# Ambience Bed for Video and Audio

> Put a seamless rain, wind, stream, fire, hum or drone bed under a video or recording, with fades. Generates a procedural ambience (rain, wind, stream, campfire, room tone with hum or an evolving drone) with the shared usefulweb audio engine, loops it seamlessly over the whole clip with a fade-in and fade-out, optionally in a room reverb, and mixes it under the original sound — or replaces the sound. Videos keep their picture untouched (stream copy); recordings come back as WAV. The report lists the bed level, how far it sits under the original and the peak of the mix.

Part of usefulweb.app (https://usefulweb.app/). Part of [Reelcraft](https://reelcraft.usefulweb.app/) (ownerApp: reelcraft). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://reelcraftambiencebed.usefulweb.app/ (interactive tool at https://reelcraftambiencebed.usefulweb.app/tool)
- JSON API: POST https://reelcraftambiencebed.usefulweb.app/api/v1/calculate with a JSON object of inputs (schema: https://reelcraftambiencebed.usefulweb.app/api/v1/tool)
- OpenAPI: https://reelcraftambiencebed.usefulweb.app/openapi.json
- MCP (Streamable HTTP): https://reelcraftambiencebed.usefulweb.app/mcp - tool `reelcraft_ambience_bed`

## Inputs
- file (file, required): Video or audio file - MP4/MOV (H.264, HEVC, AV1), WebM (VP8/VP9/AV1) or an audio file; up to 5 minutes and 250 MB. Processed in a private temporary folder that is deleted afterwards; the result can be downloaded for about 2 hours.
- kind (select): Ambience
- level (number): Bed level (RMS) - Loudness of the bed itself. Speech usually sits around −20 dBFS RMS, so −32 keeps the bed about 12 dB under it.
- fadeIn (number): Fade-in
- fadeOut (number): Fade-out
- room (select): Room reverb on the bed
- mode (select): Original sound
- seed (text): Seed (optional) - Same seed and settings = the same bed every time.

## Method
- ffprobe reads the clip; the first audio track (if any) is decoded by ffmpeg to 48 kHz stereo 32-bit float.
- The bed is a seamless loop (up to 20 s) from the shared audio engine’s procedural ambience models — the same generator as the Ambience Generator tool — at the chosen RMS level, optionally through the engine’s room reverb. It repeats over the clip with linear fades in and out (each at most half the clip).
- The bed is added to the original sound (or replaces it). If the sum comes within 0.2 dB of full scale a soft limiter rounds off only those peaks; the number of affected samples is reported.
- Videos: the picture stream is copied bit for bit and the new sound is encoded as AAC 192 kb/s in MP4 (H.264/HEVC/MPEG-4/AV1) or Opus 160 kb/s in WebM (VP8/VP9). Audio files come back as 16-bit WAV.

## Limits
- The ambiences are synthesized impressions, not recordings of a real place. Check the balance by ear: an RMS level says little about how loud speech and a bed are perceived together.