]

OK, so they’re not perfect.

[Update a few minutes later]

OK, this is nightmare fuel.

2 thoughts on “This Face Does Not Exist”

  1. There’s only about 70 or so eigenvectors (or eigenfaces) to describe any face. That’s why we can find identical-looking people who are not related in any way.

    I myself have at least one doppelganger that I know of – Burkard Polster. You can find him on YouTube on the Mathologer channel, and me on the robotguy channel. He’s a couple of years older than me but we could easily be mistaken for twins.

  2. I’ve looked at the page a number of times, and I have to say, the uncanny valley effect is in force there. Having felt under the weather just a tad the past couple of days made me really sensitive to it. Some of the images made me just a bit queasy.

    In a way, it’s like the AI’s lack of detail (that we see immediately) means an AI definitely doesn’t value we humans as much as we value ourselves. The human facial image is regarded just as the most efficient set of factors to define it, and no more. What other minimum, efficiency-minded value judgements will AI make in the future as the technology advances?

Comments are closed.