book-to-media

name	book-to-media
description	Turn a bookkit book into other media — an audiobook (via podcastkit) or a storyboard for a comic/video. Use when the user asks to "make an audiobook", "narrate this book", "do a full-cast reading", "storyboard a chapter", or otherwise render a book project into audio or visuals. Assumes a book directory with book.yaml (+ optional bible.yaml).

book-to-media

bookkit and podcastkit are decoupled sibling CLIs. bookkit owns the canonical source (Markdown chapters, book.yaml, bible.yaml); podcastkit owns audio rendering. They never import each other — you are the orchestrator that chains them. One canon, many rendered media.

The pipeline

book.yaml + chapters/ + bible.yaml          (canonical source — bookkit)
        │
        ├─ bookkit audiobook  ──►  podcastkit project (script.json + episode.yaml)
        │                              │
        │                              └─ podcastkit generate && assemble ──► MP3s
        │
        └─ bookkit storyboard ──►  storyboard.json (panels) ──► comic/video renderer

How to make an audiobook

1. Emit the project (the bible's cast becomes the voice cast automatically):

   bookkit audiobook -b <book-dir> --backend kokoro --voice bm_george
   # full-cast reading (attribute dialogue to characters):
   bookkit audiobook -b <book-dir> --cast --backend openai
   

   Use --dry-run -o json first to report episodes, line counts, and total characters (the cost estimate for paid TTS).
2. Render each chapter with podcastkit:

   for ep in <book-dir>/<slug>-audiobook/chapter_*/; do
     podcastkit generate -e "$ep" && podcastkit assemble -e "$ep"
   done
   

   Or just run scripts/render-media.sh <book-dir> [--cast], which does all of the above.

How to make a storyboard (comic / video script)

bookkit storyboard -b <book-dir>

Produces one storyboard.json per chapter: ordered panels with a scene description, attributed dialogue, the characters present, and art notes pulled from bible.yaml so a character looks the same in every panel. There is no built-in pixel renderer yet — hand the panels to an image/video tool.

Guidance

• Voices & cost. kokoro/chatterbox are local and free (English-centric); openai/elevenlabs are paid per character and handle other languages (e.g. Spanish) far better. Always surface the character count before a paid render.
• Cast placeholders. With a bible.yaml, every character is pre-seeded in episode.yaml with voice_id: REPLACE_ME and a casting note. For a full-cast reading, fill those in before rendering.
• Idempotent & safe. Both commands preserve hand-edited files unless --force is passed, so re-running never clobbers a tuned cast or script.
• Attribution is conservative. --cast only assigns a character voice when an attribution cue ("said X", "—dijo X") clearly names one; otherwise the line stays with the narrator. First-person narration stays mostly narrated — that's correct, not a bug.