Repository documentation

Greenfield creative review cut

This is the repository-owned replacement opening, not a reusable cast template. The storyboard contains the dialogue and voice choices. Production source composes illustrated shots, a real public-repository screenshot, and local neural audio into a neutral litai video manifest. The creative brief owns the full-course direction.

The cut uses Kokoro through MLX-Audio on Apple Silicon, with original synthetic voice presets, not a cloned speaker. The course-local dependency file is not inherited and does not add speech libraries to the framework's runtime dependencies. Use an isolated environment beneath _build after reviewing those dependencies. Download the public mlx-community/Kokoro-82M-bf16 model once and retain its exact snapshot. Neither project narration nor credentials are submitted to a speech service.

python docs/courses/preview/prepare.py \
  --output _build/course-review-v2 \
  --font /path/to/Arial.ttf \
  --model /path/to/downloaded/kokoro/snapshot \
  --repository-image /path/to/public-repository-screenshot.png
litai video build _build/course-review-v2/course.json \
  --output _build/course-review-renders \
  --font /path/to/Arial.ttf --narration recorded

The first command uses the isolated neural-production Python; the second uses the framework CLI. Generated audio and visuals remain in the authoring package so a subsequent media rebuild needs neither MLX nor model access. production.json records model snapshot, voices, tool version and storyboard identity.

The installation shots are labeled walkthroughs, not recordings of successful installation. The full replacement still needs captured setup and coding sessions. This review cut is for judging voice quality, dialogue, visual direction and pace; it is not creative approval, a completed course, or release qualification.