appariciojunior/video-editor-skills/music-and-sound

Music and sound design

Only when asked. The creator doesn't want a list to choose from: pick, place, and name two alternates per cue in one line each so they can say "swap to the second one".

Tools

Epidemic Sound as an MCP server: SearchRecordings, EditRecording + PollEditRecordingJob + DownloadRecordingEdit, SearchSoundEffects, DownloadSoundEffect. Downloads are signed URLs: fetch them with curl into system/videos/<v>/music/src/ or sfx/src/.

1. Read the timeline as it is now

The creator has usually trimmed the cut. Read the B-roll windows from V1 of their timeline (times in seconds), and the speech around them.

2. Cues by section (moods)

Read the transcript and split it into 2–3 sections. A vlog that worked:

• Opening / journey (hopeful, dreamy ambient beats ~100 BPM), swelling on the travel B-roll;
• The talk (laid-back lo-fi ~73 BPM, sparse, nothing with a busy lead melody);
• Ending (warm, hopeful indie ~95 BPM) from "to close this video…" through the outro.

Search with filters vocals: false and a BPM range; avoid corporate stock sounds. Then ask Epidemic for an edit at the exact section length (EditRecording, forceDuration: true, WAV, maxResults: 1, max 300 s). Check the edit's loudness over time: one edit started fading 25 s before its end, killing the B-roll it was meant to lift. Fix: ask for a longer edit and cut it yourself with your own short fade.

3. One bed with a baked envelope (on A2)

Normalize each cue to the full level, place them with 2.5–3 s crossfades, then apply a gain envelope:

• Full on B-roll (and other speech-free moments): -16 LUFS. -19 was not enough; the creator asked for louder.
• Under speech: -14 dB below full (≈ -30 LUFS against a -16 LKFS voice).
• Rise 1.2 s, starting ~0.9 s before the B-roll (inside the speaker's tail room) so the music flows into the B-roll; fall 1.0 s ending as the next sentence starts.
• Write a plain 16-bit PCM WAV with a new file name per render (bed_v3.wav) so Resolve never serves a cached old file. Verify levels per section after rendering (speech ≈ -30, B-roll ≈ -15 LUFS).

4. Sound effects: taste over literal

What failed: a zipper when the bag opens, cloth swooshes, a jet pass-by on the plane shot. The creator called it "the most cheesy and weird". Literal foley that mimics objects on screen feels cheap in a reflective vlog. What works:

• One soft swell into each B-roll: an untuned reversed cymbal ("riser, no impact") that peaks exactly on the cut, ~10 dB under the music. Untuned so it never clashes with the music's key.
• Quiet room tone only where it sets the place (airport lounge, a city street), ~ -37 LUFS, long fades.
• The B-roll's own natural sound is real and usually better than library SFX.
• Find where the event happens inside each SFX file (loudness envelope) and align that point, not the file start.
• Pre-mix each scene into one mono clip (A3 can't hold overlapping clips), placed at the scene start.

5. Place and verify

Bed on A2, scene clips on A3 (the creator's own timeline uses the same layout). Re-placing must never bring back clips the creator deleted: place "bed only" or "SFX only" when updating one of them. Then make sure the bed is actually audible (render a few seconds and correlate, see resolve-scripting): a template's music track can be silenced by a track-level AI Music Remixer.

Report

Each cue (section, title - artist, two alternates), the envelope numbers, each SFX scene and why, and that you couldn't listen: "this is a judgement from the files, tell me if it feels off".