Mix Master Agent Skill

mix-master

Purpose

The v03 Ironman LEARNINGS describe 6+ tuning rounds because no default mix protocol existed: round 1 music too loud, round 2 VO too quiet, round 3 Whisper mis-transcribed, round 4 sidechain ratio finally right, round 5 music ended too early, round 6 climax sat at the same loudness as the rest.

This atom encodes the protocol that landed and ships it as the default. Operators tune by flag; they don't redesign the chain.

Inputs

• --video <path> — video file with no audio (or whose audio gets dropped) (required)
• --vo <path,path,...> — comma-separated list of per-scene VO mp3s in scene order (required)
• --vo-starts <ms,ms,...> — start time in milliseconds for each VO clip on the timeline (required, same length as --vo)
• --music <path> — music bed file (required)
• --sfx <path,path,...> — optional comma-separated SFX list
• --sfx-starts <ms,ms,...> — optional SFX start times
• --output <path> — output mp4 (required)
• --total-duration <s> — target total duration in seconds (required — video AND music are sync-fit to this)
• --vo-boost-line <N> — 1-indexed VO clip that is the climax; gets +20% additional volume (default unset)
• --vo-volume <float> — per-clip VO volume multiplier (default 3.0)
• --vo-mix-volume <float> — extra mix-bus volume on VO (default 2.0)
• --music-volume <float> — base music volume (default 0.13)
• --music-swell-volume <float> — peak music volume during apad/swell (default 0.21)
• --sidechain-ratio <float> — sidechain ratio, capped at 20 by FFmpeg (default 20)
• --sidechain-threshold <float> — sidechain threshold (default 0.01)
• --target-i <lufs> — VO loudnorm integrated target (default -16)
• --target-tp <dbtp> — VO loudnorm true-peak (default -1.5)
• --target-lra <lu> — VO loudnorm LRA (default 11)

Workflow

• Validate inputs — --vo and --vo-starts must be same length; --sfx and --sfx-starts must be same length; ffmpeg + ffprobe must be on PATH.
• Probe music duration (ffprobe). If music shorter than --total-duration, append apad=pad_dur=2.0 then trim back to target. If music longer, trim with afade=out over last 1.5s.
• Build a filter_complex_script file (long chains exceed shell arg limits):
• Per-VO: [N:a]loudnorm=I=...:TP=...:LRA=...,adelay=<start>|<start>,volume=<vo_volume>[voN];
• Climax VO (if --vo-boost-line N): volume becomes vo_volume * 1.2.
• Mix all VO: amix=inputs=K:duration=longest,volume=<vo_mix_volume>[vo_pre];
• Split VO for sidechain key: [vo_pre]asplit=2[vo_final][vo_sc];
• Music base: apad=pad_dur=2.0,atrim=0:<TOTAL>,afade=t=out:st=<TOTAL-1.5>:d=1.5,volume=<music_volume>[music_base];
• Sidechain: [music_base][vo_sc]sidechaincompress=threshold=<thresh>:ratio=<ratio>:attack=20:release=600[music_ducked];
• SFX (optional): per-SFX adelay + volume then mix into [sfx_bus].
• Final: amix=inputs=<2 or 3>:duration=longest[a_out].
• Run a single ffmpeg invocation that consumes video, all VO, music, all SFX, applies the filter chain, and outputs the mixed mp4 (video re-encoded only if duration trim needed; otherwise -c:v copy).
• Run a verification probe: ffprobe loudness summary on [a_out], write manifest.json and verification.md.

Output

• <output> — mixed mp4 with VO + music + SFX.
• manifest.json — chain parameters, vo timing, climax line, integrated LUFS after.
• verification.md — pass/fail report (integrated LUFS within ±0.5 of expected target, no clipping).

Quality Checks

• Output integrated LUFS within ±0.5 of <target-i> (post-protocol; should be approximately -14 social-ready after the boosts).
• True-peak strictly below <target-tp> (no clipping).
• Output duration equals <total-duration> ± 100ms.
• Music tail has afade=out — last 100ms RMS < -30 dB.
• If --vo-boost-line was set, the corresponding clip's volume= value is vo_volume * 1.2.

Failure Modes

• Sidechain ratio > 20 — FFmpeg errors "Result too large." Cap at 20 (the protocol default).
• VO and starts arrays mismatched — fail fast at validation.
• Music too short and apad insufficient — emit a warning; video may end before music. Operator should re-source music or shorten --total-duration.
• VO clip > 5KB but actually error JSON — out of scope here; create-clips Phase 4 catches it.
• Re-encoding video unnecessarily — if total duration matches input video duration, use -c:v copy to preserve quality.

Example

scripts/mix.sh \
  --video peloton/ads/video-04-new-job/edits/master-no-audio.mp4 \
  --vo "peloton/ads/video-04-new-job/audio/vo-scene-01.mp3,...,peloton/ads/video-04-new-job/audio/vo-scene-14.mp3" \
  --vo-starts "0,2000,3500,...,28000" \
  --music peloton/ads/video-04-new-job/audio/music-cue-01.wav \
  --output peloton/ads/video-04-new-job/edits/master-final.mp4 \
  --total-duration 30.0 \
  --vo-boost-line 14

Consumed by: video-orchestrator/edit-video Phase 3–5, auto-fix-from-review-notes dispatcher (when vo-intelligibility or music-vo-balance polish-notes items reference mix-master).