Audio

All audio is generated locally, generated by a provider with your key, or imported by you with an explicit license flag. demovie never downloads audio. Every file gets an entry in the video's audio/provenance.json.

Music (default: free, deterministic)#

npx demovie audio music <slug> [--mood uplifting|tech|calm|energetic|minimal] [--bpm 80-140] [--key "F#m"] [--seed s]

A pure-TypeScript synthesizer (kick, clap, hats, bass, pads, plucked arpeggios, risers, a limiter) writes audio/music.wav (48 kHz stereo) and audio/beats.json. It follows the storyboard: an intro on the first shot, a build, the main groove, and an ending hit on the final shot with a bar line exactly there. The mood defaults to the style; the BPM to the storyboard's bpm; the seed to the video slug — the same inputs always give the same file.

beats.json ({ bpm, offset, beats[], bars[], sections[] }) powers v.beat(n) and v.bar(n) in compositions.

Alternatives:

--provider elevenlabs — ElevenLabs Music with ELEVENLABS_API_KEY; prints the cost and needs --yes.
Your own licensed track: npx demovie add track.mp3 --licensed --describe "Artlist license #…", set audio.music.src to "assets/track.mp3" in video.json, then npx demovie audio music <slug> --provider file for the beat grid.

SFX (bundled, CC0)#

npx demovie audio sfx --list: whoosh-short, whoosh-long, swipe, click, soft-click, pop, tick, type-key, riser-2s, impact, shimmer, notification. Cue them in video.json:

"audio": { "sfx": [{ "name": "click", "at": 9.85, "gain": -14 }] }

Voiceover (optional)#

bash

npx demovie audio voice <slug> --provider elevenlabs --voice <voice id>      # prints the estimate, asks
npx demovie audio voice <slug> --provider openai --yes                        # after you approved

Lines come from the storyboard's VO column (verbatim) or from --script vo.txt (one line each, optional [12.5] start).
ElevenLabs returns character timings, converted to word timings; OpenAI doesn't, so word timings are estimated and marked estimated.
Results are cached by provider, model, voice and text in .demovie/.cache/voice/.
Keys: ELEVENLABS_API_KEY / OPENAI_API_KEY from the environment or .demovie/.env. Never logged.
Output: audio/vo-*.mp3|wav, audio/voice.json, audio/captions.vtt and .srt.

Captions#

Burned in by the runtime (captions(v, { style })) for the formats in video.json captions.burnIn (default 9:16) when there is VO; render also writes out/captions.vtt and out/captions.srt.

Mix#

npx demovie audio mix <slug>

Music at its gain (default −14 dB), ducked −8 dB under VO (150 ms attack, 400 ms release), faded out over the last second; VO and SFX at their times; then two-pass loudnorm to the target in video.json (default −16 LUFS, −1.5 dBTP) → audio/mix.wav, which render encodes as AAC 192 kbps. QA (DM-S01) measures it.