Random game idea generator, and what to do the moment after

A generated combination — a fishing game about grief, a stealth game where you are the building — is genuinely useful as a jolt. It gets you out of the ruts you already have and it is a decent way to start a game jam.

The problem is that an idea has no quality until it is played. "Time slows when you stand still" sounds brilliant and may be tedious. "You can only move one square per hour of real time" sounds terrible and might be the best thing you make. There is no way to tell from the sentence.

So the generator is not the bottleneck. Getting to the point where you can tell is the bottleneck, and traditionally that took a week of code.

In Flockbay you describe the core mechanic. It asks what it needs to, one round at a time with a recommendation each, writes a short brief, and once you approve it, builds the mechanic and writes the GDScript into your project. It runs the result itself first. Then you press play, on your own machine, with the game running locally on the bundled Godot.

You will usually know within about ten minutes whether the idea has anything in it. That is the whole value. Nine ideas out of ten die at that point, which is exactly what should happen to them, and dying cheaply is what makes the tenth affordable.

Flockbay builds the world around the mechanic too, 2D or 3D, as real Godot nodes. A 3D idea gets terrain, water, weather and buildings, wherever it needs to live. For a first test all of that can be crude on purpose, and the brief should say so: what gets built first, and what is not in this version.

A flat idea is cheaper still. Tile ground, a band of scenery along the horizon, a ledge and a character who walks, falls and jumps is about the smallest playable thing there is, which on a page about killing ideas cheaply is most of the argument: you are arguing with something on the screen before you have decided what the game is. If the ledge is in the wrong place, Paint in the Godot editor moves it in a few strokes, and Godot's undo takes it back.

More on this: AI sandbox game maker: systems that surprise their author

Strip it to the one thing that is interesting and build only that. No menus, no art pass, no title screen. If the idea is about a rope, make a rope and nothing else.

Play it yourself, then have someone else play it without explaining it first. Their confusion is data. If they need the explanation to enjoy it, the idea is not carrying itself yet.

Then either sharpen it or throw it out. Being willing to throw out something you spent two hours on is much easier than throwing out something you spent two weeks on, which is the real argument for making the two hours possible.

A Godot project on your disk, which is a real one rather than a file only this tool understands. If a prototype turns out to be worth years, it is not stranded in a sandbox.

Flockbay is a desktop app for Windows, macOS and Linux. Building needs an internet connection, because the chat, asset generation and world data all come over the network; playing what you built does not. Flockbay is free, AI included: free hosted AI every day, or your Claude or ChatGPT plan, your own agent, or an API key.

The generator is not trying to hand you a good idea and neither is anyone else. What it does is widen the pool, and a wider pool means a better best item even though the average gets worse.

So treat volume as the strategy. Ten prototypes that die in an afternoon each is a better month than one prototype you nursed for four weeks because you were too invested to admit it was dull.

Keep a note of the ones that half worked. A mechanic that was boring on its own is sometimes exactly what another idea was missing, and you will not remember it in six months if you did not write it down.

More on this: Best AI game generator, and the difference between a demo and a game