AI tower defense generator: waves are easy, tuning is not

Tower defense has an unusually clean structure. A path, some towers with a cost and a rate and a range, waves of enemies with health and speed, and an economy connecting them. You could write the whole specification on one page.

That makes it easy to generate. Ask for six towers and twenty waves and you have them in minutes as working GDScript in your Godot project. The trouble is that having them is not the hard part and never was.

The hard part is whether wave eleven is the moment the player realises their build was wrong, or the moment they stop playing. That is decided by numbers that look identical on paper.

Every good tower defense game is about a decision under pressure: spend now on coverage or save for the upgrade. Every bad one has an optimal build that survives everything, and once players find it the game is over even though the levels continue.

Avoiding that means towers that are good against something and poor against something else, and waves arranged so the composition keeps changing. It also means an economy tight enough that you cannot have everything.

None of that survives being specified in advance. You set the numbers, you play, you notice that one tower is always correct, and you say so. The change takes a minute. Then you play again. Twenty rounds of that is what separates a tower defense game from a tower defense demonstration.

The path and the ground around it are the level design, and you build them by hand in the builder. Where the path doubles back so one tower covers two lanes. Where the ground is too narrow to build on. Where elevation gives range and costs coverage.

A generated map produces a generic map, and in this genre that is fatal because the map is what makes each level a different puzzle. Placing terrain, water and structures directly is the work, and it is the enjoyable part.

Stage Engine is a desktop app for Mac and Windows. It is downloaded and installed, and it bundles Godot so the game runs locally on your machine. A late wave with several hundred enemies and projectiles is a real load, and local is what makes it comfortable.

Building needs an internet connection since Director 1, asset generation and world data come over the network. Playing does not. There is no browser version.

Free to download with a free plan. Indie is $10 a month and Pro is $50, buying more usage.

There are three questions worth asking every time you finish a run. Did I change my mind about anything? Was there a moment I thought I had lost and then recovered? Did I build the same thing I built last time? A good tower defense game answers yes, yes, no.

Same-build syndrome is the one to fear. If the opening is solved, the first third of every level is a formality, and players notice long before they can explain it. The usual causes are a tower that is efficient against everything and an early economy that is too generous to punish a mistake.

Both are number changes, and both are quick to try. The reason to have this list is that a run can feel fine while failing all three tests, and without the questions you tend to conclude that the game is basically working.

There is a fork every tower defense design reaches. Either the player expresses themselves through which towers they place, or through how they upgrade a smaller number. Both work, they pull in opposite directions, and games that try to do both fully tend to end up with an economy that supports neither.

Picking one early makes every later decision easier, including the map layout, because a game about upgrades needs fewer build sites and a game about placement needs more. It is worth deciding before you generate a roster.