A Godot metroidvania template is a platformer with a map screen

Godot has no official metroidvania demo. Its own demo projects run to a 2D role-playing demo, a truck-driving demo, a multiplayer bomber, and nothing shaped like this. So the projects people share are community ones, and the one that circulates most is the Darkcom Metroidvania repository on GitHub: a 2D side-view starter with a character controller, enemies, connected level scenes, map data, and the menu and options flow around them.

That list is the standard shape. A metroidvania starter is a platformer starter with three things bolted on: a dash or similar ability, a health bar, and a minimap. It is a real project and it will save you real time, but it is a smaller head start than the genre label suggests.

The honest comparison is a plain 2D platformer starter, of which Godot has more and better maintained ones. What you would be missing is ability gating, save persistence and map data, and two of those three you will write yourself anyway.

A metroidvania is a graph of locks and keys wearing the costume of a place. You see a ledge you cannot reach in the first ten minutes, you find the double jump in the fortieth, and the game is good in proportion to how well it remembered that ledge and how satisfying the walk back is. That structure is the creative act of the genre.

Nothing about it is code. It is a decision about what order abilities arrive in, which regions each one opens, how many routes exist through the middle of the game, and where the shortcuts unlock so that backtracking feels like mastery rather than a chore. A starter project cannot contain a line of it, because it is not the kind of thing that lives in a file.

The practical consequence is a working order. Sketch the region graph on paper first. Decide which ability opens which door, then check that every region has an exit reachable with the abilities the player must already hold to be standing there. Most abandoned metroidvanias have an excellent dash and a map that is still three rooms in a row.

Movement feel decides whether people keep playing, and it is almost entirely a tuning problem. Jump height and the arc it makes. How much you can steer in the air. Coyote time, so a jump pressed just after leaving the ledge still counts. Jump buffering, so a jump pressed just before landing still counts. Whether the dash cancels into a jump.

A starter arrives with somebody else's answers to all of those, and they will feel wrong to you, because feel is personal and theirs was tuned against their game. So the first thing you do with an inherited controller is change it, which means the first code you have to read properly is the code the author cared most about and structured most tightly.

This is why a metroidvania starter is a weaker deal than a racing or tower defense one. There you can leave the inherited systems alone for months. Here the one thing you must change on day one sits at the centre of the project.

A metroidvania has to remember things across a save file, and the list is longer than it looks. Which doors are open. Which bosses stay dead. Which items are gone from the world. Which abilities you hold. Which rooms have been seen, so the map can draw them. Where the last save point was.

Starters usually stop short of that, because a demo with three rooms does not need it. Retrofitting persistence into a project that assumed rooms were disposable is one of the genuinely expensive jobs in this genre. Every scene that spawns something has to ask whether that something is already gone.

The map screen is the same problem in different clothes. A map is a data structure that has to stay in step with the level layout, and maintained separately the two will drift the first week you start moving rooms. Deriving the map from the layout costs more up front and is the version that survives.

Metroidvanias are large. Not in systems, in rooms. The genre's well-known games run to hundreds of screens, and the feeling of a connected world is a direct product of that count. There is no version of this genre that is small and still feels like the genre.

A starter gives you two or three rooms. The gap between that and a shippable game is a content pipeline: a way to build a room quickly, connect it to its neighbours, place enemies and pickups, and test it from the nearest save point rather than the title screen. Building that early is what makes the rest possible, and no starter contains it because it is specific to how you work.

If that sounds like more than you want to take on, a single-screen action game or a linear platformer is not a lesser goal. It is a different one, and it finishes.

Stage Engine approaches this from the other side, and it is not a template library. There is no picker in the app, and the Darkcom project cannot be opened from inside it. What changes is who writes the code: you say the dash should cancel into a jump, Director 1 writes the GDScript in your project, and you play it and say whether it is right. That loop answers the controller problem better than inheriting somebody's and editing it.

The honest limit is that the hand-built world tooling is aimed at 3D worlds, so a 2D side-scroller uses less of it than an open world would. It is a desktop app for Mac and Windows, the engine underneath is Godot, and your game runs locally on the bundled copy. Building needs an internet connection; playing what you built does not.