[darkwater]

For all the people boasting about how easily they can detect it: yes, you have to deeply look at possible artifacts (especially in teeth/ears) but sometimes it's not that easy and I'm pretty sure it would fool most of the population, especially if not giving a reference, real image on the side. Photoshopped images can also be spotted easily by keen eyes, but they still do their job, which is deceiving the majority.

Edit: typos

[thom]

This is a bit like chess puzzles. When you know there’s some winning tactic, you’ll sit and look for it until you find it. But in most real positions in actual games you don’t know and sometimes have to trust your gut as to whether to spend time on the details. If you know one picture is fake you’ll find it. If it’s just a social media avatar, you’ll assume it’s a real person.

That said, even without looking deeply for weird smooshy patterns, inconsistent curves, lack of symmetry or nonsense clothing, the biggest giveaway is that most AIs are pretty bad a realistic lighting. I got most of these at a glance because it’s a very pronounced difference.

[andybak]

I've spent a lot of time playing with AI image gen and I had to think really hard about most of them. I can confidently say I would be fooled by nearly all of them if I wasn't on the lookout.

[gehsty]

As an avatar on twitter or wherever 100% would trick me, if I even clicked the image to take a closer look I wouldn’t know if it’s the compression by the social network or the image being generated…

[jcims]

Also the model is trained on faces, not backgrounds. Pretty soon we’re going to see entire 3D scenes generated and rendered photorealistically through a camera model.

[relaxing]

I don't think that's true. If they masked out the backgrounds on training, how would the model be able to synthesize a background at all?

The problem is there's too much variety in the backgrounds of the training set. They don't follow a pattern the way growing a human does.