Magnum Opus Pipeline

The Magnum Opus pipeline turns any raw recording (Zoom call, consultation, interview, demo) into fully produced shorts, longs, and a full-length cut.

The One File

Everything runs through magnum-opus.py at the root of create-video/. It is the single orchestrator — the only thing you run, the only thing launchd points at.

python3 magnum-opus.py <project> --all           # full pipeline
python3 magnum-opus.py <project> --from post      # resume from post-production
python3 magnum-opus.py <project> --assemble       # just build output hierarchy
python3 magnum-opus.py <project> --all --dry-run  # preview without doing anything

Five Stages

#	Stage	What it does	Key scripts called
1	Analyze	7-step analysis: discover segments, detect screen share, classify visuals, trim, highlights, reconcile, build viewer	magnum-opus-pipeline.sh
2	Render	Render all compositions through make-video.sh (which runs normalize-pts + level-audio + audit)	make-video.sh
3	Post	Level audio (acompressor 6:1 + loudnorm -14 LUFS) and audit every render	level-audio.sh, audit-render.sh
4	Assemble	Build the output hierarchy — per-segment folders, long/ and shorts/ subdirectories, duration in filenames	built into magnum-opus.py
5	Deliver	Copy to output dir (default ~/Desktop/OVEN/), send email summary, self-unload LaunchAgent	gmail-notify

Each stage checks what already exists and skips completed work.

Output Hierarchy

OVEN/
└── matt-romero-full-call-2026-06-04/
    ├── source/
    │   ├── shared_screen_with_speaker_view.mp4
    │   ├── audio_only.m4a
    │   └── transcript.vtt
    ├── local-pricing-content/
    │   ├── long/
    │   │   └── content-drip-strategy-long-5m44s.mp4
    │   └── shorts/
    │       └── no-shorts.txt
    ├── crm-automation-mistake/
    │   ├── long/
    │   │   └── crm-import-disaster-long-22m39s.mp4
    │   └── shorts/
    │       ├── klingon-0m54s.mp4
    │       ├── klingon-pip-0m10s.mp4
    │       └── new-employee-1m50s.mp4
    ├── ai-browser-automation/
    │   ├── long/
    │   │   └── ai-browser-automation-long-13m53s.mp4
    │   └── shorts/
    │       ├── green-screen-0m51s.mp4
    │       └── remote-mac-closet-0m36s.mp4
    └── magnum-opus.md

Duration format: XmYYs — always both parts, even for sub-minute (0m28s) or exact minutes (5m00s).

clips.json — Single Source of Truth

clips.json in each project is the source of truth for segments and their composition mappings:

{
  "segments": [
    {
      "slug": "crm-automation-mistake",
      "title": "The CRM Automation Mistake...",
      "start_s": 726,
      "end_s": 2223,
      "compositions": {
        "long": "crm-import-disaster",
        "shorts": ["klingon", "klingon-pip", "new-employee"]
      }
    }
  ]
}

The compositions field maps segment slugs (analysis names) to render slugs (composition names). These often differ — the analysis names the topic, the composition names the video.

Overnight Runs

The LaunchAgent com.cc.pipeline-run fires magnum-opus.py at 01:00:

magnum-opus.py matt-romero-full-call --from post --output-dir ~/Desktop/OVEN

It runs post-production (level + audit), assembles the output hierarchy, delivers to OVEN, sends email, and self-unloads.

Checklist

Open Checklist (Build 5 · 85 steps · 4 phases)

The checklist tracks its own build number. Constraints on how to execute each step live in create-video/CLAUDE.md and magnum-opus.py — the checklist only contains actions and decisions.

Source of truth

The canonical checklist file is create-video/matt-romero-full-call/magnum-opus-working-checklist.html. The copy hosted here is synced manually when the build number bumps.