script-to-reellisted
Install: claude install-skill ruturajjena/claude-animate-pack
# Script to reel
Anything the voice says lands on screen on the word that says it. Read the `/hyperframes` skill once if it is installed: this skill already follows its render contract, so you only write a script and a `beats.json`.
Scripts live next to this file in `scripts/`. Call them by absolute path; below, `<skill>` means this skill's folder.
## 1. Project
```bash
node "<skill>/scripts/new.mjs" my-reel
```
Then write `my-reel/script.txt`: one spoken line per line. Short sentences. Spell numbers the way they are said ("six thousand"). 20 to 45 seconds suits most reels, about 2.5 words a second.
## 2. Voice
Pick one, in this order of quality:
1. **Their own recording** (phone voice memo is fine): `node "<skill>/scripts/voice.mjs" my-reel --file take.m4a`
2. **A cloned voice** they already have (ElevenLabs, Voicebox): export the audio and use `--file`.
3. **Free on-device voice** (Kokoro, no account): `node "<skill>/scripts/voice.mjs" my-reel --voice am_michael`. Voices: `npx hyperframes tts --list`. Needs Python with `pip install kokoro-onnx soundfile` once.
Either way it writes `assets/voice.wav` and `transcript.json` (word times, transcribed on the device). Read the transcript before going on. If it misheard a word, the captions still use the script's spelling as long as both have the same number of words.
## 3. Beats
Write `my-reel/beats.json`. One beat per idea, usually one per sentence. A beat starts on the first word of its `at` phrase (searched in order