An AI game maker built on Godot, not instead of it

The first is a chat box that emits a small web game. It is genuinely impressive for about ten minutes, and then you want to change something structural and discover there is nowhere to stand. The game exists only as whatever the model last produced.

The second is a plugin bolted onto an existing engine, which assumes you already know the engine. That is useful if you do and useless if you do not, because the plugin cannot help with the part where you did not know what a signal was.

Stage Engine is neither. It is a full desktop application with Godot bundled inside it, and it assumes you have never opened an engine before but do have an opinion about games.

You build the world with your hands. Terrain, water, weather, buildings, props, characters — placed in a builder that works the way a level editor works, because arranging a space is a spatial job and describing a space in a sentence is slow and lossy.

Director 1 writes the code. You say what should happen and it writes GDScript inside your project. Inventory, doors, enemy behaviour, scoring, whatever the game needs. You never open a script unless you want to.

Then you play it and say whether it is right. That handing-back is the normal loop. Not writing the code does not mean not judging the result, and the tool is built on the assumption that you are the one with the taste.

Godot is open source and real games ship on it. When Stage Engine makes something, it makes a Godot project: scenes, nodes, scripts, the actual thing. Nothing is held in a proprietary format that stops existing if we do.

It also means the ceiling is Godot's ceiling rather than a menu of supported behaviours. There is no point at which the tool says the feature you want is not on the list, because the code is being written rather than selected.

And it means performance is Godot performance, running natively on your machine, which is what makes an actual 3D world with terrain and weather a reasonable thing to attempt rather than a thing you gesture at.

The download is free and there is a free plan. Indie is $10 a month, Pro is $50. What paid plans buy is more usage — more work through Director 1, more generation — not a different engine or unlocked features.

It runs on Mac and Windows, both shipping. It needs an internet connection to build, because Director 1, asset generation and world data all come over the network. The game you build runs locally, so playing it does not.

There is no team or seat pricing, because it is a desktop app and the licence is per person.

It suits someone with a specific game in mind who does not want to spend six months learning an engine first. It suits people who will happily spend an evening moving rocks around because the ridge is wrong.

It suits people less well if what they want is a finished game in twenty minutes. Nothing does that. If the appeal was that AI removes the work, the honest answer is that it removes the coding and leaves the game design, which was always the larger half.

Generation, the web, or you. There is no fixed catalogue built into the app, and that is deliberate: a catalogue means every game made with the tool ends up containing the same crate and the same tree.

Some things are better built directly as geometry, and that happens too — when it is the fastest honest way to get something playable in front of you, or when simple shapes are the style you want. Whichever route a thing came from, you are told which.

Terrain, water and weather come from generation and world data rather than a preset list of environments, then you shape them by hand in the builder until the place is the place you meant.