Battle of The AI Art Engines: Midjourney vs. DALL-E

August 21, 2022

dalle-midjourney

Created by Midjourney

We’ve all been marveling at DALL-E 2 for the last few months. For good reason. It’s stunning.

Well, now there are multiple competitors. I just spent some time messing with the latest one that has a ton of hype, called Midjourney.

Midjourney is macOS and DALL-E is the Linux CLI.

The obvious question for both beginners and experts is, “What’s the difference?”

The easiest way to answer that is to say that Midjourney is like using a Mac. And DALL-E is like using the Linux command line. This is true in three main dimensions:

  1. Ease of Use: Midjourney does better with simpler prompts.

  2. Weaker But Safer: Midjourney is less powerful, but does better if you are lazy.

  3. A Bias Towards Beauty: Midjourney produces better-looking output with very little effort

midjourney-showcase

Some samples from the Midjourney Showcase

Examples

I just wrote another post for the site about censorship vs. company autonomy, and I wanted to use Midjourney to make the blog header image. I used the prompt:

An art piece showing the violence between internet censorship vs. free speech.

Here’s what DALL-E gave me:

dalle-censorship

Interesting but not immediately compelling

And after only a few minutes from research to creating, I got this from Midjourney:

midjourney-censorship

Midjourney’s attempts at the same prompt

I have tried some other queries where I went from basic to extremely nuanced DALL-E queries (vs. Midjourney), and I was able to get DALL-E to win. But it took a lot of time. A lot of trial and error. Midjourney gets to something very cool with very little prompt engineering.

It’s like DALL-E was trained off of all art, and Midjourney was trained off of Deviant Art.

It’s almost like Midjourney was trained off of images that people already love, whereas DALL-E can generate much more, but with a correspondingly larger Margin for Meh. To say that another way, I’ve made a million DALL-E images that were total throwaway misses. With Midjourney, almost everything you make is at least a little bit cool.

Interface differences

The other big difference is that DALL-E is run from a full website or via API. Midjourney on the other hand is run via Discord bot. It’s an interesting choice, and likely a brilliant one. It’s basically taken over the mindshare within the gaming world very quickly, with so many people there being 1) inclined towards pretty looking DeviantArt-like images, 2) not super interested in reading a 75-page prompt engineering book, and 3) already highly familiar with Discord.

Bottom line: you have to go play with this thing. So:

  1. If you’re a hardcore AI and/or Art person, you should be messing with DALL-E first and Midjourney second.

  2. If you’re light on the AI stuff and more interested in gaming or NFTs or just having cool art for various purposes, start with Midjourney.