In 2026, narrative games built with artificial intelligence still face a basic question: if players can create almost anything through text, how do they decide what story to tell?
Six years earlier, when Book of Infinity: 1001 Nights was a prototype built on GPT-2, producer Sun Yuqian was already trying to answer that question. After years of development, the game now gives players prompts and feedback to help them enjoy storytelling.
Led by Sun, Ada Eden is the outfit behind Book of Infinity: 1001 Nights, one of the earliest projects in China to experiment with AI-driven narrative gameplay. Its setting draws on the classic “One Thousand and One Nights.” Players take the role of Scheherazade, who discovers, while taking turns telling stories with her younger sister, that she can turn weapons described in stories into physical objects. Players must keep telling stories to an AI-controlled king, creating weapons along the way, with the ultimate goal of killing the tyrannical ruler.
For a game built around AI conversation, getting players to start talking is only the first step. What should they say? How should the AI understand them? And how should the game respond to what they create? Those questions have shaped its development.
Finding joy in telling stories
An early demo of Book of Infinity: 1001 Nights gave players one task: tell stories to the king, naturally work weapons into them, then use those weapons to defeat him. The king would respond to and steer the story based on his character and backstory. If a player’s intentions became too obvious, he could become enraged and have the player executed on the spot.
For players who are already good at storytelling, the appeal is clear. Matching wits with the king is part of the fun, but creating a brilliant or outrageous tale brings its own sense of accomplishment.
But how can players who are not natural storytellers, or who simply find the creative effort exhausting, enjoy it too? Sun said they need more guidance.
First, the game needs to show players how to begin. The early demo offered a single sentence as a prompt. The current version introduces an “Inspiration Card” system, which lets players use events that happen before they enter the king’s chamber to open their story. On subsequent playthroughs, players can also select a “Storybook” and continue a story left unfinished in an earlier run.
The game has also added a “Name Assist” feature. Sun came up with the idea after noticing that players often created more entertaining moments when they wrote people they knew, or familiar characters, into their stories. The team designed a feature around that behavior.
Once players have a way into a story, they need to see how their words affect the game.
In early versions, the king’s emotional state was partly hidden. Players often did not realize they had changed his mood until he was already angry. The team later found that, instead of scoring the story’s quality, the game could award points for actions taken while telling it, such as making the king angry, cheering him up, or triggering multiple items in a single sentence.
This feedback helps players understand what is happening and gives them a clearer sense of what to do next.
Sun said the design aims to encourage every act of storytelling, making it feel enjoyable, worth the effort, and capable of producing a reward.
That is also why Book of Infinity: 1001 Nights accommodates players’ unconventional tricks.
For example, players can enter a word such as “blade” in Name Assist. That makes it easy for the king to repeat the name, which can then trigger a weapon.
Sun does not necessarily see this behavior as a bug to fix. Finding a way to make the system produce an unexpected result is itself a form of play, she said, and part of the fun.
From telling stories to playing them
The earliest version of Book of Infinity: 1001 Nights began as a research project during Sun’s doctoral studies and led to an academic paper. As the team sought to explore more ideas, the project gradually evolved from an experimental demo into a full game.
That required a complete plot and world, capable of turning fragmented improvisation into a story with warmth, memorable characters, and enough depth to sustain players’ interest.
Sun believes her biggest mistake during development was underestimating how difficult the story would be to write. Reproducing the plot of “One Thousand and One Nights” was not enough. She wanted the protagonist to feel like someone who had truly existed, with a life and history of her own.
Discarding concept after concept gave Sun firsthand insight into why people in the industry often say that “the second level is the hardest one to make.”
The team initially gave the protagonist a reason to assassinate the king by having her younger sister suffer an accident. But that setup left the sister with little control over her own fate.
After repeated revisions with the writers, the team introduced a curse mark as a central plot device. It gave the protagonist’s struggle and journey toward redemption a stronger foundation and made the story more coherent.
With a fuller world and character backstories, Sun believes Book of Infinity: 1001 Nights now feels much more like a narrative game. Later chapters will introduce characters including a gardener, a doctor, and a general. As the plot unfolds, it will reveal the meaning of the curse marks on the girls’ wrists, why Scheherazade first seeks redemption, and the significance of the story’s clues and deliberate gaps.
A team that grew out of the game
Book of Infinity: 1001 Nights has developed alongside rapid advances in the AI technology on which it depends.
Its earliest demo dates back six years to a university project Sun worked on when GPT-2 had only recently become popular. The development team, however, was formally assembled and began operating only a little more than a year ago.
There was no single moment when Book of Infinity: 1001 Nights became a commercial game headed for Steam.
Sun describes a gradual process. As large language models matured, an ecosystem formed around the technology. The team grew, while the demo gradually proved that it was worth developing further.
With no comparable games to draw on, the project has never had a fixed direction. At times, even defining its genre has been difficult.
The team has relied on player feedback to guide successive changes.
While the demo was still an academic project, the team experimented with real-time scene generation to make play more immersive and provide immediate feedback. The results were disappointing, and players did not respond well, so the team cut the feature.
Sun said decisions like these have largely emerged from repeated encounters with players and their reactions to the game.
The team has also developed differently from those at most independent game studios.
Sun describes it as a group that “slowly grew out of the game as the game became more complete,” rather than one built through the more conventional sequence of setting up a project, raising investment, developing the game, and publishing it.
Advances in AI have changed its workflow, too. Nontechnical team members can now use AI coding tools to create gameplay demos and develop some smaller features themselves. When designers propose a feature, they first build a demo and hand it to the programmers, simplifying communication and development.
The game, its team, and the technology have grown together.
In a field with few established reference points, product updates, team building, and technological advances continue to shape one another.
Asked about her view of AI, Sun said that without her optimism about technology, she probably would not have chosen to work so deeply in AI-driven entertainment.
She expects token costs to keep falling and AI technology to keep improving, leaving substantial room for the field to grow.
For Sun, today’s AI games feel less like competitors than projects keeping one another company as they develop.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Han Ren for 36Kr.
