Created through the Digtal Pubishing Initiative at The Creative Media and Digital Culture program, with support of the OER Grants at Washington State University Vancouver.
From the enigmatic "Codex Seraphinianus" depicting an encyclopedia of an alien civilization to the deeply realized fantasy universes of Tolkien and Le Guin to Lucas's Star Wars movies and VR games, world-builders use rich details to transport us into their immersive territories of the imagination. But beyond its storytelling allure, world-building serves as an engaging demonstration of how AI can be leveraged to execute a complex set of tasks.
An exercise in AI world-building is particularly valuable because it encompasses a broad spectrum of research, creative strategy, and practical skills. This process may involve multiple modalities—text, image generation, video, audio, and coding—integrating various media to realize a cohesive representation of a world. Participants will explore skills in research, accessing and synthesizing complex information, and the practicalities of building and refining a dedicated GPT for specific tasks. This holistic approach enhances one's ability to harness AI tools effectively across diverse applications, making it a potent exercise for developing comprehensive AI proficiency.
At its core, successful world-building flows from a creative spark - an idea, image, or feeling expands iteratively into a self-consistent canvas upon which narratives can be painted. Each new element added, whether an ethnic group's traditions or the flora and fauna of a particular biome, must reinforce rather than undermine prior elements. World-builders become cultural anthropologists of the realities they conceive. Critically, the most transporting imaginary worlds challenge our assumptions and expand our conception of what is possible.
The above student project by Shannon Guo was for a world-builing assignment in my AI in the Arts class. The assignment involved using AI tools for researching a fictional world, copyediting (English is the student's second language), image generation, video and audio generation and coding assistance for web design.
Language Models and Generative AI open up many possibilities for creative world-builing within many different contexts. A movie production might use AI generators to work out the look, feel and compositions of shots well before any actual shooting. Video Game designers already use AI for responsive and interactive elements in gameplay. An historian might want to work with students to research and model an ancient city. Together, human and machine, can elaborate on conceived environments, characters and histories at a speed and scale exceeding our individual faculties.
Because AI generators are good at following specific tasks, human creators who can articulate with prompts and meta-prompts what they want can achieve remarkable representations of complex worlds. Precise language and detailed prompts are key. These creators can guide the AI to generate more specific details and make corrections by refining their prompts iteratively. For example, if a creator envisions a desert city with towering sandstone structures, they can start with a basic prompt and gradually add details about the architectural style, the types of plants in the oasis, or the way the light reflects off the buildings at sunset. By continuously adjusting the prompt, the AI can produce images that closely align with the creator’s vision.
But human co-creators who are not sure about what they want can also use AI generators to explore and iterate through possibilities. AI processes facilitate human imagining, especially for creators who may not have a vivid mental image of their world. Such creators can use AI as a brainstorming partner. They can start a chat by asking broad questions about a possible world, such as "What kind of civilizations could exist on a planet with three moons?" The AI can provide suggestions and options, helping the creator explore different traits and characteristics. This iterative dialogue can help creators discover aspects of their world they hadn't initially considered. For instance, they might ask, "What flora and fauna would thrive in a jungle with bioluminescent plants?" and then refine their ideas based on the AI's responses.
More than just straightforward text generation, AI language models excel as creative conversation partners to help brainstorm and refine the high-level foundations underlying an imagined world. By engaging in iterative dialogue, creators can envision:
Rather than static worldbuilding guides or setting bibles, this dynamic chat process allows creators and AI to develop a shared context spanning the world's deepest roots to its furthest branches. The humans supply visionary sparks which the AI then expands into elaborated, internally cohesive concepts to consider and refine.
Unlike working alone, this symbiosis benefits from AI's ability to hold vast knowledge spaces and generate logical extrapolations - freeing the creator to prioritize higher-level oversight, quality evaluation, and maintaining a unified creative voice.
While engaging generalized AI tools can be powerful, creating a customized world-building model allows for even tighter collaborative harmonization between human and machine intelligence.
One effective way to achieve this customization is by using ChatGPT Plus with GPT-4 to create a custom GPT tailored to your specific world-building needs. This process starts by defining the unique instructions and personality traits you want the AI to embody, which helps guide the model's responses to be more aligned with your creative vision. For instance, you might instruct the GPT to prioritize vivid descriptions, maintain consistent cultural and linguistic rules, or emphasize particular themes like mysticism or technological advancement.
The customization process often begins with an ideation phase, where you brainstorm key world elements such as geography, species, cultures, languages, and technologies. You can also generate or gather reference materials—like images, detailed descriptions, maps, and character profiles—which will serve as part of the training dataset for the custom GPT. By incorporating these elements, the AI model becomes increasingly adept at generating content that fits the specific lore and logic of your world.
