Using GenAI for Worldbuilding
I recently published a new creative writing piece on my website, A Geographical Survey of Lethir in the Year 1323, a worldbuilding piece for my fantasy continent of Lethir. Written in the style of a lecture delivered by a professor at a college within the world, this piece is meant to be a guide for some of my future and in-progress work, set in this world. In fact, my already published short story, The Architect and Her Assistant, is actually set in this continent, thousands of years before the lecture is delivered.
But while the text of this piece is entirely my creation, I utilized OpenArt AI to create illustrations to visualize the locations in my fantasy world, some of which I have imagined within my mind for several years. OpenArt AI uses DALL-E to generate images, and allows you to select a consistent style for your images, which was appealing to me so that the images for this piece would be thematically consistent.
Let me be clear, I have grave doubts and mixed feelings about the use of generative AI for art. As a writer, I know full well the potential for plagiarism and unethical behavior, as well as the real concerns that artists have about their livelihoods. I will state, first and foremost, the art produced was not the quality I would expect from a professional artist. I’ll elaborate further on the limitations of the AI tool in this discussion. However, as I am not gifted in the visual arts, and I simply wanted to create a visual representation of the world I see so clearly in my mind, a free trial of OpenArt AI made sense for this project. Furthermore, as my day job is product marketing for an AI-powered technology solution, I have learned quite a bit about AI’s capabilities and limitations. I use OpenAI daily in my job to write marketing copy and build out campaigns, and find it useful (though not perfect), and wanted to apply that to my creative pursuits to see if it could be a useful aid.
Overall, I would say I was pleasantly surprised by the results and enjoyed the images created. They helped make my writing and worldbuilding come to life. OpenArt AI was easy to use, but requires you to have a good command of writing to prompt it to build what you want. The tool has some definite issues, though, which you can read about below in my detailed breakdown.
You can take a look at the piece and tell me if you feel the AI generated art is helpful and adds to the writing, or if it is terrible and ugly and everything that’s wrong with the world. I’m not convinced that AI art is good or bad yet, but I’m not afraid to experiment. The truth is, it’s not going away, so as artists and creatives, we have to learn to use it… and master it.
Here’s a more detailed breakdown of my experience:
The Good
Images capture about 85% of my vision
This was the most important, and surprising reaction to this exercise. I definitely had to tweak the prompts sometimes and give more detailed instructions, but generally the images it produced created the scenes I had in mind with only a few clear issues or discrepancies.
Sar Kailaz Monastery came out exactly as I imagined
Easy to learn
As a first time user of OpenArt/DALL-E, I was able to generate all the images for my writing project in under two hours. I didn’t use all the advanced features on the free version, but could quickly identify my desired art style, input descriptions, tweak them, and have usable, relevant art within minutes. Anyone with a creative mindset and good command of the English language can use this tool.
Empowers the less artistically talented
I like to think I’m a pretty good writer, but I’m an awful visual artist. I have grand ideas and images in my mind, and I can put them in words, but I rely heavily on software and tools to create them in a visual medium.
Consistent art style
The tool allows you to generate visually consistent images that add to an achieved overall vision. It’s very helpful for this type of project.
Shares my world with non-readers
Let’s be real. Most people these days don’t read, especially not fiction. I have friends and family whom I love, but I know will realistically never read a page of my writing. And that’s okay! I don’t enjoy watching TikTok reels but I support my friends who make them. In particular, I really enjoyed showing my parents images of my world, even though neither of them are fantasy fans, and seeing how they could appreciate and see the images in my head and learn a little about my creative process.
The Bad
The tool works best when you tell it to steal
Okay, that’s a loaded statement. What I mean is that describing a scene or art style doesn’t work as well as telling it to use an example and derive from that. For one image of a monastery beneath the mountains, it was struggling for quite a bit to generate the mountain style I wanted until I told it to make something like the Matterhorn. Similarly, it doesn’t do well reading a user created description of what art style it wants. But tell it to ape Rembrandt, and it can really pull that off. Is this a problem? That’s a debate for another day. But I will say this:
“Good artists borrow, great artists steal!”
- Picasso or Shakespeare or Gandhi for all I or the internet know
Really sucks at details for humans and animals
Honestly, this really wasn’t a huge issue for this particular project. I was mostly generating dramatic images of cities and landscapes, so having distinct and detailed faces wasn’t a priority. The blurred faces and shapes of humans actually add to the style of the art for me, so I was happy with the result. If I wanted actual details, I would have disliked it. There are two images with animals in them, and they are not well-rendered, if you look closely. The animals are elephants and mammoths, and I stuck with those animals because elephants are apparently the most well-generated savanna animals this tool can muster.
Try asking it to add buffalo or antelope in a Rembrandt style to one of my images. It came out really bad for me. I’m quite curious about why certain creatures or images are easier for the tool to generate so please contact me if you’re a smart person who understands these things better than me.
Can you spot what’s wrong with the mammoths?
It lacks sophisticated reasoning to blend certain concepts
As a fantasy worldbuilder, I like blending elements of cultures and history from different time periods. DALL-E does not understand this, at all. If you tell it to have a mix of Mughal and Japanese style buildings, it will give you one (usually Japanese because OpenArt AI is definitely for weaboos… lot of anime art styles built in). If I ask for it to generate Lombard Street in San Francisco but with Islamic architecture and no modern elements, it gives me Victorian houses with minarets and steamships on the Bay. Maybe later models will improve on this, but it was important to understand the limits of the tool’s cognition and conceptual understanding.
If I had cash to spare, I’d still get a real artist
Real art is still better. The details, the tooling, and the soul. There’s a certain genericness to AI art that really can’t be shaken, and I’m relieved by that. For a fun creative project, this is nice, but I’d love a real artist to create illustrations for one of my novels.