Skip to content

CLAUDE.md

Latest commit

 

History

History
52 lines (40 loc) · 2.25 KB

CLAUDE.md

File metadata and controls

52 lines (40 loc) · 2.25 KB
Raw

ClaudeCut — guide for Claude Code

You are the "brain" of this pipeline. When the user hands you a footage folder + a briefing ("make X out of this"), you run the flow below. The scripts do the mechanical part; you make the editorial selection.

Flow

  1. Run prep.py

    python prep.py "<footage folder>" --briefing "<the user's one sentence>"

    This transcribes every clip (Whisper, JSON is cached), measures the metadata and writes transcripts.md + a scaffold cut.py.

  2. Read transcripts.md — the transcripts are in there with clip-relative timecodes in seconds. Make the editorial decision:

    • which soundbites stay, which go (misspeaks, repetitions, weak takes, filler),
    • in what order (thematic: hook → blocks → outro, not necessarily chronological),
    • a reason per choice.

  3. Fill the EDL in cut.py — format (filename, in-second, out-second, reason). Cut on word boundaries (the words[] in the JSON give the exact times). Also fill DROPPED (discarded clips + reason) — it lands in the cut plan.

  4. Check the sequence format. prep.py suggests one. On ambiguity, do NOT decide silently — ask the user:

    • mixed frame rate (e.g. 30 + 60) → which target fps?
    • portrait + landscape mixed → 16:9 or 9:16?
    • 4K footage → deliver in 4K or 1080p? This is their delivery decision. Adjust the suggested value in cut.py (SEQ_FPS, SEQ_W, SEQ_H) if needed.

  5. Show the user the cut plan BEFORE building (their control stop). Only after their OK:

    python build_xml.py

    → cut.xml + cut_plan.md.

  6. Announce the Premiere import: File → Import → cut.xml.

Rules

  • Subtitles come at the very end, freshly from the cut audio — not from the raw transcript.
  • Call Whisper per clip (that's what prep.py does) → timecodes stay clip-relative = EDL-ready.
  • cut.py is per project; do not commit it to the repo (it's in .gitignore).
  • With many/long clips, transcription takes a while — existing .json files are skipped.

When paths are missing

config.py holds the Whisper paths (or CLAUDECUT_WHISPER_PY env). If prep.py fails with "Whisper Python not found" → point the user to docs/SETUP.md.