Hands...touching...hands...reaching out...

creeping me out...

One of the overarching themes of my AI art experiments so far has been the quality, and lack thereof, of the hand images generated. I find this very interesting because when I was studying drawing and painting, a friend of mine who was much more talented than I told me about how hard it can be to draw and paint hands.

At the time it was profound because I realized that I had trouble with hands but never realized that they were actually hard to draw.

Fast forward to 2022, and I find it completely fascinating that artificial intelligence has the same limitation, only it manifests itself in a different way.

To learn more about AI’s limitations in generating original images of hands I decided to test some specific and easy-to-interpret prompts in hopes of some clear results.

My first prompt was: “A person with exactly five fingers on each hand.”

Fascinating. The image clearly shows two hands, so it gets that there are two hands. But there are three thumbs, the hands are merged together, and there are less than ten total fingers. So while this is a fail to accurately generate an image that matches the prompt, the resulting image is nonetheless intriguing.

I thought that maybe being less specific and actually simplifying the prompt would be the way to go.

The next prompt: “human hands.”

Here, we have a decently rendered pair of human hands that are reasonably realistic. But the top three hands!? Whoa!

From left to right we have a two-thumb hand, an anemone-like six-finger hand, and a five-finger hand that is connected to a double arm. Fascinating.

I like to imagine that the AI is a human going through thoughts that lead to this. It makes my head hurt a bit, but is also kinda funny.

After these two mixed results, I decided to add an action to my prompt and give fewer details about the hands.

Next prompt: “High five”

The hands aren’t so bad, but still an utter fail. How is this even a high five? It’s like somebody who doesn’t know what a high five is, and doesn’t have command of the language in which a high five was explained to them, then tried to draw a high five. Next.

“Holding hands.”

This is the last prompt for today. I thought that ‘holding hands’ was easier to interpret than high five.

Here again, we have a strange number of strangely arranged fingers. I guess if there is any conclusion I have reached, it’s that drawing hands is hard. Even though I thought a computer would be better than a person at something technical, I might have been wrong…at least from now.

I remember reading that water and hair were big hurdles in early computer animation. But, as we can see in current computer animated movies, that hurdle was passed pretty clearly. It should be only a matter of time until this hand hurdle has also been passed.

Tomorrow, I will feature some more hand experiments. Thanks as always for reading.