The Making of Mr. Illustrious: How I Turned My Novel Into an Interactive 3D Game With AI

Contributor Story · Christian Smooth

I wrote Mr. Illustrious long before I tried to turn it into a game. The story, characters and world already existed. AI did not give me the idea. What it gave me was a new set of departments I could direct as I tried to move that story into a medium I had never built before.

The story came first

I first began writing Mr. Illustrious when I was 19 years old after a friend asked me to write her a story. What began as a one-off story quickly evolved into a series, which I later collected and self-published in 2014 after graduating from university.

As I matured as a writer and storyteller, I became less satisfied with what I had published. I felt like the story no longer represented who I was as a writer. So, in the spring of 2025, I unpublished the original Mr. Illustrious from Amazon and began rewriting the story from the ground up.

I was also an early adopter of generative AI tools such as ChatGPT, DALL·E 2 and Midjourney. I followed the technology closely, but the goal was never to have AI invent Mr. Illustrious for me. I wanted to give existing readers—and people who had never encountered the book—a new way to enter a world I had already created.

I wanted the game to feel like 2009

Because the book is set in 2009, I did not want the game to look like a modern AAA production wearing an old story as a costume. I wanted it to feel like something that could have existed during the late 2000s. The PS2-inspired character models, bold menus, R&B soundtrack and arcade-style minigames were deliberate creative decisions.

The visual limitations became part of the identity of the game rather than something I wanted to hide. That decision also gave me a standard I could use when evaluating every tool: I was not asking which model could make the most technically impressive image. I was asking which result belonged in Mr. Illustrious.

Putting different AI systems against the same manuscript

I initially put Claude and ChatGPT against each other. I gave both systems my manuscript and asked each to help me build an RPG from it. Starting from the same source material made their strengths and weaknesses very obvious.

I started with ChatGPT, and I found the initial results unappealing and lackluster. The user interface took up most of the screen, the minigames were overly complicated and, as a whole, the game was not fun to play.

Claude impressed me much more quickly with the 3D backgrounds it produced. The major problem was character rendering. Some of the SVG characters looked like they had been created by Picasso, and some of the 3D humans looked like balloons. One thing I will give ChatGPT is its image generation: it gave me a much better starting point for visual character work.

When safeguards start changing an adaptation

Given the mature nature of Mr. Illustrious, I also had to deal with both language models censoring or sanitizing parts of the book. Even though the characters involved were adults, the systems would sometimes soften the language, remove moments entirely or replace sexuality with vague, overly polite descriptions.

That was a creative problem, not simply a frustration with a filter. The explicit material is not randomly placed in the story. It reveals character, establishes Maximus’s lifestyle and drives many of the consequences that follow. If an adaptation removes those functions, it is no longer preserving the same story.

I had to constantly compare the game against my manuscript, restore the original tone and find interactive ways to represent those scenes without stripping away their purpose. The rhythm games became part of my solution. They allowed the player to participate in the intensity and pacing of those encounters without the game needing to show everything literally.

That process taught me something I now consider essential: an AI model’s safeguards can directly shape an adaptation if the creator is not paying attention. It was my responsibility to make sure the technology did not sanitize the identity out of my story.

The toolchain became a production crew

To work around Claude’s character-art issues, I used Midjourney and Google’s Nano Banana image-generation tools to design Maximus and the other characters. I then used those images as references for image-to-3D generators such as Meshy.

Meshy created models, Mixamo rigged and animated them, Blender compressed and stabilized them, and Claude helped integrate the resulting assets into the game. The major visual improvements did not come from finding one magical model. They came from supplying better assets and letting different tools handle the jobs they were better at.

Connecting Claude to Blender made a major difference. Before Blender entered the workflow, many of my assets were unstable and bloated. Claude helped me build a scripted pipeline that compressed and stabilized the models, reduced the size of the game and made the workflow much more repeatable.

Blender did not make creative decisions for me. It removed a barrier between my decisions and a playable result. I could create a character in Meshy, rig it in Mixamo and get it into the game with a much cleaner process.

The rhythm game exposed the difference between “working” and playable

The minigames are the most important part of the experience, with the rhythm system at the centre. In the original build, there was only one falling-note highway, and it was even repurposed as the fighting game. During testing, players would tap the correct spot and still miss.

What appeared to be one problem turned out to be four separate bugs. I uploaded the game’s ZIP file into ChatGPT and asked it to diagnose the problems. In that moment, ChatGPT became less of a generator and more of a code auditor. It unpacked the project, inspected the HTML, CSS and JavaScript, traced the input logic and fixed four problems that had survived my earlier Claude iterations.

When I asked Claude why it had not caught the problem, it explained—paraphrasing the exchange—that the sequences had effectively been validated as winnable by a bot, not as playable by a person. That distinction stayed with me. Something can technically function without being enjoyable or even realistically usable by a human being.

Human playtesting changed the game

Playtesting shaped much more than the rhythm system. Players discovered that driving controls did not always respond correctly on mobile, rhythm notes fell too quickly at the beginning, interface elements appeared before some minigames started, and the fighting-game camera occasionally clipped through characters during takedowns. The J-Nova fight also ran smoothly on computers but struggled on phones.

These were not problems an AI model could reliably discover by reading code in isolation. Someone had to play the game, notice what felt wrong and explain the experience in human terms. Each round of feedback changed the project.

The most challenging operational issue for me was dealing with usage limits and long waits between Claude sessions. I estimate that it took more than 200 exchanges with Claude before the project reached the point where I brought it back into ChatGPT for additional debugging and finishing work. The number matters less than the lesson: iteration was the work.

AI felt less like an author and more like a film crew

I was surprised by how much Claude was eventually able to help pack into the experience, especially the fighting minigames. I did not expect to get a functioning 3D fighter into the project. The character models and collision detection needed extensive revision, but the fights eventually became one of the game’s biggest features.

I was also able to incorporate my own music and sound effects. Those elements mattered because they made the experience feel less like a technology demonstration and more like something that belonged inside the world of Mr. Illustrious.

Using AI to help create the game ultimately felt collaborative in the same way a film set feels collaborative. I was still the writer, director, editor and tester. The AI tools functioned like different departments, each with its own strengths, weaknesses and occasional misunderstandings. I had to communicate what I wanted, review the results, diagnose what went wrong and keep revising until the experience matched my vision.

Designing a video game was a childhood dream of mine. AI helped me make that dream possible—but the part I value most is that the finished project still begins with the story I chose to write.