Midjourney is a Trip

Of the several AI-powered systems that can create images from text prompts, MidJourney is the most easily accessible one right now. I’ve had some fun playing with it,

Midjourney produces images from prompts, which are basically text descriptions. They don’t have to be specific or concrete in any way. “Astronaut riding a horse in space” works just as well as abstract words. In fact, the latter can sometimes produce more interesting images.

This post might not work very well in a feed reader, so maybe head to the web version. There, you can click on the little galleries to get much larger versions of the images.

Of course I had to try it on a variety of variations of “eager eyes,” and it did not disappoint. Some of the prompts will seem a little random, but they’re usually the result of a few attempts while adding and removing words.

It’s kind of a fascinating setup. You interact with the MidJourney Bot through Discord (if you’re not familiar, it’s like a less refined version of Slack). This is especially addictive when you have the Discord app on your phone, because you can just send it prompts whenever the fancy strikes you, and it’ll ping to let you know when your images are ready.

Other systems like DALL-E produce more photorealistic images, which in my experience MidJourney does not do very well at. But what really nails are styles like art nouveau, art deco, and 1960s futurism. The latter is particularly fun.

Clearly the most interesting use is to mix concepts and styles that are nonsensical or that would not have existed. What would harvest robots in ancient Egypt have looked like? How about musical robots during the height of art nouveau?

It’s also funny to realize that while it can create images of animals that look good at first glance, it doesn’t have a clue how animals work. You quickly realize that details don’t make sense, there are extra legs, etc. But it does work for surrealist-style images at times.

It can also create some very surprising images, like Picasso-style paintings of UFOs. I was quite surprised by how well it did these, not just in terms of them looking like paintings (some even with frames!), but capturing this playful style.

Abstract terms can lead to some interesting surprises too, like the prompt “tonality.” I have no idea where it is finding these shapes, but they

I also tried a few prompts I saw people use with DALL-E on Twitter, like “Pharaoh Darth Vader of Egypt.” Midjourney did a better job here than DALL-E in my humble opinion, but I also can’t find the DALL-E tweet anymore to show them side by side.

This is just a selection of what I’ve created with MidJourney so far. If there’s any interest, I can put together a second post with some more examples. If you can find an invite, it’s worth checking out for sure.