multi-cam-editing
Edit multi-camera / shared-audio sessions (podcasts, interviews, panels) inside Palmier Pro — create a multicam group first, verify sync and mic-to-person mapping, then cut angles to the speaker with wide/layout moments for overlap. Use whenever the footage has multiple cameras, multiple mics, dual-system sound, or the user mentions multicam, multi-cam, podcast edit, interview cut, camera switching, or synced audio.
palmier-io · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add palmier-io/palmier-skills --skill multi-cam-editingAbout this skill
name: multi-cam-editing description: Edit multi-camera / shared-audio sessions (podcasts, interviews, panels) inside Palmier Pro — create a multicam group first, verify sync and mic-to-person mapping, then cut angles to the speaker with wide/layout moments for overlap. Use whenever the footage has multiple cameras, multiple mics, dual-system sound, or the user mentions multicam, multi-cam, podcast edit, interview cut, camera switching, or synced audio.
Multi-Cam Editing
Core principle
Always start from a multicam group. Do not manually stack angles and hope they stay aligned — manage_multicam syncs members by audio, locks their timing, and lets change_cam switch what the viewer sees without breaking sync. Editing is deliberate: know whether cameras are speaker angles or A/B coverage, trim with soft landings at sentence ends and tighter joins mid-thought, and put picture cuts on the same seams as audio cuts (especially B-cam). For tight, rapid, or chaotic dialogue, err on the side of not cutting — messy joins hurt more than a few extra seconds.
Workflow checklist
- [ ] 1. Inventory media — cameras, mics, who is on which file
- [ ] 2. Create multicam group (manage_multicam) — never skip this
- [ ] 3. Verify sync + mic ↔ person mapping
- [ ] 4. Agree trim goals with the user (format-specific)
- [ ] 5. Trim / tighten (get_transcript → remove_words / remove_silence)
- [ ] 6. Program the cut (change_cam — speaker cams + wide/layout)
- [ ] 7. Review with get_multicam + inspect_timeline
Step 1: Inventory the session
get_media()
get_timeline()
Classify every source before grouping:
| Role | What it is | kind in manage_multicam |
|---|---|---|
| Close / single | One person framed | angle (or both if that file's audio is their mic) |
| Master wide | Whole set / all participants in frame | angle (label it wide) |
| Dedicated mic | Lav, recorder, mix bus with no useful picture | mic |
| Cam + own audio | Remote podcast / Zoom-style per-person file | both |
Decide what the cameras represent — this drives the whole program cut:
| Setup | Cameras mean | Cut strategy |
|---|---|---|
| Speaker cams | Each angle is a different person (host / guest / panel) | Default to whoever is talking; wide/layout for overlap |
| A-cam / B-cam | Same subject or scene, different angles (hero + alternate) | Stay on A-cam most of the time; use B-cam sparingly where it earns its keep (reaction, emphasis, cover a jump, visual variety on a long hold) |
Label angles for the setup you actually have (host/guest/wide, or a-cam/b-cam). Prefer short, stable labels over filenames.
Use inspect_media (overview + a short dialogue window) when filenames don't make roles obvious.
Step 2: Always create the multicam group first
manage_multicam({
create: {
name: "Session",
master: "host-mic", // or best all-speakers source — see below
members: [
{ mediaRef: WIDE_ID, kind: "angle", angleLabel: "wide" },
{ mediaRef: HOST_ID, kind: "angle", angleLabel: "host" },
{ mediaRef: GUEST_ID, kind: "angle", angleLabel: "guest" },
{ mediaRef: MIC_ID, kind: "mic", angleLabel: "host-mic" },
]
}
})
Member-kind rules (from the tool — follow exactly):
- Separate cameras + mics →
angle+mic - One file per participant with its own audio (remote podcast) →
both - Cameras + one master mic →
angles + onemic - Always include at least one
micorboth. Pureangleaudio is scratch for sync only and never plays — an all-angle group is silent. If camera sound is program audio, mark those membersboth, notangle.
Master: the member whose clock defines group time and whose audio feeds transcript/waveforms. Default is the first mic/both. When every member is both, pass master explicitly — pick the source that hears every speaker best (usually the host).
Create places stamped program clips on the timeline (one program video track + one audio track per mic). Read the response: every member should report a confident offsetSeconds. Weak correlation → pin offsetSeconds (slate/clap) and recreate. sync_clips is refused on multicam clips (already aligned) — only use it on plain clips before grouping. Do not angle-cut on a weak sync.
Step 3: Verify sync and audio ↔ person
get_multicam({ groupId: GROUP_ID })
get_transcript({ clipId: MULTICAM_CLIP_ID, granularity: "segments" })
inspect_timeline({ startFrame: SAMPLE_A, endFrame: SAMPLE_B, maxFrames: 4 })
Check before any creative cutting:
- Sync — clap/slate/first overlapping speech lands together across angles (spot-check 2–3 moments with
inspect_timeline/ short listens). - Mic ↔ person — when person A talks, the waveform/transcript energy matches A's angle; no swapped lavs or host/guest flip.
- Master choice — if a guest mic is clearer for the whole conversation, recreate with that as
masterrather than living with a bad transcript source.
If audio is wrong for a person, fix membership/kinds/master and recreate the group. Do not "fix it in the cut."
Step 4: Agree what to trim (work with the user)
Do not silently gut the recording. Propose a trim plan from the transcript, then confirm:
| Format | Usually remove | Usually keep |
|---|---|---|
| User discovery / interview | Interviewer's questions (when the deliverable is guest-only answers); pre-roll setup; "can you hear me"; retakes of the same answer | Guest answers in full; natural pauses that carry meaning; emotional beats |
| Podcast / panel | Beginning slate, tech checks, "we're recording," ending "ok that's a wrap," long off-topic tangents the hosts already flagged | Banter that defines the show; overlapping laughs; cold open if one exists |
| Either | Long silences, repeated false starts, obvious "start over" takes | Rapid back-and-forth, crosstalk, punchlines th |