DALL·E vs. Midjourney: How AI Image Generators See Human Concepts Much Differently

A side-by-side comparison of results from two popular platforms

If you’ve been reading my work (thanks if you have!), you’ll know I’ve plunged face-first into the world of AI imaging recently with DALL·E from OpenAI. Since then I have also discovered Midjourney, another ambitious project that produces images from simple text prompts in Discord. The results from both platforms are truly incredible, and I have already spent money on both after my free trials ended.

While the trained neural networks can produce incredible images that would take a human artist hours or days to complete, the “art” of text prompting is learning how to use the right words in the right order to bring your original ideas to life. Generally, the more descriptive and specific the text you enter, the closer it may be to what’s in your mind’s eye. It takes some practice (I’m still a rookie), but when you nail a concept, it’s very exciting.

Since I’m now on board with both DALL·E and Midjourney, I thought I’d try a head-to-head experiment. Not to see which one is better — that’s a hard call so far — but to see how each would interpret single-word prompts. Below are some samples I generated from both, showing how both labs envision these human concepts (without me influencing results by adding descriptions.)

Keep in mind that each lab generates four results from a single text prompt. I chose the most compelling images among the results from both to feature, and then stitched them together for easier comparison.

Here they are, in no particular order. DALL·E images are on the left, Midjourney generations are on the right:

“Women”

It seems Midjourney has a “type” — I’ve noticed when generating pictures of women, it tends to default to a dark-haired beauty with striking eyes and lips. I do love what she’s wearing, seemingly pulled straight out of an inter-dimensional closet. I also like the look on her face, like she’s thinking about what to have for dinner that evening.

Meanwhile, DALL·E generated a woman of colour in its results. Actually, all four of the women in this particular sample appear to be non-Caucasian, which is interesting.

“War”

I’m not 100% sure what DALL·E is trying to show me based on this one-word prompt. At first I thought it was a war hammer, but it’s more likely some kind of armoured gun with a huge turret.

Meanwhile, the Midjourney version is entirely different. It shows the bleakness of war in a more human context. However, the biggest surprise about this version is that there seems to be a heart hidden in the fire — is that a subtle message to love one another? Or a happy accident? Either way, I like it.

“Money”

Midjourney went with a varied approach to this word. You can see how it features coins, but also what appears to be credit/debit cards. The stacks/buildings in the foreground and background suggest it’s pointing to industrial production in relation to wealth.

DALL·E chose a more simple approach, presenting paper bills that are quite convincing, and resemble existing currency in some countries. I like both versions equally.

“Men”

Now this one was a real eye-opener because of the wide divide in interpretations. As you can see, DALL·E’s version of a man looks like a dude from a 1990’s band (ahem.) Poor guy may be thinning a little on top though, judging by his bangs.

Midjourney definitely has its preconceived notions of men. With no other prompts, it made him into a scary, almost demon-like being with what looks like fire in his eyes. Either that, or this dude needs to see a doctor very soon.

“Love”

Again, these results really threw me. On the right, we have a nice, “safe” interpretation of love — a pretty young woman with a heart in her hair. I’m not sure how it’s staying in there, though — perhaps she used glue?

The other side is more open for human interpretation. The powdery white substance surrounding the finger-drawn words could be a number of things — but I’m going with flour on a cutting board.

“God”

Here we have another wildly different interpretation from both AI labs. Midjourney (on the right) has produced what looks like a Caucasian God, with a straggly beard. He looks a bit worn out and sad, and you can’t blame him with the current state of the world.

Meanwhile, DALL·E surprised me by choosing Ganesha, a symbol of Hinduism. I like this version, as the elephant-like god appears to be chilling out in a tree. Ganesha also appears to be meditating and gesturing at the same time with his two right hands. The two left hands seem to be telling us to stop, while also waving a sharpened lollipop. (Ganesha lore includes him having four arms, so bravo to DALL·E for attention to details on this one.)

“Death”

I chose very different versions from the two labs for contrast. DALL·E seems to have created some kind of charm or hood ornament in the shape of a skull. I’m not sure why it took the shape it did, unless it’s a supposed to represent a cross? Your guess is probably better than mine on this.

Midjourney went with a more traditional view of death, and I think it’s a masterpiece. The treatment of the “brush” strokes, the web in the eye sockets, the fading flowers — all of it works so well together. I would almost print and hang this on my wall.

“Capitalism”

I’m not exactly sure DALL·E understood the assignment. It didn’t produce a scene representing capitalism, but rather randomly placed words. This generation featured a face of a man (named Midall?) who looks like he has a lot of money. Or, at least his suit implies he will do what it takes to make it.

Again, Midjourney wowed me with its interpretation, choose a magazine illustration style to depict capitalism. I like how the man is standing tall against the skyscrapers, with the red stuff coming up from the earth. Is that hellfire? Fruit juice? The overwhelming stress to make money so we can stay alive to make more money?

It’s anyone’s guess.

Give AI a try

Both DALL·E vs. Midjourney are mind-blowing tools. Once you start using them, it’s hard to stop. The possibilities really are bound only by your imagination.

I’ve already started using these machine generated images for editorial purposes, and I have started an AI art account on Instagram if you want to follow my artificial creations.

Thanks for reading! Share your thoughts, or post links to your own AI creations!