Making a Post-Apocalyptic Arcology in Midjourney

Alex Tully
7 min readSep 18, 2023

--

Two things inspired this piece:

  1. One was J.G. Ballard’s The Drowned World, a post-apocalyptic SF novel set in a future where the sun has grown brighter and forced humans to relocate to the polar regions.
  2. The second was the mega-scale arcology under construction in Saudi Arabia. Called NEOM, it will be 500 metres high and stretch for hundreds of kilometres across the desert.

Now while J.G. Ballard imagined people living in “half a mess tin” in Northern Greenland, I’m imagining that a newly thawed Antarctica would be home to an arcology built by people relocating from the Middle East (which would be the first area to be evacuated if the sun became stronger, since its temperatures are the closest to the upper limit of what’s survivable for humans).

I wanted to create an image with the following elements:

  • An aerial view of a giant futuristic arcology, with architecture similar to the mega projects of the Middle East
  • For this arcology to be on whatever kind of landscape would emerge from under the melting ice around the Shouwa Antarctic Research Station, which is the closest area of Antarctica to Mecca - This would look most similar to Greenland or Norway, where fjords carve into continental shields of incredibly ancient gneiss rock dating from before there were trilobites.
  • A barren landscape, where only moss patches have had time to grow
  • A muddy ocean with blooms of algae feeding off all the nutrients washed into the ocean from the melting ice and the rising sea levels - I wanted a long exposure image to make the water look eerily smooth.
  • For there to be a supercell thunderstorm with a tornado - The Drowned World mentioned “inversion layers” that create unstable weather when they disappeared, this site also models a temperature inversion plus a high level of directional wind shear in the relevant scenario, and I can imagine a lot of atmospheric instability on the boundary between a steaming ocean and a continent that only ever gets low-angled sunlight (if that). I wanted to capture the definition of the clouds using a circular polariser.
  • Ominous cloudy weather - Ideally there’d be thick fog because of the previously discussed temperature inversion, and I hoped that the long exposure would make that show up. But on the other hand I knew that the circular polariser would cut through the fog.
  • A rusting shipwreck, creating a post-apocalyptic feel
  • To enhance the post-apocalyptic unease with a fisheye lens, which would also get more of the surrounding landscape into the image
  • Leading lines to draw the viewer’s eye to the arcology - If necessary I’d inpaint these in like I’d done in Aurora Rice Cavates.

Here’s the eventual result:

Title: Apocalypse Catches Up (Made in Midjourney) https://www.instagram.com/p/CxVHGMMLoSG/

I started off by creating the image of the storm over the water, to inpaint everything else onto. I could do this easily by prompting: long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer, at dawn during thick_fog, of a supercell_thunderstorm with a tornado, ominously towering over a silted muddy_ocean, high_horizon — s 50 — style raw and selecting the below:

Made in Midjourney

I chose this thunderstorm because it looked almost like a nuke, which enhanced the post apocalyptic feel.

In this post, I explain my reasons for putting underscores between some of the words.

To change the land, I planned to use my tried and true method of inpainting with an image prompt (first discussed here). But I did something new to get the keywords. First I asked ChatGPT to list fjords in Greenland and Norway that might look like what I was after. Then I searched for these places in Google Images. GPT being what it is, some of the locations weren’t what I wanted. But for others, I downloaded some suitable images, then used Midjourney’s /describe function on them. One keyword that kept cropping up again and again was “gravity-defying”. So I included that in the prompt, which was: long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and a circular_polarizer, at dawn during thick_fog, of a silted muddy_ocean with algal_blooms, gravity-defying gneiss_cliffs towering out of deep_fjords on a jagged_coastline with dark_sand and gravel_beaches — s 50 — style raw

I chose:

Made in Midjourney

Going back to my image with the storm, I clicked Vary (Region) and selected all of the land. Then I inpainted with the prompt: <URL of Image of Fjord> long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and a circular_polarizer, at dawn during thick_fog, of a silted muddy_ocean with algal_blooms, gravity-defying gneiss_cliffs towering out of deep_fjords on a jagged_coastline with dark_sand and gravel_beaches
The text portion of the prompt was exactly the same as what I’d used to generate the image of the fjord.

