AI game maker comparison: the five axes that matter

In a browser tab, or natively on your machine. This is the most predictive single fact about any tool in the category.

A tab constrains object counts, world size, texture budget and framerate, and no amount of cleverness in the tool changes that. Tools in this position steer you toward small, flat scenes whether or not their marketing says so.

Native means the limit is your hardware, which is at least a limit you can see and upgrade. The cost is that it is a download, on specific operating systems, with a machine requirement.

A project in a standard engine format that other software can open, or a document that exists only inside the service that made it.

The second is a completely legitimate product and it is what makes instant sharing possible. It becomes a problem exactly once: when you want to do something the tool does not do, and there is nowhere to take the work.

The test is one question. If this company shut down, or if you wanted to hand the project to a developer, what would you send them?

More on this: What is a no-code game engine?

Assembled from a fixed list of behaviours, wired together as a graph, or written as real script by a model.

The list is fastest to start and has a ceiling shaped exactly like the list. The graph composes further but is programming with a different notation. Written script has the engine's ceiling rather than the tool's, and costs a round or two of iteration to get right.

The useful question to ask any tool: what happens when I need a behaviour nobody anticipated? The answer separates the three cleanly.

More on this: Who writes the code in Flockbay?

The model, or you. Tools that generate levels are faster to a first result and produce something generic, because a landscape has no correctness test and a description only ever gets close.

Tools that make you build it are slower to a first result and produce the place you meant. Which one is right depends entirely on whether the world is part of what you care about.

This axis is the one most buried in marketing, because "generates your entire world" sounds better than "you build the world", and the second one is what people who finish games end up wanting.

More on this: Can AI make a 3D game?

Some tools meter how much of the same thing you get. Others hold whole capabilities behind the paywall. Either can be honest; what decides it is whether the free tier lets you judge the real product, and whether you can try the paid parts before committing.

Watch for the version where the free tier is a demo of a smaller engine. You cannot learn anything about the real product from it, and by the time you find that out you have spent a fortnight.

Runs natively, on Windows, on a bundled copy of Godot. You hold a normal Godot project in a folder. Code is written by Flockbay as real GDScript, with no behaviour list anywhere in the product. The world is built by you, by hand, deliberately.

Which world is the first thing the app asks: 2D or 3D, shown as two pictures, with no default and nothing created until you pick. Neither is a mode of the other and neither is the cut-down one, so every position above holds either way — the same project in the same folder, the same native runtime, the same written GDScript, the same world laid out by you. A flat project trades terrain, water and weather for ground you drag into place and sprites you position by hand.

The platform is free: the whole builder, your own art, play and playtesting. Flockbay is free on your own AI: your Claude or ChatGPT plan, your own agent, or an API key. Flockbay builds from a library of ready-made objects, or with Godot's own shapes. No team or enterprise tier. Driving the app with your own agent over MCP is free too.

The honest costs of that position: it is a download rather than a tab, building needs an internet connection, there is no phone or Chromebook version, and reaching something playable takes an evening rather than ninety seconds.

More on this: Six questions to ask any AI game maker