An honest answer about "unlimited" AI game making

Every AI request costs the company running it real money, every time. A tool advertising unlimited AI is either about to change its pricing, silently degrading the model when you use it a lot, or defining "unlimited" against a rate limit written somewhere you have not read yet.

We are not going to claim it. Flockbay is free on your own AI: your Claude or ChatGPT plan, your own agent, or an API key. On your own AI the meter is your provider's, not ours, and it is exactly as generous as the plan or balance you have with them.

The more useful question is not "is it unlimited" but "how much of the work needs the meter at all". That answer is different here than in a tool where every single change goes through a prompt.

More on this: Best free AI game generator: what free usually means

Building the world by hand. In a 3D project that is sculpting terrain, carving water, setting the weather and positioning buildings and characters. In a 2D project it is dragging to say which cells are ground and placing sprites exactly where you want them. Either way it is local work on your machine, done with your hands, and it is the majority of the time you will spend. No credits are consumed by moving a hill, or by painting a ledge.

Playing your game. Flockbay ships with Godot inside it and your game runs locally when you press play. Testing a build twenty times in an evening costs nothing, which matters because testing repeatedly is how a game gets good.

The project itself. There are no per-project limits, no export gate, no watermark on what you made. Your project is a real Godot project sitting in a folder on your disk.

Flockbay. He is the part that writes the GDScript in your project. Each time you ask for a behaviour and he writes or revises code, that uses your AI. Flockbay builds from a library of ready-made objects, or with Godot's own shapes.

This is why the split between hand-building and code generation matters commercially as well as creatively. A tool where the only way to move a rock is to ask an AI to move it will burn through any allowance quickly, and it will also be a worse way to move a rock. Here the prompting is reserved for the part where prompting is genuinely better than clicking: describing behaviour.

Building always needs an internet connection for the same reason. Flockbay, generation and world data are network services. Playing what you already built does not need one.

Be specific. "Make the enemies smarter" spends a request and comes back with a guess. "When the player is behind cover, enemies should flank instead of walking forward" spends the same request and comes back with something you can judge. Vagueness is the main way people waste usage.

Play before you ask again. The instinct is to fire three requests and then look. It is cheaper and faster to change one thing, play it, and know what the change actually did.

Do the buildable parts by hand. If the fix is that a doorway is in the wrong place, drag the doorway. Asking for it is slower and it costs usage you could spend on the parts you cannot do yourself.

A few evenings are enough to find out whether the loop suits you: build some world, ask for some behaviour, play it, judge it. That is the whole product in miniature and you should decide on the basis of it rather than on a feature list.

It suits somebody working on a game regularly — evenings, weekends, a project they intend to finish. A plan runs the AI for you with a month of usage, and on your own AI Flockbay writes your gameplay logic for free. Generation — textures, sounds, props and rigs in a 3D project, and a custom flat image when a 2D one needs a sprite the Catalogue does not have — uses your own keys on Free, and the plan's usage on a plan. There is no team, per-seat or enterprise tier, because this is a desktop app that one person installs and uses.

More on this: An AI game maker with no sign up: Flockbay needs an account, and here is why