Sound Is Half the Shot

The problem

You show someone your AI clip. They say "cool" in the voice people use for a screensaver. You show them the same clip with three layers of sound underneath and they lean in. Nothing changed visually. A silent or music-only clip reads as a rendering. A clip with a believable sound bed reads as a place.

Most AI shorts have one sound layer, usually a music track. Real films have three at all times, and you can build them in fifteen minutes.

The three layers

1. Room tone or ambience. The continuous bed of the location. Rain on asphalt, diner hum with a fridge, wind on a rooftop, night traffic three streets away. It never stops, and it is quiet enough that you stop noticing it, which is the point.
2. Spot effects. Sounds tied to things you can see happening: a cup set down, a door, footsteps, a lighter, a coat rustle. These are what make the image feel physical. Two or three per shot is plenty.
3. Music or drone. The emotional layer. Often the only layer people add, and the least important for believability. Keep it under the other two, not on top of them.

Steps

1. Watch the clip muted and write down every sound you would hear if you were standing there. Be boring. "Rain. Distant car. Her coat. The bus brakes." That list is your spotting sheet.
2. Generate the ambience first. In ElevenLabs Sound Effects, prompt the bed: steady night rain on asphalt with distant city traffic, no music, seamless loop. Ask for 20 to 30 seconds and loop it under the whole scene. Set it low, around minus 24 to minus 20 dB.
3. Generate the spot effects one at a time. single ceramic cup set down on a wooden counter, close, dry room or city bus air brakes hissing to a stop, medium distance, wet street. Short, specific, one sound per prompt. Place each exactly on the frame where the action happens.
4. Add music last, and lower than you think. If you can hear the melody clearly over the ambience during dialogue, it is too loud. Try minus 18 dB under dialogue and let it come up only where there is no speech.
5. Cut sound across the picture cut. Let the ambience run through edits without a break. When the picture cuts to the reverse angle, the rain should not restart. This one habit makes five clips feel like one scene.
6. Listen on headphones and on a phone speaker. If the spot effects vanish on the phone, raise them a little. That is where most people will watch.

Worked example

Clip: 6 seconds, woman at a night bus stop, bus arrives.

Spotting sheet: rain (bed), distant traffic (bed), her coat shifting (spot, 0:01), bus approaching and brakes (spot, 0:03 to 0:05), doors opening (spot, 0:05), low string drone (music, under everything).

Six sounds. About ten minutes of work. The clip goes from "AI demo" to "scene".