DALL-E Mini Who? It’s All About Midjourney Now

An image created by Midjourney of two shapes that look a bit like mobile sim cards containing multiple squares of different colours

A couple of months back, I wrote a blog on DALL-E MIni, the mini open-access version of OpenAI’s artificial intelligence text-to-image generator DALL-E. The week I put the piece together, the system was taking the internet by storm thanks to its capacity for creating original but often hilariously warped artworks. Here are my two attempts, based on the daft prompts “Mr Bean wearing a wedding dress while riding a fish” and “bald bear chip shop hat wiggle”:

Multiple warped cartoon-like images of Mr Bean wearing a wedding dress riding a fish made on DALL-E Mini
Multiple images of caps with warped pictures of bears on the front made on DALL-E Mini

Then I got busy — as I tend to do — and took weeks to edit the article. The internet — as it tends to do — quickly moved on, and I missed my window to post. Such is digital life.

A month later, along came another tool, which confidently instructed DALL-E Mini, “Hold my beer, mate”, as it rolled up its sleeves and showed us what we’ve been missing. Introducing Midjourney, an AI-powered art generator developed by an independent research lab focused on exploring new mediums of thought and expanding the imaginative powers of the human species.

Midjourney can only be accessed with an account on Discord, a social app that allows people to communicate with others through text chats or voice/video calls in communities called servers. Once registered, users can join the Midjourney server as a beta tester, giving them a limited free trial of around 25 prompts before membership fees apply (min $10 per month).

You only need to scroll through the ‘highlights’ and ‘community showcase’ sections for a few seconds to realise those fees would totally be worth every penny to anyone working in the creative industries. I was instantly blown away by creepy scenes of brutalist architecture beneath swirling blood-red skies and of Jesus standing among crouching American soldiers as fighter jets fly by just feet above.

I could have looked at the weird and wonderful creations of other users all day, but I was keen to make the most of my freebies. I followed the instructions laid out in the ‘getting started’ tab to the left, which saw me joining one of the newcomer rooms and typing “/imagine” before each prompt.

For a direct comparison with DALL-E Mini, I kicked off with the same ridiculous prompts I mentioned earlier, and Midjourney gave me four different options for each within seconds. Here they are:

Four abstract images of Mr Bean wearing a wedding dress riding a fish made on Midjourney
Four abstract images, two of bears wearing hats, and the other two of hats with some element of a bear on them, made on Midjourney

As you’ll no doubt notice, the images are considerably better than DALL-E Mini’s — richer in detail, less distorted, and more varied. That said, they’re still abstract and nowhere near as jaw-dropping as I’d expected after seeing what the server made for other people. I chalked this up to the silliness of my prompts and decided to go back into both Midjourney and DALL-E Mini with something a little less absurd but still a lot of fun: “Star Wars droids in the style of Picasso”. Let’s look at Midjourney’s efforts first:

Four images of Star Wars droids in the style of Picasso made on Midjourney

Now we’re talking. Probably down to a much more specific prompt. And what about DALL-E Mini’s designs?

Multiple cartoon-like images of Star Wars droids supposedly in the style of Picasso (but not really) made on DALL-E Mini

Meh, I see the path it’s taken, but not loving those as much. I have to say, it’s pretty hard to know with these text-to-image generators how much and what sort of detail to include in prompts. I wish there were a little more guidance on the issue. But based on the prompts of other users on Midjourney, as far as I can tell, fewer or amateur search terms get you naff, cartoony images, while plentiful or sophisticated search terms result in sharper, more lifelike artworks.

The addition of such terms on both systems completely transformed my creations, but Midjourney won every time — that is, until I ran out of free credits. Here’s just one example, where I added the word “photorealistic” to “Star Wars droids in the style of Picasso” on Midjourney and then DALL-E Mini:

Four photorealistic images of Star Wars Droids supposedly in the style of Picasso (but more like totem faces) made on Midjourney
Multiple images of Star Wars droids supposedly in the style of Picasso (but more like Star Wars toys) made on DALL-E Mini

Altogether different results, with Midjourney’s selection looking more like totem faces and DALL-E Mini’s closer to Star Wars toys. Interesting, huh?

DALL-E Mini has been great for a laugh, for sure, but Midjourney produces pictures that people might actually want to use in their work. Take the main image at the top of this blog. I’d normally source something from a site like Unsplash or Pixabay, but I created that one using Midjourney. I love a bit of the meta, so I asked the server to deliver “Midjourney artificial intelligence text-to-image generator”. I have no idea what I’m seeing up there, but it’s beautiful and I’d certainly hang it in my living room.

Have you had a chance to try out either system? If so, let me know what you think in the comments.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Catriona Campbell

Catriona Campbell

Behavioural psychologist; AI-quisitive; EY UK&I Client Technology & Innovation Officer. Views my own & don't represent EY’s position. catrionacampbell.com