Video Script
Intelligent short-form & long-form video scripting microservice: turn a topic into scroll-stopping hooks, a full beat-structured script for a target duration, a retention-beat map...
video-script
Intelligent video-scripting microservice. Turns a topic into everything you need to shoot a short- or long-form video: scroll-stopping hooks, a beat-structured script sized to a target duration, a retention-beat timeline, timed captions (SRT), a scene/B-roll list, and call-to-action variants.
Every tool has a deterministic heuristic/template core that runs fully offline with no model. The generative tools (hook, script, cta) also try an optional LLM (via ../_shared/llm.js) and silently fall back to the template when no model is reachable. Results carry mode: 'heuristic' | 'llm'.
Tools
| Tool | Args | Returns |
|---|---|---|
hook | { topic, count?, tone?, audience?, platform? } | ranked hook variants (formula-driven; optional LLM rewrite) |
script | { topic, duration?, wpm?, tone?, platform?, audience? } | beat framework (hook→context→value beats→payoff→CTA) with per-beat word/time budgets + full scriptText |
retentionBeats | { topic, duration? } | pattern-interrupt / open-loop / re-hook timeline + predicted retention curve |
captions | { script, wpm?, options.maxCharsPerCaption? } | timed subtitle cues + a ready .srt string |
sceneList | { topic?, script?, duration?, wpm? } | shot-by-shot scene plan (shot type, narration, text overlay, B-roll) |
cta | { goal?, platform?, count?, audience?, options.product?, options.link? } | ranked call-to-action variants |
duration accepts 60, '60s', '2m', '1m30s'. platform ∈ tiktok|reels|shorts|youtube|ad|explainer.
Example
import pack from './index.js';
const s = await pack.adapters.script({ topic: 'cold email that gets replies', duration: '60s', platform: 'shorts' });
console.log(s.beats.map(b => `${b.label} (${b.durationSec}s): ${b.line}`));
const caps = pack.adapters.captions({ script: s.scriptText, wpm: 150 });
console.log(caps.srt);DRY boundaries
- Not seo (that generates page meta tags / sitemaps / JSON-LD, not spoken video narrative).
- Not a-text (generic string transforms) or chatbots (conversational agents).
- Caption timing here is spoken-word cue timing, distinct from datetime clock math.
- Self-contained: no cross-pack imports except
../_shared/llm.js.