How to make generative ambient that plays itself

What "generative" actually means

Brian Eno coined the term for music that makes itself from a set of rules. It is not a recording and it is not a loop. You give it some parts, a scale and a little randomness, and it plays itself, never quite the same way twice.

The useful picture is a wind chime. You do not write each note. You build the chime, hang it up, and let the wind play it. Your job is the design of the system. The performance is the system's.

This matters because a held pad is static and a loop repeats until the ear switches off. Generative sits between the two. It is always familiar and never identical, so attention never quite lets go, which is exactly what you want from something meant to play for an hour under a scene or behind work.

The three things that keep it alive

Drift, not repetition

Use layers of different lengths that never line up. Two loops of four bars each lock together and you hear the seam on every pass. Make them seven bars and eleven, or let them free-run with no bar at all, and the combination takes minutes to come round, if it ever does. This is the old tape-loop trick: loops of different lengths, phasing slowly past each other, so the texture is always in a slightly new place.

Randomness inside a fence

Let the machine pick notes at random, but only from one scale or one chord. It cannot play a wrong note because there are no wrong notes available to it. Randomness is only pleasant when it is fenced. Take the fence away and you get noise; draw it tight and the same randomness sounds composed.

Slow evolution

Have one or two things that move over minutes, not seconds. A filter opening across five minutes, a layer fading up over ten. That gives the piece an arc you never had to draw. If a control moves every bar, the ear stops reading it as movement and it just becomes wobble.

Building one by hand

You can make this in any DAW today. It is fiddly, but it is worth doing once so the idea is yours.

Start with harmony, then leave it alone

Pick one chord or one mode and stay there. Generative ambient almost never changes key. The interest is texture and time, not a progression. Everything else hangs off this one decision, so make it a chord you would happily hear for twenty minutes.

Layer three or four slow voices

Put three or four soft voices on notes of that chord, each with a long, slow attack. Give each one a different, unrelated length, or free-run them with no loop point, so they drift out of phase with each other. That drift is most of the magic.

Automate slowly, at different rates

Add slow movement to the filter and the level on each voice, and make the rates different so the balance is always shifting and never syncs up. Long, lazy curves. Nothing that arrives on a downbeat.

Sprinkle notes from the scale

Over the top, drop sparse single notes drawn at random from the same scale, with a long release and wide panning. A generative or random MIDI step does this for you. By hand, write a long line and delete most of it, keeping what falls in the gaps.

The part everyone gets wrong

The urge, once it is running, is to add a part. A melody, a swell, a bit of arrangement. That urge is the enemy here. Generative ambient works because it is under-composed, and every fixed gesture you add is one more thing the ear can latch onto, get bored of, and start waiting for.

• Stay in one register. If the layers spread across five octaves it reads as an arrangement, not a field. Keep them close and let the drift do the moving.
• One harmony is plenty. A second chord is a temptation, not an improvement. If you must move, move once, slowly, and never come back on a schedule.
• Silence counts. The random notes should be sparse. Gaps are where the ear rests, and a wall of sound has none.
• Resist the reverb crutch. A huge reverb hides a boring source. Get the notes and the drift right first, then the space is decoration rather than disguise.