Anyway that yielded:

Made in Midjourney

I had my algae, but the land enclosed looked too much like an oil slick, and the other land looked too alive. It was time for a second round of inpainting. To use as the image prompt, I generated an image in Midjourney with the text: long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer, at dawn during thick_fog, of waterlogged dark_soil with moss_patches and gneiss_boulders, gravel_beaches, dark_sand and gneiss_cliffs towering out of a silted muddy_ocean — s 50 — style raw and chose the below:

Made in Midjourney

Then in the main image I selected all of the land for inpainting, but kept some algae out of the selection so Midjourney would be likely to create it back in that area (as I discovered here). Then I used the inpainting prompt: <URL of Above Image> long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer, at dawn during thick_fog, of waterlogged dark_soil with moss_patches and gneiss_boulders, gravel_beaches, dark_sand and gneiss_cliffs towering out of a silted muddy_ocean — s 50 — style raw
(Again the text portion of the above prompt was identical to what had generated the image portion of the prompt)

Made in Midjourney

This was more like it!

Time for the arcology. To get more keywords I collected some images of futuristic architecture from the Middle East and used Midjourney’s /describe function on them. I decided to use the recurring keywords “gigantic scale”, “futurist”, “islamic architecture”, “bold designs”, thin steel forms” and “sleek metallic finish”, to make the prompt: long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer, at dawn during thick_fog, of gigantic scale futurist arcology_towers in the style of islamic architecture with bold designs, thin steel forms and a sleek metallic finish, on barren_mountains towering out of a muddy_ocean — s 50 — style raw

I kept the below image for further use:

Made in Midjourney

Then in my main image I clicked Vary (Region), selected a rectangle going up from the central island almost all the way to the top of the image, and prompted <URL of Above Image of Tower> long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer, at dawn during thick_fog, of gigantic scale futurist arcology_towers in the style of islamic architecture with bold designs, thin steel forms and a sleek metallic finish, on barren_mountains towering out of a muddy_ocean

After that and then a few further small cycles of inpainting to fix some ugly bits, I got:

Made in Midjourney

The last stage was putting in the ship, but it was the most challenging. To make the image for the inpainting prompt, I typed: long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer at dawn during thick_fog, of the heavily_rusted shipwreck of a bulk_carrier on gneiss_boulders in a silted muddy_ocean with algal_blooms — s 50 — style raw
I tried using the below two images for inpainting:

Made in Midjourney

But I only got good results with by inpainting with the second image in the prompt, and also by selecting an area of water that was much larger than necessary (fortunately Midjourney didn’t put in anything unwanted). The prompt was: <URL of Above Right Image of Shipwreck> long_exposure aerial_photograph shot by Sony A7R II with a fisheye_lens and circular_polarizer at dawn during thick_fog, of the heavily_rusted shipwreck of a bulk_carrier on gneiss_boulders in a silted muddy_ocean with algal_blooms — s 50 — style raw

After cropping, I got my final result:

Title: Apocalypse Catches Up (Made in Midjourney) https://www.instagram.com/p/CxVHGMMLoSG/

Fortuitously, there were already enough leading lines, so I didn’t need to put any more in manually.

I hope you like the image, and even if it’s not your thing, hopefully you can try getting prompts for your own art by using the /describe function on images you see and like on the Internet. And if you’re not sure what to search for, you can do what I did and ask GPT for suggestions.

--

--

Alex Tully

Into Generative AI, but 100% Human-Written Blog (every word)・Bachelor’s in Maths・Master’s in Linguistics (@ANU 🇦🇺 )・Taught myself 🇯🇵 and 🇹🇭・Digital Nomad