CREATE WITH MARK / THE GAME-BUILDING GUIDE

HOW TO MAKE
AI GAMES
THAT DON’T SUCK.

Don’t build it all at once. Build it in steps.

Play video: How to Make AI Games That Are Actually Fun (Step-by-Step)

WATCH THE VIDEOHow to Make AI Games That Are Actually Fun (Step-by-Step)

THE COMMON MISTAKE

ONE PROMPT.
EVERYTHING AT ONCE.

The AI has to guess all the decisions
you haven’t made yet.

THE ENTIRE REQUEST
“Build this.”
IDEA → WHOLE GAME → HOPE FOR THE BEST
Here’s the better process.Align. Build. Approve. Move on.
01 / PROVIDE THE INITIAL INPUT

Show the AI
what you mean.

Your idea.

What is the game? How does it work?
What is the player’s goal?

Your references.

Share reference games, links and images.
Point out the graphics or features you want.

What to research.

Ask it to study the useful parts of that genre:
its controls, core mechanics or pacing.

Visual reference of a small robot on a floating workshop island, with cream stone, orange energy cubes and a blue sky.
A SIMPLE EXAMPLE

“A little robot collects five sparks, avoids the patrols and reaches the exit.”

Use this image for the colors and workshop mood.

The result: shared direction, before the building starts.

02 / ESTABLISH THE FOUNDATION

Choose the right
amount of planning.

SIMPLE ARCADE-STYLE GAME

Use the initial description.

Clear controls. Clear rules. Clear goal.
Keep the game small and specific.

A SHORT BRIEF CAN BE ENOUGH.
MORE COMPLEX GAME

Create a game design document.

Have the AI organize the mechanics, systems,
visuals and scope into one agreed plan.

EVERY PIECE FOLLOWS THE SAME PLAN.
Either way: make sure you and the AI agree on what is being built.

The result: an approved foundation to build from.

03 / FORMULATE THE ASSETS

Build the pieces.
Lock in what looks good.

Make assets individually.

Characters, scenery, objects, UI and sound.
Use the references and agreed plan.

Review before moving on.

Ask for specific changes until they fit.
Approve the pieces you want to keep.

A workbench helps here.

Ask for one page with asset previews, approvals
and live updates on what the AI is working on.

Asset workbenchEXAMPLE
Working onPlayer asset · ready for your review
PlayerNeeds review
CollectibleApproved
Review → revise if needed → approve and lock in.

The result: a consistent set of approved, usable assets.

04 / ASSEMBLE AND DEPLOY

Turn the pieces
into a game you can play.

Assemble the approved pieces.

Have the AI connect the assets and game rules
into one playable experience.

Deploy it.

Publish a playable link or export a downloadable build
for your chosen platform.

Play. Then improve.

Try the actual game. Give specific feedback.
Fix one thing at a time and test again.

APPROVED ASSETS+GAME RULES→GAME
!→
ASSEMBLE → DEPLOY → PLAYPLAYABLE

The result: a game built from decisions you reviewed along the way.

YOU FINISHED THE GUIDE

You know the steps.
Here are the files.

Every step here goes faster when you are not staring at a blank prompt. The AI Creator Bundle is my own library: the prompts I actually use and the editable files behind the games, worlds and animations I build. Open one, change it, make it yours.

GAME DEVELOPMENTWORLD BUILDERARTIST TOOLKITANIMATION
Open the bundle ↗Join the Discord ↗

One payment of $9. Weekly updates included, no subscription. The Discord is free either way.

Animation packGame Development packArtist Toolkit packWorld Builder pack
THE AI CREATOR BUNDLE$9one-time
  • 250+ prompts & editable files
  • 4 packs, updated every week

createwithmark.co/bundle

OPTIONAL GUIDANCE
A QUICK WORD ON SETTINGS

Start simple.
Turn things up when needed.

Model

Which AI you use.

Start with a capable coding model; use image tools for artwork.
Try a stronger model for a hard task your current one cannot solve.

Effort

How deeply it thinks.

Default / medium is a practical starting point.
Low for easy edits. High for difficult planning or debugging.

Loops

Repeated attempts.

Usually, review after each step yourself.
Automate only a clear, testable task. Set a limit and a stop condition.

More isn’t automatically better. A clearer request is often the next thing to try.

If a loop repeats the same failure, stop and diagnose it.
Model options and effort labels vary by tool.

Example prompt

Replace the example details with your game. Keep one task in each request, then play or inspect the result.

Notes & sources.

A simple workflow for small, AI-assisted games. Provide the initial input, establish the foundation, formulate and approve assets, then assemble and deploy. Models, effort and loops support the work; they do not replace a clear process.

GitHub Docs · Prompt engineering ↗

Give useful context, break up complex work and iterate on the result.

Godot Docs · Importing images ↗

Check image formats and import settings so assets are usable in the game.

Roblox Creator Hub · Prototyping ↗

Keep experiments small and use playtesting to find problems.

Keep the plan proportionate: a short brief is enough to start a small demo. A larger game may need a fuller design document. If a mechanic is especially uncertain, ask for a small test before investing in a large asset set.

The workbench is optional: the example here is an illustrative asset-review layout. It does not monitor a real project. Ask your AI to connect the actual workbench to real task and asset data, and update it as work changes.

OpenAI · Model and reasoning-effort guidance ↗

GPT-5.5 documentation recommends medium as a balanced starting point and increasing effort when evaluation shows a useful gain. The settings slide is practical guidance, not a benchmark across all models. Labels, costs and behavior vary.

Loops: use a concrete success check and a bounded budget. Two passes is an example limit, not a universal rule. Automated checks can catch bugs; fun and feel still need human playtesting.

Real expectations: expect revisions, fixes and playtests. A smaller scope gives you fewer things to solve at once. This guide does not promise a particular build time or guaranteed result.

Artwork & font credits

Original AI-generated Spark Run reference artwork, created for this guide. Brand direction follows the supplied Create With Mark board. The small asset and game diagrams are illustrative code-native examples. Space Grotesk, Inter and IBM Plex Mono are embedded under the SIL Open Font License.