Generative art making is flourishing. Algorithms that turn text prompts into images, similar to DALL-E and Stable Diffusion, are emerging as viable creative tools. And they’re fuelling much debate about their artistic legitimacy and potential to pinch our jobs.

The sudden leap in fidelity of artificial intelligence (AI) art production has been made possible by advances in deep learning technologies, particularly natural language processing and generative adversarial networks.

In essence, a user can input a text description and the algorithm auto-translates this right into a cohesive image.

MidJourney interface showing the four-panel grid of image results from two separate text prompts. From here the user can decide to either upscale (U) or create further variations (V) from any of the 4 results.
Mitch Goodwin/MidJourney, Author provided

MidJourney – or MJ because it is understood to its passionate users – is maybe probably the most seductive technology for its painterly output and poetic interactions. The charm begins from the very first moment, with the command line prompt “/imagine”.

Augmented imagination

MidJourney founder David Holz has said users find their text-to-image interactions to be a “deeply emotional experience” with the potential for it to be therapeutic. He said:

There’s a number of beautiful stuff happening.

Triptych: Larry David as a warrior princess, a design for a phonograph by Trent Reznor, and a close-up of a rose
L-R: ‘Larry David’ from the Warrior Princess series by Brian Penny (2022-09-03); ‘An intricate schematic of a Victorian phonograph designed by Trent Reznor’ by GM Gleeson, (@NooYawkGurrl, 2022-07-15); ‘A rose is a rose is a rose’ by Mitch Goodwin (2022-07-26).

MidJourney plays with genre and form, using existing principles which have long informed media arts practice, similar to non-linearity, repetition and remix, to exploit the archive.

Holz has suggested the algorithm’s purpose is to “augment our imagination”.

My first image requests were whimsical queries, nocturnal flights of fancy, gentle tentative casts into the virtual spirit world.

As it seems, my melancholic prompts were unnervingly well-suited to the algorithm’s default aesthetic.

Triptych: a discarded surgical mask; a cyborg with glowing red eyes; a mother and a child sitting in a flooded railway carriage watching other commuters.
L-R: ‘Surgical mask discarded on a wet dirty street. In the gap, the radio plays the Beach Boys’ (2022-07-15); from the portrait series ‘Call Centre Cyborgs, 2037 AD’ (2022-07-27); ‘I believed we were all on this thing together … but I used to be mistaken’(2022-07-14)

Magic lurks inside the algorithm too. Ilya Sutskever, co-founder and chief scientist at OpenAI, describes the method as “transcendent beauty as a service”.

Artist and theorist Lev Manovich has poetically described his interactions with MidJourney as akin to working with a “memory machine”.

The recognition it’s a service but in addition a metaphysical experience is a brand new way of fascinated with tools of automation.

The technical process may be an imprecise science wherein slippages and overlaps are inevitable. As Manovich recognises, MidJourney remixes:

something from real history and popular stereotypes – real knowledge and fantasies. But we must always not blame it, because we do the exact same, on a regular basis.

An illustrated cloud; a snow-filled stage.
From the series, ‘My Favourite Century’ and ‘Stage designs for an unwritten play’
Lev Manovich/MidJourney/Facebook

Collaborative remixing

The MidJourney Bot is hosted on the social platform Discord creating an intoxicating cascade of generative screen works.

It is an inherently communal experience. The image stream also functions as a site of shared creation. If one other user’s composition catches your eye, you may co-opt their prompt – or the image itself – and refine it in accordance with your personal aesthetic preferences.

This collaborative remixing is what makes the MidJourney Discord channel as much a social experiment as a scientific one.

Joker's mask in rubbish filled alleway, grubby New York skyline; a raging bushfire consumes a blackened forest.
Text prompts: ‘The joker’s grubby mask discarded in an alleyway, Brooklyn, NYC.’ and ‘A brand new age dawns because the last of the forests are consumed.’
Mitch Goodwin & Kesson/MidJourney, Author provided

My research into the darkening aesthetics of digital media means I’m somewhat predisposed to spotting dystopian visions. The MidJourney Discord channel is actually a seductive rabbit hole for voyeurs of destruction.

Ghastly cyborg futures and post-nuclear wastelands would appear to be for the AI prompt engineer. I often see prompts citing the retro-futurist nightmares of artists similar to HR Giger and Zdzisław Beksiński and the cinematic tendencies of David Lynch and Andrei Tarkovsky.

As Bowie crooned on the cyber-noir album Outside, itself a chronicle of art world depravity: “there is no such thing as a hell, like an old hell”.

Users are also finding ways to use the technology in a moving image context. Notable efforts include a generative fashion demo, morphing amoebas narrated by a synthetic David Attenborough, Fabian Stelzer’s crowdsourced narrative SALT_VERSE and Drew Medina’s mesmerising fractal film Monsters.

The most meaningful assemblage I even have come across is Gabriele Dente’s SOLAR (the history of humanity drawn by machines), accompanied by a manifesto highlighting the associated ethical and industrial implications of neural networks.

Digital tools have long been enablers of speed, dexterity and flexibility for designers and artists. Studio professionals within the MJ community are already finding efficiencies of their workflows.

Two pages from a sci-fi comic book; a photographic studio; a hand holding a cold beer can
Comic book layout, ‘Death’s Dream Kingdom’ by Randall Rozzell (2022-07-27); beer label graphic by Bryan Launier (2022-07-26); Photo shoot using a Nine Inch Nails inspired industrial backdrop by Caleb Hoernschemeyer, Flow Productions, photo credit Jared Jasinski (2022-07-27).
Randall Rozzell, Bryan Launier, Jared Jasinski & Caleb Hoernschemeyer/MidJourney/Facebook, Author provided

A startlingly beautiful example of the chances for design and concept ideation come from architect and designer Cesare Battelli.

His series “space-kangaroo” is evocative of a mode of conceptual design pondering that blends aesthetics, functionality and fantasy.

Architectural hybrid-design, ‘space-kangaroo’ by Cesare Battelli (2022-08-19)
Cesare Battelli /MidJourney/Facebook

‘Spirit photography’

Eryk Salvaggio has described the technology of the more photo-realistic aspirations of the DALL-E platform as “a type of spirit photography” conjuring images replete with the ghosts and markings of past technologies: the fading image, the decaying medium and the corrosive chemical response.

This ability of reconstituting the past and enhancing the end result with techniques of capture and display and procedural degradation makes MidJourney especially fertile ground for “authentic” gestures of the fabulous and the fake.

A collage of ghostly faces in a severely degraded photographic images; a film studio showing Apollo 11 on a fake moon surface.
DALL-E variation of a picture from a dataset of photographs by Romanian photographer Costică Acsinte, by Eryk Salvaggio; The alternate history of the Apollo 11 moon landing, by Mitch Gates (2022-07-21).
Eryk Salvaggio/DALL.E/Mitch Gates/MidJourney

How much this sudden uptick of synthetic media will contribute to the glut of misinformation online nonetheless is uncertain. How does the visual historical record accommodate its synthetic mirror?

We also needs to consider the evolutionary implications for language and computation. With the democratisation of AI assistants, the sphere of human computer interaction is evolving rapidly as are the inherent entanglements.

And so, tonight as town sleeps I watch the feed and dream together with the machine. I punch in one other text prompt and wait impatiently for my MidJourney Bot to conjure its response. All the while I’m wondering as to the reach of the text into the algorithm’s code, and to what extent it’s, little by little, re-coding me?

This article was originally published at