Learning

Procedural Generation · Reference

RandomNumberGenerator cheat sheet

The class you should reach for whenever a procgen algorithm needs randomness. See official docs .

Why not the global randi() / randf()

Godot exposes global random functions (randi(), randf(), randi_range()) that use a single shared, engine-wide PRNG state. That’s fine for one-off effects (particle jitter), but for procgen you want an isolated, seedable stream so one system’s randomness doesn’t perturb another’s, and so you can save/restore exact state. Use an instance of RandomNumberGenerator instead.

Basic usage

var rng := RandomNumberGenerator.new()
rng.seed = 12345          # any int — same seed, same sequence
# rng.randomize()         # use instead of a fixed seed for true randomness

var x := rng.randi_range(0, 9)      # int in [0, 9]
var f := rng.randf()                # float in [0.0, 1.0)
var f2 := rng.randf_range(-1.0, 1.0)
var coin := rng.randi() % 2 == 0    # avoid — use randi_range instead

Reproducing a run

var rng := RandomNumberGenerator.new()
rng.seed = 12345
generate_dungeon(rng)   # pass the SAME rng instance through the whole generator

# Later, to reproduce exactly:
var rng2 := RandomNumberGenerator.new()
rng2.seed = 12345       # same seed →same dungeon, if generation order is unchanged
Gotcha Determinism only holds if you call the RNG in the same order every time. Branching generation order on, say, dictionary iteration (unordered) or Array.shuffle() called from a different RNG can silently break reproducibility.

Common calls

Call Returns
randi() random 32-bit unsigned int, full range
randi_range(from, to) random int, inclusive both ends
randf() random float in [0.0, 1.0)
randf_range(from, to) random float in [from, to)
randomize() reseeds from OS entropy — use for non-reproducible runs