AI arcade game maker: short games, tight numbers

There is not much to an arcade game on paper. A player, a threat, a score, and a difficulty curve. The reason good ones are rare is that all of the quality lives in numbers you can only find by playing: how fast the second wave arrives, how much the score multiplier rewards risk, how long the invulnerability window is after a hit.

Getting to the point where you can tune those numbers normally takes weeks of programming. That is the part Stage Engine removes. Director 1 writes the loop, the spawner, the scoring, the lives and the game over in GDScript inside your project, and you start playing on the first day.

From there it is all judgement, and judgement is not a technical skill. If wave four is where it stops being fun, you say so.

Object pooling so a hundred projectiles do not cause a hitch. A spawn director with a curve rather than a table. Combo timers. A high score table that persists. Screen shake and hit stop tuned so a hit reads without being unpleasant. Pause that actually pauses. Restart that resets everything, including the three things you forgot about.

These are the classic reasons an arcade project dies at 80 percent. None of them are creative work and all of them are asked for in a sentence here.

Arcade covers a lot of shapes. A top-down shooter. A one-button endless game. A 3D arena where waves come at you from every angle. Stage Engine does not have a menu of arcade templates to pick from, deliberately, because a template only fits the game somebody already made.

Instead you build the space by hand — an arena, a lane, a scrolling field — and describe the behaviour you want. The layout part is fast because arcade spaces are usually simple. The behaviour part is where the work goes.

Where a hand-built space does matter is in the games that have one: an arena with cover, a track with a pinch point, a room with a hazard in an awkward place. Even a simple arcade layout is a design decision and it is quick to make.

Because runs are short, the loop here is unusually tight. Change a number, play a run, feel the difference, change it again. Twenty minutes of that will tell you more than any amount of planning.

The game runs locally on the copy of Godot bundled with the app, so a run starts immediately at your machine's framerate. That matters more in arcade than anywhere else, because a tool that adds two seconds of latency to the iteration loop makes tuning miserable.

Stage Engine is a desktop app for Mac and Windows. Free to download, free plan available, Indie $10 a month, Pro $50. Paid plans are more usage, not a better engine.

Building needs an internet connection since Director 1 and asset generation come over the network. Playing what you built does not.

What separates an arcade game that feels good from one that does not is a set of durations most players will never consciously notice. Four frames of hit stop. A hundred milliseconds of invulnerability after a hit. A shot that fires on the press rather than the release. A death animation long enough to register and short enough not to annoy.

Individually they are trivial. Collectively they are the entire quality of the game, and they can only be found by playing, changing one thing, and playing again. Anyone who has tuned an arcade game will tell you it is dozens of these passes.

The reason this matters for a tool is iteration cost. If each change costs an hour of programming you will make ten of these passes and stop. If each change costs a sentence and a run, you will make a hundred, and the game will be a different quality of thing at the end.

You do not need to know which number to change. Saying "getting hit does not feel like it registers" is enough to work from.