Experiment #07

Future Imperfect

Original text prompt

The Future of architecture

Technical notes

A study in dystopian landscapes authored by Midjourney in collaboration with academic and photographer Adam Brown.

The prompt he posed was a fairly neutral statement: “The future of architecture”

Onace again the RunwayML frame interpolation process was used, over 70 images were used from Adam’s Future Imperfect series.

On the Midjourney side, no further stylistic or technical parameters were used in the initial prompting phase, however one of the early images generated by Midjourney was used to embellish the text prompt. The Aspect ration was set at the default 1:1.

Procedural Notes

I was somewhat horrified that such a simple unadorned request of an AI could produce image after image of crumbling cityscapes and barren wastelands. Each variation rendered in deep reds and oranges with hot hazy disappearing horizon lines of infinite destruction. In the centre of the image rose a shiny abstract tower, all glass and steel and aqua blue accents, impossible architectural forms that reached pleasingly into the temperate cooling air of the stratosphere.

There will always be inequality, the algorithm was telling me. Winners and losers, air-con and keystrokes. The Cloudalists above and the Earth hardened data clones toiling away in the sun below.

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on Adam’s author page.

Experiment #01

Sad by Design

Original text prompt

Sad by design

Technical notes

A study in AI portraiture using the text-to-image service Midjourney in collaboration with Geert Lovink. The prompt was a simple provocation from Geert’s book title, Sad by Design.

33 sequential images taken from the series were blended together using RunwayML’s frame interpolation technology.

On the Midjourney side, no stylistic or technical parameters were used beyond specifying a 2:3 aspect ratio.

Procedural Notes

By persisting with a simple unadorned prompt and iterating doggedly through many steps of variation and upscaling it is possible to achieve fleeting moments of coherence. We can see the algorithm iterate through layers of noisy abstraction as the composition evolves across four discernible figures in portrait. There is an air of inevitability as coherence emerges amidst the abstract mush of shapes and lines and shadow.

These emergent patterns demonstrate how the algorithm seeks out formal references from the chaos and abstraction – the latent instinct of the diffusion model. In turn through our witnessing of the sequential playback, we are able to attribute meaning and mediated symbolism upon the ghostly visage.

Through veils of abstraction, we can also see how this process reveals the biases and narrow bandwidth within the labeling method used to train the Midjourney algorithm as it seeks out visual stereotypes: the horror film poster featuring a blood red silhouette, the traumatized Asian/Middle Eastern figure in a scarf, the forlorn female model disappearing into advertorial collateral and is that Picasso’s Weeping Woman? Perhaps.

The prompt mentioned neither gender or any form of embodiment. What is being designed here? Are women inherently sad – or just rendered that way by men and their algorithms?

Furthermore, seeing the images assembled in this manner – a form of containerized linearity – is reminiscent of the cascading reel of synthetic images generated on the Discord server (albeit vertically) by the Midjourney Bot.

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on Geert’s author page.

Experiment #05

Smart Plants

Original text prompt

Teleporting through Smart Plants.

Technical Notes

A study in synthetic biology using the text-to-image service Midjourney in collaboration with Melbourne-Pakistani academic Wajeehah Aayeshah.

RunwayML frame interpolation process was used to blend 37 images from the Smart Plants series.

On the Midjourney side, further stylistic and technical parameters were used to amplify the prompt.

Procedural Notes

The aim of this assemblage was to explore an organic luminescent thread which began appearing in Wajeehah’s prompt, a futurist bio-tech move: “Teleporting through Smart Plants.”

To emphasize the virtual laboratory feel of the generated images, I supplemented the prompt with some stylistic and technical parameters: “Metaverse, biotech, intricate detail, fine textures, OctaneRender, trending on ArtStation.”

What is interesting here is that the initial generations are abstract designs, the type of stuff you might find hanging in a dental surgery or in a highway motel that is trying to spruce things up a bit. Soft pastels, luminescent acrylics, industrial scale mechanical reproductions .

However, this quickly evolves into the fantasy world of sci-fi – Annihilation meets The Day of the Triffids – flora-hacking for the Anthropocene. Pinks, violets, oranges and luminescent greens and electric blues become increasingly pronounced as the organic material morphs into something alien, digital otherness. Without a doubt, we could feed quantum computers on this stuff!

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images from the Smart Plants series, please see Wajeehah’s author page.

Experiment #02

Reality Check

Original text prompt

How do I know if this is real?

Technical notes

Another study in AI portraiture this time in collaboration with writer and journalist Jennifer Perkin.

The prompt she posed was a question we have all been asking ourselves lately: “How do I know if this is real?”

Onace again the RunwayML frame interpolation process was used, however this time 125 images from the Jennifer’s Reality Check series were used.

On the Midjourney side, no further stylistic or technical parameters were used in the prompt parameters. Aspect ration was set at the default 1:1.

