jamtime

Sunday mornings in the auditorium, our worship band leans on gorgeous pad sounds — chords that swell up behind the vocalists — to create a wash of sound that fills the room without pulling focus. Strums get muted with a volume pedal; what people hear is the harmony, blooming and decaying. A wide and lush sound the blends everything together.

Tonight I built this kind of pad sound in JAMtime in about fifteen minutes. The audio you'll hear in this post is the same patch played through my JAM engine — the same code that runs on the Omega FX1 pedal also runs in this browser tab, so when you play the samples below, you're running the same code that the plugins (and sometime in the next year or so, a physical Omega FX1) is running! I recorded the swells through my Yamaha EG303, captured them in Ableton, and let each iteration of the patch process them. Read along and you'll hear the conversation unfold, mistakes and all.

The wall (Chain A)

The opening prompt was a one-shot. Most days that's how I'll work — fire one prompt that captures the whole vision and tune from there:

I need an effect chain that gives me a huge ambient sound. A wall of sound. Only very slightly detuned chorus (amt Knob 1), with stereo detuning adds presence. It should use a granular effect to give me the 5th and octave, and the next 5th and octave. Those should be controlled with a semi-fast LFO (speed on Knob 2) that makes those higher sounds sparkle. And huge reverb with a 3-4 second tail (tail on Knob 3).

Here's what came back:

The signal flow was right. The granular felt great. But the knobs seemed dead — twisting Knob 1 (chorus depth) did nothing audible, same for Knob 2, and the reverb tail's range was too narrow to hear much difference. Two things were going on. First, I'd told the AI exactly which parameters to bind the knobs to, and it's hard to predict in advance which parameters will actually be expressive on a given patch — sometimes the dial you reach for is the wrong dial. Second, knobs cover a range, and the range itself can miss the interesting territory. A reverb-tail knob sweeping from 1.0s to 1.5s is technically working; it just isn't making any difference you can hear.

The fix is to zoom out. Triple the reverb range. For Knob 2, ask the AI to pick a parameter that's actually expressive on this patch. Same for Knob 1.

Knob 2 now drives spray on the granular layers — grain scatter, focused-shimmer to scattered-sparkle. That's not what I asked for; it was the AI's pick after I told it the original parameter wasn't audible enough. That's a useful workflow note: when a knob feels dead, just say this knob isn't doing anything — find me something better. It will.

The stab (Chain B)

I had the wall. What I wanted next was an upper layer that stabbed — fleeting, glassy, teasing. Not pad. Tiny sparkles that appear and vanish.

Now I need another copy of that, at 2 octaves up. But this one doesn't have a reverb at the end, it has an echo with a 50% decay rate, about 500ms. And we need to somehow make these sounds "stab" instead of flow. These are little sparkles of sound — fleeting, teasing and beautiful.

I notice I'm not being technical here. Stab instead of flow is a feel — I'm describing the experience and letting the AI figure out the mechanism. That's deliberate. I know how to talk about granular shimmer; I don't always know what DSP move turns a flowing texture into a glassy stab.

What it returned: the same upper-octave granular setup but with very short grain sizes (25-40 ms instead of 120-180), an envelope follower with a 5 ms attack and 80 ms release, and — the move I wouldn't have thought to ask for — squaring that envelope to sharpen the gate, so only the bright spike of each pick attack lets audio through. Then a stereo delay at 500 ms / 50% feedback for the trailing ghosts.

That envelope-squaring trick is exactly the kind of move I lean on the AI for. I'd have spent ten minutes searching for the right parameter dance; instead I described what I wanted to hear and got the right answer. (You'll notice this chain ended up needing a hefty gain bump later — more on that in a minute.)

The hole in the middle (Chain C)

I played A and B together. Pretty. But hollow. There was no body — nothing connecting the wide low pad to the high sparkles. A guitar swell still sounded like a guitar swell with shimmer on top. I wanted it to sound like an organ.

So I told the AI exactly that. We're missing that meat in the middle, some glowing effect that makes the guitar sound like an organ. What are we missing? It came back with a name for the gap: sustained harmonic reinforcement. A slow swell envelope so the note appears already sustaining; a clean octave-up unison via FFT pitch-shift to fatten the harmonic stack; a slow chorus for Leslie-like motion. Nothing flashy — warmth, one octave up, blended right into the middle of the spectrum.

That conversational beat — describing a missing thing in plain language and getting back a name for it — is the part of this workflow I'd miss most if it were taken away. The AI didn't just build something; it named the problem first. Yes, I told it. Build it.

Mixing by ear

Three chains, all running. The sound was coming together — but the whole pedal was quiet. Chain C in particular sounded thin against the others. This is the part nobody warns you about with multi-chain patches: it's almost impossible to predict, up front, how multiple effects will combine across the spectrum. The math can be right and the perceived loudness still wrong, because where each chain sits in the frequency range changes how loud it feels against the others.

So — by ear. I asked for four new "tunables" (things that you can assign to knobs): a per-chain gain for A, B, C, and a master. I quickly swapped around the knobs, tweaks, listened, and made adjustments.

Still weak. Same trick as before — zoom the range out. 0-6× this time, with a logarithmic curve so the lower half of the knob travel still gives fine control near unity gain.

Closer, but Chains B and C needed to be much hotter than 1.0 to sit right against A. So the last move was to bake offsets into the TUNABLEs themselves — Chain B's effective gain is now knob + 6.0, Chain C's is knob + 4.0. The knobs still default to 1.0, so the patch lands at the right mix out of the box. Pull them back toward zero if it's too much.

That last one is the keeper. Plug in a guitar with a volume pedal, mute your strums, and let chords bloom into it.

Fifteen minutes

The whole conversation took about fifteen minutes. If I'd built this in a DAW — clicking, dragging, wiring chorus into granular into reverb, twisting plugin knobs to find the right spot, soloing each chain to debug — it would have been hours. The language-first workflow is the value. And when I want to get hands-on, the knobs are right there: three physical dials on the Omega FX1, mapped however I tell the AI to map them. Best of both.

If you want this pad as a starting point, hit the fork icon next to any play button above. You'll land in JAMtime with that exact patch on the bench — tweak the knobs, re-prompt it, make it yours.