1001 Nights: AI-Native Narrative Game Driven by Large Language Models

"1001 Nights" is a narrative game developed by Yuqian Sun and collaborators at the Royal College of Art in London, built around the premise that language can shape reality. Drawing on Wittgenstein's assertion that the limits of language define the limits of one's world, the game casts the player as Shahrzad, who transforms stories narrated by an AI King into tangible in-game objects such as swords, knives, and shields. The design is organized into two phases: a storytelling phase, in which the player strategically guides the AI King's narration using keywords, and a battle phase, in which the collected weapons are used against the King. The overarching goal is to rewrite Shahrzad's fate and diverge from the traditional folklore, a narrative choice that also challenges patriarchal norms.

The project's technical foundation combines large language model GPT-4 for narrative generation with the text-to-image model Stable Diffusion, which visualizes the gradual invasion of the story world into the game's reality as weapons are collected. Generative AI thus serves both as a co-creative partner and as the engine producing the game's evolving visual and textual content. The work was also presented as a physical installation that pairs a PC with a mini printer, which outputs the corresponding story text whenever a keyword is triggered, extending the theme of stories coming to life into the exhibition space.

The game has received public and critical attention, having been exhibited at Gamescom 2023 and longlisted for the 2022 Lumen Prize. Its development is documented in two peer-reviewed publications: an ICIDS 2022 paper on a co-creative text adventure built on a story generation model, and an AIIDE 2023 paper on the co-creative storytelling experience, which is also available on arXiv.