jamtime

Most days, when I sit down with JAMtime, I write a careful prompt. I name the components — a compressor here, a short dark reverb there. Or I think structurally — one chain feeds another, this output goes there. Structured. Pre-conceived.

Today I didn't go that route. First, check out the dry sample I was working with - a sweet metal riff deserving of some sick effects. I named a song I like and a sound I like inside it, and hit send.

I want a heavy metal sound, with a lot of growl and fuzz in the low end — but it sounds great, not muddy. Presence with a very subtle bit of stereo detuning into a subtle reverb might help it really stand out in the mix. I'm imagining Black Betty (the new version) — that driving rhythm "oh yeah, alright!"

That's it. No component list. No chain descriptions. Just growl, fuzz, presence, not muddy, and a song reference.

And it came back with a keeper.

If you're into the technical details, it said it built a three-stage saturation chain — two stacked tanh stages doing the chunky lifting, then an asymmetric fuzz on top, with a 65 Hz highpass to kill the subsonic flub that turns metal patches into mush. A short dark room reverb — 0.28 size, heavy damping — to make it sit in a mix. And a barely-there stereo chorus (0.2 Hz, 1.8 ms) for width without wobble.

That great — and I may use some of those idea next time. But for now, I twisted a few knobs to taste and went home happy!

What the model is actually doing

The cool thing isn't that it built a fuzz chain. It's that it built a fitting fuzz chain from words like growl and not muddy and Black Betty. The AI knows English — including the feelings behind certain words — which gives it the ability to translate abstract concepts into the code the JAM engine uses to make DSP software. English → concept → sonic design → JDL, all in one shot.

It doesn't always work this well. Getting lost in translation is a real thing — two people can both say "a gritty sound" and mean two different things, and the model just picks on. Usually that means a back-and-forth: I describe, it builds, I ask for a refinement, it adjusts. That's a common rhythm.

But once in a while, the loose prompt lands. The vibe-words and the song reference give it enough to work from, and the first thing it hands you is the thing you wanted. You twist a knob or two and you're done.

Today was one of those days. Struck gold in 2 minutes flat.

If you want this patch as a starting point, hit the fork icon next to the play button above. You'll land in JAMtime with the patch on the bench — twist the knobs, ask for variations, make it yours.

Want to do this with synths instead? Build your own synth — just say how you want it to work! vi1 is coming, the same chat-driven workflow for synthesizers. Get notified when it lands →