JimLiu/baocut / skills/baocut/built-in-skills/production-scripts
production-scripts
Verified有命令行环境、旁白按段配、画面按段上时间线的片子(讲解、解说混剪、旁白片、歌词视频……)要跑脚本时用:配音、对齐字时间、把场景代码打成代码包、按段摆上时间线并处理变长变短的波纹、按稿子建字幕、剪实拍与水印叠上去或垫在底下、量画面占比。脚本是一条省事的路,不规定片子的形态;形态和它不合时用 baocut CLI 自己写。只是加一段动效图形或配一句旁白时不用。
| File | Kind | Size |
|---|---|---|
| .baocut-meta.json | config | 2.8 KB |
| SKILL.md | skill md | 13.4 KB |
| references/pipeline.md | markdown | 37.3 KB |
| scene-kit/geo110.js | script | 27.3 KB |
| scene-kit/scene-kit.js | script | 38.3 KB |
| scripts/align.mjs | script | 11.6 KB |
| scripts/assemble.mjs | script | 26.4 KB |
| scripts/build.mjs | script | 18.1 KB |
| scripts/captions.mjs | script | 18.6 KB |
| scripts/clips.example.json | config | 3.2 KB |
| scripts/coverage.mjs | script | 10.0 KB |
| scripts/envelope.mjs | script | 2.9 KB |
| scripts/explainer.example.json | config | 2.2 KB |
| scripts/fonts.mjs | script | 7.0 KB |
| scripts/frames.mjs | script | 7.7 KB |
| scripts/gaps.mjs | script | 11.3 KB |
| scripts/host.mjs | script | 14.8 KB |
| scripts/legibility.mjs | script | 5.7 KB |
| scripts/level.mjs | script | 9.2 KB |
| scripts/lib.mjs | script | 19.3 KB |
| scripts/overlay.mjs | script | 34.7 KB |
| scripts/preview.mjs | script | 4.6 KB |
| scripts/quotes.mjs | script | 7.4 KB |
| scripts/sources.mjs | script | 6.2 KB |
| scripts/srt.mjs | script | 5.8 KB |
| scripts/voice.mjs | script | 7.7 KB |
scripts/captions.mjs
View on GitHub#!/usr/bin/env node
// i18n-ignore-file: 给智能体拷走用的独立脚本
// 字幕:用稿子原文 + 对齐好的字时间(timing/<id>.json)给每段旁白素材写一份转写(ElevenLabs Scribe 形状),
// documents put 导入后 captions create 建字幕层。用稿子原文而不是识别出的字,专名不会写错。
// node tools/captions.mjs [--only=id,id] [--dry] [--config <配置>]
// --dry 只打印每段切出的词与前几个词的时间,不连 Runtime。
// 结果记在 captions.json(timeline.json 归 assemble.mjs 写)。某段的旁白片段换了(assemble 重放了旁白)或字时间变了,
// 先把所有这样的段的旧字幕片段一起删掉,再逐段重建:逐段「删一段建一段」时,前一段变长后的新字幕会撞上后一段还没删的旧字幕,
// Runtime 就把它放进新建的另一条字幕轨。建好后字幕不在已有的字幕轨上时报出来。
// 字幕默认比每条的第一个字提前一点出现(字幕样式的 displayTiming.leadIn,默认 0.5 秒):从第 0 帧起的那段,
// 第一条会压在封面上,所以建好后把那个字幕片段的起点裁到第一个字被说到的时刻(整帧,向后取)。
// 字幕默认在每条说完后再停 1 秒(displayTiming.tail)。旁白停下来放原声片段(clips.json 里 gap 为 true)时,这一秒会压在
// 片段开头烧进去的字幕上,所以建好后按留白收短:tail 不超过任一留白里旁白结尾到片段开始的距离,leadIn 不超过片段结尾到
// 下一句开口的距离(各少一帧)。配置 captions.displayTiming 可以另给 { leadIn, tail },同样会被收短。
// 配置 captions.style 给旁白字幕的摆放与外观(Studio 样式的键:y、verticalAlign、x、width、fontSize 等,竖版要给,
// 默认的 y 86 落在平台控件区里)。找视频里旁白的字幕样式文档(overlay.json 里原声字幕的样式不算;没有就以默认样式加上
// style 建一份),把各段字幕挂到这份样式上,再只把和配置不同的键用 setCaptionStyle 写进去(都一样时不动);显示时长
