I Just Tried DALL-E 2, the AI that Creates Art

The feature image for this post is a piece of art I dreamed up, but I didn't create it. I suppose you could say I commissioned it, in a matter of speaking, but the artist is an AI, and the only instructions I gave it were, "An octopus wearing a Captain America costume." You might notice right away that it's not exactly the costume of the first Avenger but rather an octopus who appears to be the captain of a ship and also has red, white and blue body paint. Close enough, right? This was one of six paintings the AI called DALL-E 2 created for me.

DALL-E 2 is the brainchild (or perhaps BRAINIAC?) of OpenAI, which describes it as such:

"DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language."
"DALL·E 2 has learned the relationship between images and the text used to describe them. It uses a process called “diffusion,” which starts with a pattern of random dots and gradually alters that pattern towards an image when it recognizes specific aspects of that image."

DALL-E 2 currently has some limitations. You cannot ask it to create images of real people, mainly to protect those people from whatever monstrosity you might accidentally or intentionally have it create. The API is also currently not available, as this is still described as a research project. In other words, don't expect to use it on your website anytime soon.

Anyway, you probably want me to stop describing it and just show you the images I asked it to create. Here goes.

I asked it to create "An elephant covered in carpet playing basketball."

DALL-E 2 created these:

In my mind, I think I imagined the elephant having fur that resembled carpet, sort of like Snuffleupagus. In DALL-E's mind, however, if I can call it a mind, it envisioned more of a rug draped over the elephant, so these three and the three I'm not showing you all have that look. Still, it's definitely playing basketball.

Also, here are more octopuses dressed as "Captain America."

These were easy ones, but I was working my way up to really challenging it. I next asked for an ostrich wearing a dress, dancing in the rain.

And there you go. DALL-E knows what dancing is, and those are all indeed ostriches wearing dresses. Two even have umbrellas, something I didn't ask for.

So far so good, but things got a little weird when I asked for a starship travelling through a wormhole.

To be fair, this is something theoretical, science fiction even, so DALL-E has no real-world point of reference, and I have no idea how much the AI has been trained on fiction. Or do I? Because next I asked it to draw me an Ewok holding a lightsaber. Is that too easy?

I'm pretty sure that first one is just a dog, but hey, what are ewoks really? The bottom two are kind of freaky and little more serious than I expected. DALL-E has layers, bro.

OK, time to really put the AI to the test. Can it handle more nuanced requests? I asked it to draw me a red bike riding a white horse. How many humans can even imagine that?

Oops, I broke it. To its credit, every image has a bike and a horse, and there is some element of red and white, but none of them quite grasp what I wanted to see. The top middle one is the most intriguing because there is riding happening, and the red and white part is right, but the bike is a second object being ridden.

Finally, I just got downright mean and asked DALL-E to draw me a square circle inside a round triangle. I fully expected it to halt and catch fire.

You know, this is about as good as we would get if we asked a human to do it, so I can't really complain.

This is fun, but the practical uses for DALL-E and other AI like it (Google is building one of its own) are many. What remains to be seen is how any organization that opens up such systems to the public will regulate what goes into them. DALL-E 2 currently has a closed testing term that is invite only. I had to tell OpenAI my IT credentials and then wait for an invite.

Having said that, I have no idea how strict their acceptance is, and I have seen plenty of other content creators testing it out. Still, they are asking anyone who tries it to do so responsibly, and they've restricted adult content and violence. Will that be enough when anyone in the world can get their hands on it? Moreover, when AI becomes so advanced that it can create photo-realistic images solely based on your descriptions, will we ever be able to trust our eyes again?

So many questions, but until then I'm just going to enjoy my dinosaur wearing a tuxedo while eating a taco.

Tavis J. Hampton

Tavis J. Hampton

IT worker with a master's in Library and Information Science currently working in the healthcare industry. Passionate about Free and Open Source software.
USA