In the previous chapter, I introduced what this guide will be about, namely creating a rich and detailed fictional universe using AI language and art models. So, let’s begin!

In order to bootstrap our fictional universe, we’ll first need to carefully craft our starting prompt. Fictional universes come in a huge variety of archetypes. If you’re planning on creating one, you probably already have an idea of what you want. Is your story in a fantasy setting, like those depicted in Lord of the Rings, Game of Thrones, or World of Warcraft? Is it in a science fiction setting similar to Star Wars, Stargate, or Battlestar Galactica? Is it grounded in today’s reality? In a magic-wielding, dragon-filled medieval setting? Or is it set in a high-tech future? Is your story or game set on a single planet, a single city, or across a whole network of star systems? All these choices will define your first prompt, the output of which will form the basis of subsequent prompts used to develop lore and detail.

So, getting it right is important.

It would be impossible to demonstrate prompts for every possible fictional scenario one might want to create in this book. So, for the purposes of this guide, I’ve focused on a single particular setting – a highly advanced alien planet, inhabited by numerous different alien races, somewhere else in the universe. Humans living amongst aliens – a setting similar to that depicted in Star Wars.

Here’s my initial prompt:

Our story starts on a thriving planet full of natural wonders, high-tech cities,  cultural venues, and advanced alien races who coexist peacefully. Please give the planet a graceful, memorable, and short (four letters or less) name and present some facts about its size, host star, moons, location in the universe, and nearby habitable planets that it trades with.

It won’t matter too much which large language model you run this prompt with. Smaller and older models such as the original GPT-3 should work just fine. In fact, from my own testing, newer, larger models don’t do any better when it comes to writing fiction.

Since this prompt will set the scene for your entire fictional universe, do run it over and over until it puts out something to your liking. Try also varying the model’s temperature setting, if your interface allows for it. A higher temperature should cause the model to be more creative. I normally set the temperature to about 0.8.

I personally used “lex.page” to interact with language models during the writing of this guide. It’s a word processor with integrated AI functionality that is eminently suitable for the usage I’ll describe in these articles. I highly recommend using it.

The reason why I asked the AI to give the planet a graceful, memorable, and short (four letters or less) name is because, after multiple attempts without that clause, it continually came up with cliched, seventies sci-fi names such as Xyloquinox, Xylophar, Zentorion, Zanara. Also, the model was eager to generate planets in orbit of a binary star system, something I didn’t really want. Tricking the language model into generating things that aren’t predictable and unimaginative is what this book is all about. And when that can’t be done, I’ll show you how to use what it generates to good effect.

Note that the prompt above is merely an example. If you want your fictional planet to be in the Milky Way, simply specify it. If you want the planet to have a more alien name, do that. If your world is full of wizards, dryads, elves, and dragons, state it. The sky’s the limit!

For those of you familiar with prompt engineering and common, tried-and-tested formats for prompting large language models, you’ll notice that my prompt doesn’t include a role. For those of you unfamiliar with this concept, assigning a role is as easy as including instructions such as “As a scifi novelist…” or “As a fantasy writer…” to the prompt, usually before the action – the thing you are asking the AI to do. I omitted the role in my prompt because I wanted the AI to write something formal and wikipedia-like, and that’s what language models tend to do by default. Adding a role such as “As a scifi novelist…” to my prompt encouraged the AI to write in a more long-winded, descriptive manner. As you’ll see later on, we will want to save short summaries of the outputs we obtain, so concise, formal language is more suited to that task.

Here’s the output I eventually settled upon:

Lira is a small, verdant world almost twice the size of Earth. Orbiting a stable G-type main-sequence star not unlike our sun, this planet features a single, shimmering moon known as Veil. Lira's position in the Laniakea Supercluster puts it within favorable proximity to several other habitable planets. The most prominent among its interspatial neighbors are the terrestrial worlds of Zyre, Hela, and Tora. Together, these planets form the Amity Ring—a collective known for its interplanetary trade and cultural exchanges.

Funny how it designated Lira as a small planet when it is, in fact, twice the size of Earth. Still, it sounds like a nice place to live. Probably nicer than this rock, anyway. Although I wouldn’t fancy those longer flights.

Now that we’ve named our planet, let’s use AI to define its geography. Here’s the prompt I settled on for this task:

Please name and describe each of the planet’s continents. For each continent name and describe the location and look of any notable forests, mountain ranges, or other natural wonders. In each case, come up with unique, graceful names that do not contain any English words or terms whatsoever. Start by telling me how many continents there are. List them. Then go into descriptions in subsequent paragraphs.

There are five continents on Lira: Zirah, Velnas, Ero, Sion, and Orbix.

Zirah, the largest continent, is distinguished by its vast Andara Forest, an expanse of bioluminescent trees that glow with a multitude of colors during the nightly cycle. The heart of Zirah is dominated by the towering Crystalline Spire Range, whose peaks are composed of natural gemstones, casting a sparkling tapestry against the horizon.

Velnas is a land of sharp contrasts, featuring the expansive Shrouded Plains, where perpetual mist clings to the ground, creating an ever-shifting landscape. The continent's eastern peninsula is watched over by the Nosk Volcano, a gentle giant swathed in ice and snow, known to harbor hot springs with curative properties.

Ero is an archipelagic continent formed by a cluster of interlinked islands. The Kivu Arch, a natural stone bridge, is a particularly stunning formation that arcs gracefully over the turquoise waters of the Sea of Talir, connecting two of Ero's major islands.

