stitching-multi-cliplisted

# Stitching Multi-Clip When the composition has N≥2 video overlays from independently-generated source files, **keep each clip as its own layer on the timeline.** The editor's playback engine handles seam smoothness — Tier 0 preroll/double-buffer + hold-last-frame at the cut, plus a background seam render-cache (`seam_cache`) that bakes a short MP4 spanning each boundary and plays it across the seam. There is NO clip-boundary glitch to work around, so do NOT pre-concatenate the timeline into one file. Collapsing N overlays into one destroys the non-destructive per-beat editability (re-rolling a single beat, trimming one clip, swapping audio per scene). Concatenation is genuinely needed only at FINAL EXPORT. ## Before you draft the plan — load `voiceover-production` FIRST This skill owns the stitch's CLIP plan (the partition below); **`voiceover-production` owns the stitch's VOICE plan.** Load it **before** you draft the audio for any beat — not after you present a plan. The default for a voiced AI beat is the main character's `reference-to-video` `@Audio1` carry (`generate_audio: true`), NOT laying source audio under the clip as a separate track (the mute-and-overlay anti-pattern). A beat/voice plan drafted without `voiceover-production` (and this skill) loaded is **provisional** — re-derive the audio half here before generating anything. Do not present a plan, then reconcile it to the skills afterward; load first, plan second. ## Partition the source — replace the surrou