Prompt to 3D game: questions first, then the world and the systems

Try writing down the exact shape of a hill you can picture. Height, slope on each side, where it flattens, how the ridge runs, where the trees stop. It takes a paragraph, it will still be wrong, and you would have drawn it in fifteen seconds.

This gets worse the more of a world there is. A valley, a river through it, a road that follows the river, a town where the road crosses — the relationships between those things are the design, and prose is a bad format for relationships in space.

A tool that builds straight from that sentence hands you its guess: a valley, a river and a town, none of them where you meant. Flockbay does not start there. It asks questions one round at a time, each with a recommendation and a line on why, and writes a brief: what gets built first, what comes after, what is not in this version. Nothing is built until you approve it.

More on this: Open worlds are a scale problem before they are a design problem

Once the brief is approved, Flockbay builds it with world tools: terrain sculpted and painted, water, sky, weather, plants, the player controller, characters and people, and props. Each is nodes and resources in your scene tree, so the next change has something real to change.

It writes the systems the same way, as GDScript inside your project: how the player moves and what they can climb, what the wolves do when they smell you, whether the torch burns down. You never have to read it, though it is a normal Godot project and the file is right there.

Then it play-tests its own work. It runs the game, headless or in a window, sends input events, reads stdout and the debugger for GDScript parse errors, and captures the viewport to see what the camera sees.

Then you play it, and this is the part that matters most in 3D: things feel wrong in ways you only notice at the controls. The camera swings too fast, the jump does not clear the gap you thought it would, the enemy sees you through a hedge. Say so, and Flockbay changes it, held against the brief you approved.

Some calls are pure taste: exactly where the ridge runs, which way the river bends. Describing those in a paragraph is slower than dragging them, so the same world tools are buttons around the Godot editor's own 3D view: Sculpt, Paint, Plants, Place, Water, Sky, Weather, Player controller, NPCs and Camera. Move the one hill yourself, and Godot's undo takes it back if you were wrong. It is optional; nothing asks you to touch them.

More on this: An AI game builder that builds the world, and leaves you the controls

Flockbay is a desktop app for Windows, macOS and Linux and it ships with Godot inside. When you press play, your machine renders the game. No streaming, no queue, no browser tab holding a budget over you.

For 3D that is the difference between a scene and a world. Draw distance, terrain resolution, dynamic weather, how many characters can be doing things at once — all of it comes out of the hardware actually running the frame.

Building needs an internet connection. The chat, asset generation and world data all come over the network. Playing what you built does not.

Something rough that runs: the first thing the brief said to build. A stretch of terrain, a player controller that moves the way you asked, one system doing what you described. Not a game yet. Ask for something the size of GTA 6 and you get the truth and a smaller first version, not a fake yes.

Getting from there to a game is iteration, and most of it is you deciding things: this is too big, that is too dark, this route is boring, that fight is too long. Flockbay is fast at acting on those. Having them is the work only you can do.

Free to download, and the world tools are free. Flockbay is free, AI included: free hosted AI every day, or 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. The engine and the export path are free.

More on this: Looking for an AI dungeon master, or a game to put one in?

First worlds are almost always too big. A landscape that looks reasonable from above turns out to take four minutes to cross on foot with nothing happening, and nothing kills a 3D game faster than empty walking.

The fix is to size the world to walking distance rather than to the view. Decide how long it should take to get from one interesting thing to the next, walk it, and ask for the gap to shrink until it stops being a chore.

You will notice this immediately once you play it and never notice it while looking at the editor's 3D view from above. That is a general rule for 3D: the camera you design in is not the camera anyone plays in.

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