Sion, the most culturally diverse continent, is home to the Sahiri Desert. Within this vast expanse of golden sands lies the Oasis of Mirages, famous for its illusionary visions. Sion's northern territory, however, boasts the lush Yavi Rainforest, where the flora and fauna emit soft, harmonious sounds providing a natural symphony that enchants all who traverse its depths.

Orbix is unique in its mostly aquatic terrain, punctuated by small landmasses that are teeming with life. The Tesserine Lakes, a network of interconnected bodies of water, reflect the sky in a mirror-like fashion, giving an effect of floating towns and forests.

The names and descriptions generated by the model are interesting enough. Note that it didn’t include physical locations – whether the continent was in the north, south, or on the equator. This was fine for me – creating an image of the planet’s continental structure is easier if I don’t need to re-generate until I find a map with an exact tectonic configuration.

Let’s create some oceans, too. Remember to prepend the previous output to your prompt.

Please name the planet’s oceans. For each one, describe where it is located in relation to the planet’s continents.

There are three major oceans on Lira: the Belenic Ocean, the Thalric Ocean, and the Silvian Ocean.

The Belenic Ocean, surrounding Zirah to the north and Velnas to the south, is known for its crystal-clear waters and deep underwater canyons. It serves as a crucial navigation route for the planet's inhabitants and is home to an array of luminescent marine life.

The Thalric Ocean lies to the west of Sion and Ero, characterized by its stormy waters and the legendary whirlpools near the Stratos Isles. It is known as a sea of discovery, where many ancient shipwrecks await those brave enough to delve into its depths. 

The Silvian Ocean, to the east of Orbix and south of Ero, is the most tranquil, with gentle waves caressing beaches of pearlescent sand. Underneath its serene surface, the Coral Labyrinths offer a spectacular and intricate ecosystem, hosting thousands of species unknown to any other part of Lira. Here, the harmonious relationship between the alien races and their environment is most apparent, as conservation and respect for this natural beauty is a cultural priority.

At this point, we’ve generated enough of a high-level description of our planet. And a fair amount of text. In order to continually reuse this, we’ll want to generate a short, clean summary that can be prepended to further prompts. Let’s do that. Again, if you’re using something like ChatGPT, remember to include everything generated prior into your prompt.

Please summarize the above.

Lira is a thriving, peaceful planet twice the size of Earth, with a single moon, Veil. It's part of the Laniakea Supercluster and forms the Amity Ring with nearby habitable planets Zyre, Hela, and Tora, fostering trade and cultural exchanges.

Lira features five unique continents: Zirah, home to the luminous Andara Forest and sparkling Crystalline Spire Range; Velnas, known for the Shrouded Plains and the snowy Nosk Volcano; Ero, an archipelago with the striking Kivu Arch and the Sea of Talir; Sion, culturally rich with the Sahiri Desert and the musically enigmatic Yavi Rainforest; and Orbix, with its Tesserine Lakes that mirror the sky.

The planet is embraced by three oceans: the clear-watered Belenic Ocean with its deep canyons; the tumultuous Thalric Ocean, full of mysteries and ancient shipwrecks; and the calm Silvian Ocean, with its Coral Labyrinths teeming with a myriad of life.

Copy-paste your own version of that summary into a text file. You’re going to be using it in subsequent prompts and we’ll be adding to it as we move along. For now, we’re done setting the scene. Time to visualize what we have so far.

For the purposes of AI art in this book, I’ll be using Midjourney. This is because it’s the only art model I’m familiar with. Prompting Midjourney is a lot more difficult than prompting ChatGPT or indeed any LLM. The prompts I provide here may or may not work with other art models.

To visualize our planet, Lira, I’ve opted for a two-hemisphere map. Here’s the prompt I used:

Midjourney prompt: a two hemisphere map of an offworld planet with five continents; blue on black; simple annotated line drawing. –ar 2:1 –v 5.2

While I was writing this chapter, Midjourney was still on version 5.2. That model was incapable of counting beyond two or three. Thus, I had to re-run the above prompt many times. In some cases, it would output a map of Earth. In others, one hemisphere featured the Americas, while the other contained a unique set of coastlines. As with everything AI, just keep going until you get something that conforms to the description the LLM provided. And be patient with this one – I ran over fifty jobs to get one image that sorta worked. The 2:1 aspect ratio should be enough for a two-hemisphere map, but note how the one I finally chose still overlaps the two hemispheres slightly, as it did with many of the images it created. Here, we’re looking for at least three large continents and some archipelagos. Here’s an annotated version of the map I finally chose.

I tried similar prompts in Midjourney 6 and they provided me with very similar images. It still couldn’t count.

We can also visualize Lira’s position in the Laniakea Supercluster with an image generated using the following prompt:

Midjourney prompt: a celestial map of the Laniakea Supercluster; simple blue on black; annotated; –ar 3:2 –v 5.2

Here’s an annotated version. I added the label and arrow.

That’s a wrap for our first chapter. To summarize, we used AI to name a planet and tell us something about its host star, moons, location in the universe. We then created some basic geographical facts with descriptions of continents and oceans. Finally, we generated a summary to be used in subsequent steps and a few images illustrating our creation. 

In the next chapter we’ll expand upon that lore further by creating more detailed descriptions of each continent.