noise

type: "noise"

Plays generated white noise for the effect's lifetime — no audio file in any resource pack. One voice per effect instance: it opens when the instance starts, follows the animated volume, brightness, jitter, jitter_rate and seed every frame, and closes when the instance ends (or is stopped, faded out, or rebuilt by a replay seek).

Fields

Audio effects also accept the shared definition fields (duration, easing, loop, persistent, fade_ticks, sound). They do not accept screen_layer or order, and they render no geometry.

Field	Type	Default	Meaning
volume	float	0.0	Linear gain, 0..1. Fades with the effect weight, so fade_ticks fades the noise in and out
brightness	float	0.5	Crisp/soft blend, 0..1. 1 is raw white noise, 0 its lowpassed (mellow) copy
jitter	float	0.0	Random-walk step size for the playback-rate and volume wander. 0 is steady noise
jitter_rate	float	1.0	Wander speed multiplier: how fast the walk drifts per audio block (5 ms, so the wander does not depend on the frame rate)
seed	float	0.0	Walk seed. Changing it reseeds and restarts the texture, so a value replays the same wander

All are ordinary animatable params: start/end, keyframes, expr, graphs, setParam and bind all work. volume's neutral is 0.0 (that is what the fade blends toward); the rest have no neutral and are read as authored.

The voice gain each frame is volume × weight × distance × wander, where weight is the effect's fade weight, distance is a linear falloff from the camera (full volume within 8 blocks, silent past 48 blocks), and wander is the gentle volume random walk (jitter scaled, mean-reverted, clamped to 0..2 — silent only while volume or weight is).

Positioning

Like a world overlay, the voice is heard from a world point (mono — there is no stereo panning, only distance). First source wins:

1. a runtime move (sendMove / network MOVE),
2. the first static positions entry (block-centred),
3. the first entity anchor's live position (an untracked entity is silent that frame),
4. the authored pos_x / pos_y / pos_z triple,
5. otherwise the local-player position (interpolated snapshot), so a plain /vfx play vfxweaver:noise follows the viewer.

positions and region work as on Datapack format.

Example

{
    "type": "noise",
    "duration": 100,
    "fade_ticks": 20,
    "params": {
        "volume": 0.4,
        "brightness": 0.8,
        "jitter": 0.05,
        "seed": 7.0
    }
}

Five seconds of bright radio static that fades in and out over one second each. Loop it for rain:

{
    "type": "noise",
    "loop": true,
    "persistent": true,
    "fade_ticks": 40,
    "params": {
        "volume": 0.3,
        "brightness": 0.25,
        "jitter": 0.02,
        "jitter_rate": 0.5,
        "seed": 3.0
    }
}

/vfx play vfxweaver:noise

Honest limits

• The frame rate does not reach the sound. The voices are synthesised on a dedicated audio thread, in fixed 5 ms blocks from that thread's own wall clock, and every line stays about 60 ms ahead of its playhead. A slow frame, a lost frame or an unfocused window therefore cannot starve the line or stretch a ramp - the sound is the same at 30 fps and at 200 fps. A paused game (or a paused replay) is the one thing that stops it: the shared effect clock freezes, the thread notices within a quarter of a second and renders silence until it moves again. A replay seek resyncs silently instead of bursting.

• The vanilla mixer is bypassed. The noise is synthesised into the OS mixer directly, so the game's volume sliders, mute and resource-pack sounds do not affect it — control loudness with the volume param. A machine with no audio device (or a headless JVM) stays silent and logs once instead of failing.

• stopAll cuts immediately. Stopping one instance fades it out over fade_ticks; stopping everything (world exit, /vfx stopall, client shutdown) closes the voices at once, like the aux windows.