Video Editing Pipeline

Core Workflow

1. Ingest → download/copy all source files, probe metadata (duration, resolution, codec)
2. Transcribe → Whisper verbose_json with language=en + context prompt; get word/segment timestamps
3. Analyze → map each source file's content, identify usable segments, classify (talking head / b-roll / ambient)
4. Plan edit → build scene list from transcript segments, respecting sentence boundaries
5. Assemble → create Remotion composition or ffmpeg concat from scene list
6. Render → output at target specs, compress for delivery

Critical Rules (Learned from Feedback)

Multi-source usage

• USE ALL source files. Never build a reel from just one clip when multiple are provided.
• Map every source file's content before planning the edit. Each file gets a content summary.
• Decide per-scene which file provides the best visual AND which provides the best audio.
• Audio and visual sources can differ per scene (split A/V editing).