AI Images

Some time ago, I played around with the Midjourney AI for a bit, a website that allows you to generate images from verbal commands. You can say things to it like “Imagine a jazz singer in a dark, smokey lounge.” It then generates four images, which you can make further variations on.

I would say the results were reasonably good, although it does make strange errors. It sometimes makes too many or not enough fingers. You can also see that it does strange things with the microphone, either having it float in midair or changing the shape of it to something that looks more like a shower head.

In terms of the faces, it pretty much has that down when it comes to randomly generating faces.

When it comes to facial recognition, not so much. I uploaded some photos of myself and tried to make myself appear in different smoky lounge jazz settings, etc. That didn’t work. It distorted my face shape to such a degree that I was unrecognizable (even slightly deformed in some photos). The distortion was so great that I didn’t end up even saving any of those pictures (although in hindsight, maybe I should have!)

The AI’s ability to render “generic” beautiful human faces with such accuracy does raise the question: Where has all this “facial input” come from?

I fear the answer is: from us.

We have been feeding the Internet our data and photos for years, both on social media and on private website accessible by Google, etc. And that data is now being harvested and used by AI (without our permission). Yikes indeed. More on that later.

As I was looking into all this and trying to see if I could generate a cool picture of myself singing in some kind of jazz lounge that would have been difficult if not virtually impossible to obtain for myself in real life, I discovered that there was a “plug-in” (addition to a piece of software) that you could use to “map out your face” in 3D which would then render the Midjourney generated images of yourself more accurate. Some people had had success with it apparently, according to reviews online. I decided not to pursue it in the end.

Generating these images from simple text commands takes a surprising amount of time. I know an artist who has spent hundreds if not thousands of hours refining his commands (known as “prompts”) so that he can generate the exact kind of art he wanted. It’s not simple, and it’s not for the faint of heart.

After a few hours, I decided I had spent enough time on it already.

Besides, is it really that “cool” to use AI-generated images of yourself? Would it not be more inspiring to use “real” photos, even if they are perhaps simpler and have less elaborate backgrounds?

The debate rages on.

Let me know what you think.