Procedural Notes

Clearly visible in the orb like faces are landscapes, sunsets, cracked earth, bleached valleys and occasionally hints of a baking cityscape – even a stack of old newspapers signposts the decay. In two frames a small figure briefly appears – a familiar narrative object present in images generated by Midjourney Version 3 (Jul-Nov 2022).

In total, for this prompt, 265 images were generated over several days in October 2022. The compositions that emerged from this simple query of the algorithm included Escher-type architectures, strange neon chambers, ghostly figures drifting down a deserted apocalyptic street and in one peculiar outlier thread, what seemed like a rolled-up length of pillows in the shape of an icy cave.

The images featured in this video permeated from the very first four-grid image output in Midjourney – an abstract portrait of a female, with an orb-like visage of liquid glass (or burning lava?) and a bob of black hair. After 80+ iterations the hair morphed into two faces in profile looking out to the edges of the frame.

The internal orb also changed to express synergy between these two “minds” – roots, branches, valleys, stars and finally in the last frame – where I stopped the process – tears. A sadness had emerged – was reality no longer shared? Or was it mourning the fact that the two could not really be as one, permanently staring off in separate directions.

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on Jennifer’s author page.

Experiment #03

1770 Futurism

Original text prompt

17th century etching in the style of Breugel of a Georgian vision of a 1970s modern research institute.

Technical notes

A study in retro-futurist re-animation in collaboration with artist producer Lev Manovich.

Again RunwayML frame interpolation has been used to build the sequence.

This video has also been more heavily edited to give the impression of entering a fantasy world of retro-futurism circa. 1770.

Procedural Notes

By blending twenty-five images into a video sequence we can examine a larger narrative within this ‘illustrated’ fantasy world of Lev Manovich’s prompt. Descending into the labyrinthine chambers of looping time we find a 1970s research institute as envisioned by an algorithm in 2023, pretending to be designing the future through a 1770 etching.

Looking at these distinctly sci-fi imaginings, we might ask:

“What are they building in there?”

“Why are they collecting everything?”

“Where are those long lines of people going?”

Lev’s original prompt was: “17th century etching, 1770s vision of 1970s modern research institute, white paper, lots of details, in the style of Breugel, cross-hatching,”

However, this produced somewhat muddled and inconsistent results. So the prompt parameters were modified into the following:

“17th century etching of a Georgian vision of a late 20th Century modern research institute. In the style of Breugel, on white archival paper, highly detailed, cross-hatching, rare.”

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on Lev’s author page.

Experiment #04

The Pink Dress,
2100 AD

Text prompt

If I was diagnosed ‘mad’ in the year 2100, what would I be wearing? Where would I be? Who would surround me?

Technical Notes

A study in AI fashion using the text-to-image service Midjourney in collaboration with artist producer Sarah-Mace Dennis.

This video sequences uses 19 unique images blended into one sequence using RunwayML’s frame interpolation process.

Procedural Notes

The original Discord image grid produced an image pair, a quirk that has became increasingly common in Midjoiurney Version 5 (March 2023). I have separated these in Photoshop and composited them as single image models. Furthermore, these are presented out of sequence to make the interpolation process as fluid as possible.

The prompt Sarah posed was a question: “If I was diagnosed ‘mad’ in the year 2100, what would I be wearing? Where would I be? Who would surround me?”

To focus the output I enhanced the prompt with stylistic and content parameters: “Futurism in the Anthropocene. Outsider fashion. Mixed-media, wearable art, street wise, artist impression.”

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on Sarah’s author page.

Experiment #06

3AM

Original text prompt

Monday morning 3.37am

Lover

Pining

Technical notes

A study of the early hours and the emotion of distance. A collaboration with artist/ researcher/ curator James Carey and the Midjourney bot.

The prompt James posed was a private moment, an observation perhaps, but a feeling all the same: “Monday morning 3.37am / Lover / Pining.”

On the Midjourney side, no further stylistic and technical parameters were used to amplify the prompt. The RunwayML frame interpolation process was used to blend 22 images from the 3AM series into a 01m 35s sequence.

Procedural Notes

What is interesting about this series of images is that they have been lifted from two separate threads, generated, firstly on October 12 and then again on November 8, 2022. Both of these threads began with images of streetscapes, as indicated by the first frame in this video. What is fascinating, is the algorithm’s tendency towards the ambiguous image, suggestive of both a physical and emotional presence.

The image construction is evocative of an entanglement. Are they two figures leaning towards each other? Or is that a phoenix or a dove swelling between them in the early AM light? And while the second sequence might suggest two bodies entangled on the sofa – replete with typical Midjourney disjointedness of hands and arms folding over one another – however, the resolution of the sequence is a solitary figure sitting on a bed … pining.

/////

For additional technical information including the original Midjourney Job ID and Discord instances for the images used in this video, please see the last entry on James’ author page.