Ben Goertz

...And Then I Fed Lines From My Short Story Into Midjourney

In my previous post I fed lines from my short story, Subversion, into Open AI’s DALL·E. It took some experimenting to land on the postfix “… in the style of John Harris” to produce images I felt matched the style for the short story. The results were interesting enough to share and continue learning more about AI image generators.

Example from DALL·E

The day after I published, I read this amazing article on The Verge interviewing the founder of Midjourney.

Where DALL·E felt like a curiosity (with limited credits to play with) Midjourney is this boisterous, open Discord where you use /imagine ... as much as you want in the public channels for free. If you pay $10/month you gain full commercial license of the images you generate. The results have left my mind racing!

Midjourney Video Render

From the start, Midjourney leans towards stylized art rather than almost photo realistic output. The fact you can watch the system working to create the images feels like watching dreams or memories coalesce. Almost any input has beautiful outputs. The system is a black box but the illusion and feeling of control, picking and choosing variants, changing input keywords, all while a mass of other people are running their own commands in the same Discord channel is overwhelming. After a few hours I started to feel the limits of my imagination in what I could generate.

Midjourney The Feeling of a Poem

In a way, Midjourney sidesteps the Turing test. I wasn’t chatting with the AI to determine if it was sentient. But I also don’t chat with a friend and ask for dozens of incredible images (posthaste!) and then revisions all in a short period of time. It’s not humanly possible. This is something new.

... we see this technology as an engine for the imagination.
Midjourney is trying to figure out is, okay, how do we use this for people? How do we teach people to swim? How do we make boats? How do we dam it up?

The strangest thing to me is the human faces that I started to see after many runs.

Midjourney A Character's Face

I don’t have the control over the system to say, “Yep! That’s my main character. Keep them and now put them in a new scene.”

Midjourney A Variant of The Character's Face

I learned a few tips about inputting the same seed value to try to produce similar results to a previous image. But again, the levers seem far removed from actual control.

I still haven’t settled on what these swirls of feelings are from using Midjourney. It feels like a universe of stories I can only dip into and come back to reality with a few images. When I plunge back down into the river, the stories are different, the images change. There’s no firm control, but it’s an incredible opening into new worlds.

Midjourney A Telescope in the Distance