As you refine the GPT through iterative cycles, you provide new prompts, evaluate the outputs, and give feedback on how well the model captures the essence of your world. Each round of interaction helps the AI internalize more nuanced aspects of the world's internal logic, cultural norms, and interconnected complexities. This iterative process allows the rough sketches of your world to evolve into a richly coherent alternate reality, with the AI offering creative contributions that are more in sync with your vision.
The value of using a custom GPT goes beyond merely avoiding the constraints of generic models. It lies in cultivating an AI that develops a unified creative voice that synergizes with its human collaborator—almost becoming a co-author with a shared vision. By tailoring the AI to your specific needs and preferences, you can create a tool that not only generates relevant content but also inspires new ideas, deepens the narrative complexity, and enhances the overall storytelling experience.
Advanced meta-prompting techniques allow world-builders to harness the recursive generative potential of AI systems. Rather than simply having the AI generate static descriptions, meta-prompts initiate dynamic feedback loops where the AI's outputs influence its subsequent outputs. This creates worlds that are self-referential, continuously evolving, and seemingly imbued with emergent properties.
For example, a creator could meta-prompt an AI to describe a civilization, then have it generate cultural traditions influenced by that initial description, which in turn impact the civilization's origins and future—an interlinked tapestry steadily enriching itself. The AI effectively co-authors an open-ended narrative by turning its outputs into new depth-compounding inputs.
Meta-prompting allows worlds to exhibit patterns of complexity found in living systems - adaptation, self-organized criticality, nested feedback cycles spiraling across scales. Historic events impact future technologies which reshape societal norms, looping back to redefine origin myths. Like our own universe, the world's evolution encodes remnant traces of all prior states, accumulating densely layered meaning.
World-builders utilizing these strategies are like epigeneticists, subtly modifying parameters to steer accelerated evolution down novel paths. The resulting worlds contain unforeseen emergent qualities arising from AI-facilitated recombination and mutation of core ideas over recursive generative iterations.
To make AI world-building an immersive, dynamic experience, this unit culminates in a multi-stage creative exercise that encourages exploration, imagination, and collaboration:
Participants begin by collaborating with an AI brainstorming assistant to develop a high-concept world premise. This could be anything from a dystopian future on a floating city to a fantasy realm where magic is powered by music. Use open-ended prompts like, "Imagine a world where..." to spark creativity. The AI will help generate initial ideas and suggest unique twists. For example, "Imagine a world where cities are built on the backs of giant, ancient creatures, and civilizations must learn to coexist with their movements."
Once the premise is set, use a world-building GPT to flesh out the details of the world. Prompt the AI to describe various aspects like geography, cultures, history, and technological developments. For example, ask, "Describe the history and culture of the nomadic tribes that navigate the oceans on the backs of colossal sea turtles." Use the AI's outputs to create a comprehensive world document that includes maps, timelines, and cultural notes. Participants can also play with the AI's responses to introduce unexpected elements, like a lost civilization beneath the waves.
With the world’s geography and locations described, use an AI image model to visually render these spaces. For example, prompt the AI to generate images of "a sprawling market suspended between two mountain peaks" or "a glowing city submerged under a crystal-clear sea." Compare these visualizations to the original descriptions and refine the text to better match the imagery, or adjust the images to align with the written world details.
Next, create vibrant character profiles using the AI, contextualizing each character within the world's lore. For instance, ask the AI to "Generate a character profile for a young inventor who creates devices powered by the magical energy of the sea." Once characters are established, write narrative vignettes that place them in the world’s key locations. For example, "Write a scene where the young inventor tests her latest invention in the bustling marketplace in the sky." These vignettes help bring the world to life and provide a glimpse into how characters interact with their environment.
Take the narrative vignettes a step further by using meta-prompting to evolve the story. Ask the AI to extend these scenes into plot continuations. For example, "What happens after the inventor’s device accidentally causes a storm that disrupts the floating market?" Use the AI’s narrative to explore how the world changes and grows over time, introducing new challenges or opportunities for the characters.
Finally, reflect on the world that has been co-created. Discuss with participants how the world’s complexities evolved, and consider what elements could be expanded or refined further. Engage the AI in a reflective dialogue by asking it, "What are the key themes that emerged in this world?" or "How could the world’s lore be expanded in future stories?" Use these reflections to iterate on the world, adding layers of detail and depth in subsequent rounds of creation.
To conclude, participants can present their world as a complete, hyperreal experience, showcasing the text, visuals, and narrative elements they’ve developed. This can be done through a collaborative storytelling session, a digital portfolio, or even an interactive map that allows others to explore the world. The presentation should highlight the AI's role in shaping the world, as well as the creative choices made by the participants.
This exercise not only produces richly detailed worlds but also highlights AI’s capacity to seamlessly blend text, visuals, and iterative world-building. By immersing themselves in this collaborative process, participants gain a deeper understanding of how structured invention and freeform creativity can combine to unlock the full potential of AI-enhanced storytelling.