I found inspiration for using Midjourney during my last project, an AI discovery for a client in the UK. We ran a generative AI ideation workshop for business stakeholders. To make AI concepts easy to understand, we simplified them into a limited number of patterns.
My original idea was to create a deck of cards for these patterns to be used in the client workshop as a stimulus. Unfortunately I didn't find time to do this during the project but I resolved to revisit as a creative exercise for trying out Midjourney.
Creating card decks for workshops is something I am quite familiar with. I've created several buy-a-feature card decks for past client projects (not to mention several personal projects including my long-running Megacities Top Trumps deck). I have a simple template in Pages that I use so I knew I could easily reuse this. I also had all the AI pattern titles and descriptions I created for the client workshop for the individual card content. The only thing missing was the visuals to make the cards more playful and interesting.
Early image test using Firefly
I had done some early experiments with creating quick images of little robots in Adobe Firefly. They were pretty good as placeholders but not really what I had in mind, aesthetically. For a printed deck of cards, I really wanted a simpler illustrative style that wasn't like a 3D render.
Prompt for inspiration
How to describe the style I wanted? I found some inspiration on Twitter from someone who was using a range of different prompts to compare the outputs from various AI tools.
Illustration of a man in a red hoodie, minimalist, graphic design poster art, dark cyan and sky - blue, honeycore
(I didn't know what honeycore was but it led me to the fascinating aesthetics wiki)
I left the aesthetic part of the prompt as it was and changed the “man in a red hoodie" for ”a red robot" and whatever concept the pattern description called for.
illustration of a red robot pointing to the distance, minimalist, graphic design poster art, dark cyan and sky - blue, honeycore
The Midjourney results were pretty good and after a few different prompts I became reasonably confident I could generate enough similar images to make them look superficially like like a suite of images for a card deck (I needed 20 in total).
Not feeling the flow
It was fast to generate the images but after a while it became quite tireseome and laborious. Half a day of entering text on screen is not a very satisfying process as you are never really in the flow of doing creative work.
I opted not to edit or change the results outside of Midjourney even though this is quite possible to do (I found the random noise in images quite annoying to have to fix although the zoom out function was very useful to create more breathing room for the central image and some accidental landscape that enhanced the concept of something artifical in an environment).
The process of editing prompts, refreshing the results, satisficing choices felt a bit like spending an afternoon as a hovering art director at the shoulder of a brilliant but soulless artworker.
Since the images are generated on the fly every time, there was also a sense of a better result being just around the corner from the next promp, or if it wasn't, would I be able to keep a track of earlier drafts? Discord is a strange tool for creating images at first although after a while the whole continuous conversational interface kind of makes sense.
Forwards with Firefly
Havig completed a set of images in Midjourney I was happy with, I thought about the tweet that had inspired me and decided to try and recreate the images using Firefly to compare the results.
The Firefly interface is more intuitive as you would expect from Adobe. Being able to select the aspect ratio made me realise I had totally overlooked this in Midjourney (it's perfectly easy to do but you have to remember to append it to your prompt using --ar 16:9).
Accidental Eric Carle
The aethetic results from Firefly were very different. Inferior I'd say but mostly just… different (accidental Eric Carle is how I thought of it if you have ever read the children's book The Hungry Caterpillar).
This makes sense when you think about how the different tools have been trained. Firefly has a much more graphic design aesthetic in keeping with the professional stock library it has been trained on.
One thing that bugged me about Firefly was that you had to download the images as jpegs not pngs. It didn't make much difference to the quality but struck me as a bit of an oversight for a company that knows the difference between photographs and illustration.
Abstract for abstractions sake
By the time I had recreated my efforts using Firefly I was actually starting to enjoy the use of the AI tools and the possibility it afforded me for fast experimentation. Over the weekend, I had a brainwave that the best visual style to represent abstract concepts would be abstract art.
I found an abstract painter whose style I liked and went back again to Midjourney to recreate the images in the style of Robert Delauney. I found this more satisfying than creating the illustrations as each result was like a new piece of art that required more interpretative judgement.
An abstract interpretation of “answering” from Midjourney in the style or Robert Delauney.
Doing the whole thing over again using a completely different style was a very interesting, reflective process that made me question the original text I had been using, making many more adjustments on the fly to get to the result I wanted. This finally felt like creative flow.
Full card deck with variations, using images from Midjourney and Firefly.