How to build a continuous ambience bed under your AI clips so every cut stops sounding like a cut

You have a scene of six AI clips, and the picture holds together. Then you play it with sound and every cut announces itself. One clip has a faint synthetic hiss, the next is dead silent, the third has a washed-out wash of generated noise that changes character halfway through. Your ear hears each cut as a jump to a new place, even when the image says you have not moved.

The cause is simple. Each generation carries its own audio, or none, and none of it was recorded in the same room. Real films solve this with room tone, a continuous layer of the location's background sound that runs under every shot, whatever the dialogue or cutting does. You can build the same thing in one sitting, and it fixes more than any single effect you add on top.

Strip the native audio first

Mute or delete the audio on every generated clip in the scene. Do not try to blend it. Generated ambience is rarely consistent from clip to clip, and it will fight the bed you are about to build. If a clip has a sound you want, such as a door closing or a footstep, keep that one moment, trim it tight, and put it on its own track.

You are now starting from silence, which is the right place to start.

Build the bed in three layers

Pick one location and treat the whole scene as happening in it. Then stack three layers on separate tracks:

```
Track A1 Room tone low, steady, no events -28 dB to -32 dB
Track A2 Location bed the place itself -24 dB to -28 dB
Track A3 Spot events sparse, one-off sounds -18 dB to -24 dB
```

Room tone is the quiet hum of the space: air handling indoors, wind floor outdoors. Location bed is what tells the viewer where they are, such as distant traffic, a café murmur, or rain on a window. Spot events are the occasional specifics, a car passing or a cup set down, placed by hand and never looped.

The levels above are starting points, not rules. Set room tone so you only notice it when you mute it.

Run it unbroken across the cuts

This is the step that does the work. Lay the room tone and location bed as single long clips that run straight through every cut in the scene. Do not split them at picture edits. Extend them a second or two past the first and last shots of the scene, and fade them over about a second at the scene boundaries.

If your source loop is short, overlap two copies with a one-second crossfade, offset so any repeating feature never lands in the same place twice. Listen for a recognisable event, such as a dog bark, that returns on a clock. Cut it out or pick a cleaner source.

Then add your spot events. Place them near, not on, the cuts, since a sound that starts just before a new shot reads as the world carrying on across the edit.

What you should hear

Before: each cut changes the sonic texture. Silence between clips makes the scene feel like a slideshow, and any generated noise surfaces and drops out on the edit points.

After: the cuts become invisible to the ear. The floor of the sound never moves, so the viewer's sense of place holds even when the camera angle, the lighting or the subject shifts. Dialogue and music, when you add them, sit in a space that already exists.

A quick test is to play the scene with your eyes closed. If you cannot tell where the cuts fall, the bed is working.

Try this next

Take your most recent finished scene, mute every clip, and build the three tracks above in twenty minutes. Export it with and without the bed and play both back to back on headphones. Keep the bed as a template and reuse its structure for the next location.