What It Means to Design Scenarios with AI

 

From the moment my kid said "visual novel," we have come this far. Twenty parts. Looking back, we have covered quite a distance.

Not a single line of code was written. Ren'Py was never installed. And yet the most important work of this project is done. Scenario design. The work of understanding the structure of emotion and translating it into systems.

What I Discovered on This Journey

At the start, the goal was "making a visual novel." By the end, the goal had shifted.

Making a visual novel is not technically difficult. Ren'Py handles most of the technical challenges. The difficult part is making a visual novel that truly moves the player's emotions. And for that, what you need is not technology but an understanding of emotion.

Love is not a single variable but the result of conditions. A scene is not a backdrop but an amplifier of emotion. The consequences of choices arrive not immediately but later. Good choices can become tragedy. Actions and inner selves diverge. Conflict is not created by villains. Endings are not scores but states.

These principles are principles of visual novel design, but they are simultaneously observations about relationships. The paradoxical experience of coming to understand relationships through designing a game.

There were unexpected discoveries too. When I asked AI directly "What is love?" it reeled off dictionary definitions. Not very useful. But when I asked "If trust is at 40 and intimacy is at 30, how would this character treat the other person?" the AI gave a concrete and persuasive behavioral description. Concrete questions within a system draw far more useful answers than abstract questions. This was a core lesson about using AI. Ask about emotion directly and you get cliches; ask indirectly through structure and you get real insight.

What Role Did AI Play?

In this project, AI served three roles.

First, the architect who establishes structure. The three-layer model for the emotion system, the variable structure for the context system, the flag design for the memory system. AI quickly built these structural skeletons. Work that would have taken many times longer for a person alone.

Second, the reviewer who catches what is missing. When I asked "If this variable changes like this, what should happen to that variable?" it verified logical consistency. Effective at catching edge cases that humans overlook.

Third, the conversation partner who clarifies thinking. To explain something to AI, I first have to know it precisely myself. In the process of telling AI "This character is this kind of person," my thinking became concrete. More often than AI giving me an answer, I found the answer through the process of asking AI a question.

And what AI could not do. Judging emotional weight. Feeling "why this scene is sad." Writing a single nuanced line of dialogue. These remain the human domain. And they probably will for a while yet.

To give a concrete example, when I asked AI to write the final line for the Ruin ending, it proposed "We loved each other too much and that's what destroyed us." Not bad. But too explanatory. The line I wrote was "The days I cried because of you outnumbered the days I cried because you were gone." The same emotion, but the latter hurts more. Because it is specific. Emotional weight comes not from abstract declarations but from concrete experience. AI provides the structure, but the single line that goes inside that structure must be written by a person.

Can I Show It to My Kid?

I remember that this project started with a single remark from my kid. My middle-school daughter saying "visual novel."

There is no game yet. No code either. But there is a design. A structure for how emotion should flow, what weight choices should carry, and how relationships are built and broken.

What comes next is translating this into code. Implementing the emotion system in Ren'Py, writing scenes, placing choices, and testing. That will be the next season of this series.

What I honestly worry about most is the design getting compromised during implementation. The emotion system designed across 20 parts could get simplified when translated to code. The 14 variables might be cut to 10 because they are too many, or the context combinations might be trimmed to essentials because they are too complex. That is fine. Design is the ideal, and implementation is reality. What matters is preserving the core principles. The principle that "love is not a variable but the result of conditions." The principle that "the consequences of choices arrive later." The principle that "good choices can become tragedy." As long as these three survive, even if a few variables are dropped during implementation, this visual novel will be meaningful.

Will the day come when I show my kid a finished game? Probably. And watching which choices she makes while playing, what emotions she feels — that will be the true ending of this project.

I allow myself one fun thought experiment. What if she plays and reaches the "Late Love" ending? She is in middle school, so she would probably choose the caring option every time. Then she would experience that caring coming back as avoidance. What she would think in that moment. Perhaps she would understand in a single playthrough what her father designed across 20 parts.

To the Reader of This Series

This series is not a visual novel production tutorial. Twenty parts were written without a single line of code. This is an exploration of emotion, a record of collaboration with AI, and an analysis of relationships.

If you also want to make a visual novel, do not start with code. First, answer this question. "Why do people feel real emotion in a fictional relationship?" When you have your own answer to that question, only then can you design emotion.

And if you are going to use AI, remember just this. Do not tell AI "Write me a moving story." Ask AI "Given this combination of variables, how would this character behave?" AI is strongest when it works within a structure. It is worse than a human at free-form creative writing, but faster and more accurate at logical reasoning within a system. The core of AI collaboration learned through this project ultimately comes down to this. Give AI good constraints. The more specific the constraints, the more useful the answers become.

AI can give you structure. But the emotion — you have to put that in yourself.

Next Steps

This series is the scenario design arc. Next comes the implementation arc.

• Implementing the three-layer emotion system in Ren'Py
• Coding the context system and weather system
• Memory system and flag management
• Writing one full episode of actual scenario
• AI-assisted dialogue generation and testing

What I felt most while writing 20 parts was that collaborating with AI is not "faster than doing it alone" but "reaching places you could not reach alone." The three-layer structure of the emotion system could have been conceived without AI. But the principle that Layer A does not change and only Layer C changes, and the design implication that this preserves a character's personality while only the relationship shifts — that was a conclusion reached through dozens of rounds of "Why?" exchanged with AI. Not a matter of speed, but of depth.

Starting from my kid's single remark, a journey of designing emotion together with AI. The scenario arc ends here. Next time, we return with code.

---

Series "Building with AI: Visual Novel Project — Scenario Design Arc" Complete

Continues in the Implementation Arc.