ComfyUI lab 1: Design sound effects from first principles

A convincing sound effect is a tiny physical story. Something stores energy, something releases it, a material resonates, and a room plus microphone decide what survives. “Sci-fi sound” hides all of those decisions; a useful prompt names them.

The six-field sound model

Write source → action → material → space → capture → envelope. For an airlock: the source is a powered door, the action is a fast close, the materials are heavy steel and hydraulics, the space is a dry corridor, capture is a close microphone, and the envelope is a sharp impact with a short pressure release. Add “no music” when you need editable production audio rather than a cinematic cue.

Duration is part of the design. A UI click may need 150 ms; a door can use two seconds; an ambience needs enough time to reveal its texture. Generating eight seconds and trimming later often makes the event feel vague because the model spends its budget filling time.

A controlled experiment

Keep the action and seed-like request context stable. Change only material: steel, ceramic, dense wood, then soft polymer. Compare transient sharpness, resonant tail, and low-frequency energy. Next keep material stable and change the room. This is the audio equivalent of holding a diffusion seed while sweeping CFG.

Normalize only after choosing a take. Loudness processing cannot repair the wrong event, and a limiter can erase the transient that made the effect readable.

Run the lab

• Open this graph in Comfy Studio
• Download the exact API-format workflow
• CLI: app comfy download appnz-sound-effects

Record the graph JSON, input assets, seed where relevant, model or service version, custom-node commit, and final artifact together. That bundle—not a screenshot—is the reproducible creative unit.

Open-source map

• Open app.nz audio nodes — MIT node used by the downloadable graph
• Stable Audio Open — Open audio model training and inference tools

Creative Systems Labs: 1 of 12.

Next: Build loopable music beds