Set up the included free local synthetic narration, planning and word-alignment workflow for this Make Video Skill toolkit so I can make new spoken stories. Read references/free-local.md, references/local-third-party.md, requirements-local-apple.txt and scripts/download-local-models.py before acting.

This is the Apple Silicon MLX route tested on macOS with 32 GB memory and Python 3.12. Confirm the actual computer, free disk space and a working Python 3.12. Tell me the roughly 6.7 GB model download estimate plus package/cache/output space before starting. I authorize these free pinned public model and Python-package downloads if this machine fits that documented route and has enough space. On an unsupported or low-resource computer, explain the unverified part and offer my own permitted recorded voice instead. Do not claim broad platform support.

Preserve the production-finishing environment. Create a separate .local-voice environment only if it does not already exist; if it exists, inspect and reuse it safely rather than deleting it. Install the pinned requirements-local-apple.txt there. Use scripts/download-local-models.py --out models to download the included pinned voice, aligner and planner revisions, record their sources and verify the returned revisions. Do not use an unpinned substitute or download any private sales-presentation speaker reference. Check required executables and use the toolkit-local production environment for finishing when needed.

Use the included original stock synthetic voice for a short clearly marked test, save the clean script/audio, align the final spoken words, process pacing, and inspect the resulting word clock. Models can mispronounce or omit words; forced alignment is not proof that every word was said. Ask me to listen to the sample and approve the voice before a long story. Use my own voice reference only if I supply it and confirm permission; do not clone a third party without permission.

When ready, show me how to run the documented make-local.py route into a new project, then how to keep its audio/source editable and render on my computer. Local inference and rendering use no paid API after setup; AI assistant access, hardware, electricity and optional premium media remain separate. Do not buy credits, change subscriptions, publish or share my files.