# Why Telling an AI "No" Broke My Game

Dive into the making of RPG Slots with creator Elric and learn practical, structured approach to working with AI in game development. The AI needs more than a good idea.

By Elric · October 13, 2026

![Why Telling an AI "No" Broke My Game](https://res.cloudinary.com/venusseriesai/image/upload/v1791871478/upload_hipqye.jpg)

By: [Elric](https://run.world/Elric)

I’m a single guy in my 50s, and I manage IT employees for a living. Before that, I was a programmer, and I’ve been playing video games long enough that my earliest adventures came as text on a screen. You read the description, typed what you wanted to do, and imagined the rest. The graphics department was between your ears.

That background came with me when I started building RPG Slots. I had ideas about what I wanted, enough programming experience to have opinions, and an AI that needed those opinions explained very clearly. Apparently, writing requirements follows me home.

> One thing that helps is having a working process:\
> REFERENCES -> ASSETS -> GAME -> SOUNDS -> EFFICIENCY

Start by gathering references and establishing the direction. Create the visual assets. Build the game using those assets. Add and refine the sounds. Then check performance and efficiency. You’ll still go back and change things. Game development has a way of introducing you to decisions you thought you had already made. But having an order gives each stage a foundation and keeps the whole project moving in a clear direction.

For [RPG Slots](https://w.run/elric/rpg-slots), one of my first priorities was making the entire screen look consistent. I wanted a game where the characters, backgrounds, effects, and interface all belonged together. That meant committing to a pixel-art style. Getting there required an instruction I learned to be specific about: IMAGEGEN for the artwork. No procedurally generated graphics.

If you’ve worked with AI on a game, you may have encountered the little people it can construct out of basic shapes. A square for the body. A circle for the head. A few ovals for arms and legs. Technically, we have a man. We have not necessarily established an art direction. I wanted actual image assets created for the game, with a consistent pixel-art style. That became part of the instructions from the beginning, and something I kept checking as the game developed.

There is an important detail here, though: “no procedural graphics” and “no procedural generation” are very different instructions.

> I once told it “no procgen.” It broke everything.

Procedural generation can still be useful in how a game works. My objection was to the graphics, and I had managed to turn a specific visual requirement into a much broader instruction. The AI followed the words more literally than I intended. That was a memorable lesson in choosing the words.

I used Codex throughout the process, and one of the most useful parts of my workflow was having it create animation overview pages. Creating a sprite sheet is one step. Seeing the animations play, together on a page where I can review them, is where I can actually make decisions. I can point to an individual animation and explain what I like or what needs changing. An attack might need more impact. A movement might look awkward. Something might look fine by itself but feel out of place next to everything else. “Make the animations better” leaves a lot open to interpretation. “That attack needs a clearer windup, and this movement looks too stiff” gives us something to work on. It also saves a lot of hunting through the game just to see the same animation again.

Once those assets are in the game, they need to work together in actual play. An animation that looks good on a preview page still has to feel right when the player presses a button. That is where the separate pieces start becoming a game. I take the same approach with sound. Attacks, taking damage, button clicks, dice rolling. All of it gets previewed. I want an overview where I can listen to each sound, request a change, or accept it and move on. I’ll tell Codex to search the internet for sounds that fit whatever we’re working on. Then I listen. A sound can fit its description perfectly and still feel wrong in the game. There’s also a difference between a sound you enjoy hearing once and one you’re willing to hear several hundred times. A button click should not become your enemy.

That leads to some of the less exciting details that help round out a game: let people mute the music, mute the sound effects, and adjust the volumes separately. Players have their own preferences, and sometimes they just want to play while listening to something else.

The last stage in that process is efficiency. Check the graphics and file sizes. Look for unnecessary work happening while the game runs. Pay attention to how it behaves on a phone, including how much heat it produces. Getting something running on my screen is one milestone. Making it comfortable to play on someone else’s device deserves its own attention. And that brings me back to the first stage, references, and the biggest lesson I’ve taken from creating with AI:

> CONTEXT IS KEY.

That applies to a game, a song, a book, or almost anything else you want to make. You can give an AI a short prompt and say, “Create my idea!” It will produce something. But you’ve also left it to guess about most of the decisions that make that idea yours. Give it something useful to work from. If an existing game captures part of what you’re aiming for, use it as a reference. Have the AI study the gameplay, examine relevant code where it’s available, research the mechanics, and keep useful notes. Give it videos and examples, then ask it to explain what it learned. Be specific about what matters in those references. Is it the visual style? The pace? The way an attack feels? The way the game teaches a new player?

> Here’s an example outside gaming.

Ask an AI songwriter to write a love song about a boy meeting a girl and falling in love. You’ll probably get something recognizable as a love song. You’ve given it a subject, but very little emotional detail.

Now ask it to analyse the relationship and emotional journey in a movie like The Notebook, then use that understanding to help develop an original song. There’s much more to work with: characters, obstacles, separation, memories, and reasons the relationship matters.

You’ve supplied a story with emotional weight. You still need to add your own secret sauce. Your taste, your experiences, and your decisions are what give the work a personal direction. A reference gives you a stronger starting point.

Building [RPG Slots](https://w.run/elric/rpg-slots) has meant making those decisions over and over. Choosing the style. Reviewing the animations. Listening to the sounds. Catching the details. Explaining what I meant when what I said wasn’t quite enough. I started playing games when the computer supplied the words and I supplied the pictures. Now it can help make the pictures, too. \
\
I still have to bring the imagination, and apparently be very careful where I put the word “no.”\
\
[Build on RUN ](https://run.game/catalog)· [Join the Discord](http://discord.gg/rundotcreators)