// displayTiming 不是 setCaptionStyle 的键,用 patchDocument 写。用的样式记在 captions.json 的 _style,各段记 style
// (重建的字幕是默认样式,记录里 style 不是这份的挂一遍)。
// 旁白字幕不做逐字高亮:引擎的默认样式会把念到的字染成青色,所以什么都不给也要建样式文档,写 wordAnimation none。
// 要 KTV 时给 captions.style.sungColor(#RRGGBB):念过的部分从字色(fontColor,没念到的颜色)渐变成 sungColor,
// 写成 setCaptionStyle 的 textMotion.karaoke.color(Runtime 同时把当前词换成 none,textMotion 里原有的入场、出场保留)。
// 两个颜色都从这期的色板取:浅底上 fontColor 取浅一档、sungColor 取墨色;深底上 fontColor 取暗一点的中间调、
// sungColor 取更亮或更暖的强调色。
// 默认样式的黑描边与黑投影是给实拍上的白字用的:fontColor 或 sungColor 是深色(和黑色的对比度不到 4.5:1,浅底上取的
// 墨色就是)时,黑边会把字糊成一团。这时配置里没写 textOutline、dropShadow 的,脚本按关掉算({ on: false });
// 要描边就在配置里写出来(例如浅色描边)。
import fs from 'node:fs';
import { fileURLToPath } from 'node:url';
import { loadConfig, parseArgs, listFlag, makeBc, readJson, writeJson, isCjk, textHash, allItems } from './lib.mjs';
const OPEN = /[“‘(《「『〈【([{¿¡]/u;
const WORD = c => /[\p{L}\p{N}]/u.test(c) && !isCjk(c);
const JOIN = /['’.,%/+\-_:]/;
// 一句稿子切成词:汉字与假名一个字一个词;其余文字按词(字母数字连着,中间夹的 ' . , % / + - _ : 后面还跟字母数字时算词内);
// 左引号、左括号贴到后一个词,其余标点贴到前一个词(前面隔着空格的独立标点,例如破折号,连空格一起贴);空白记为 spacing。
// 时间取词首字的对齐时刻,结尾取下一个词的开始(最后一个取这一句的结尾),全程不往回走(floor 是上一句最后一个词的结尾)。
export function words(line, floor = 0) {
const s = line.text, out = [];
let lead = '';
const last = () => out.findLast(w => w.type === 'word');
for (let i = 0; i < s.length;) {
const c = s[i];
if (/\s/.test(c)) { if (out.length && out.at(-1).type !== 'spacing') out.push({ type: 'spacing', text: ' ' }); i++; continue; }
if (OPEN.test(c) || (/["']/.test(c) && (i === 0 || /\s/.test(s[i - 1])))) { lead += c; i++; continue; } // 直引号在词首算左引号
if (isCjk(c) || WORD(c)) {
let j = i + 1;
if (WORD(c)) while (j < s.length && (WORD(s[j]) || (JOIN.test(s[j]) && j + 1 < s.length && WORD(s[j + 1])))) j++;
out.push({ type: 'word', text: lead + s.slice(i, j), start: line.chars[i] });
lead = '';
i = j;
continue;
}
// 标点与符号
const prev = last();
if (!prev) { lead += c; i++; continue; }
if (out.at(-1).type === 'spacing') { out.pop(); prev.text += ' ' + c; } else prev.text += c;
i++;
}
while (out.length && out.at(-1).type === 'spacing') out.pop();
const ws = out.filter(w => w.type === 'word');
let t = floor;
for (const w of ws) w.start = t = Math.max(t, w.start);
ws.forEach((w, k) => {
const next = ws[k + 1] ? ws[k + 1].start : Math.max(line.end, w.start);
w.end = +Math.max(w.start, next).toFixed(3);
w.start = +w.start.toFixed(3);
});
return out;
}
const TIMING = { leadIn: 0.5, tail: 1 };
// 留白给字幕显示时长的上限(秒):tail 是旁白结尾到原声片段开始,leadIn 是片段结尾到下一句开口,取所有留白里最小的
export function timingLimits(cfg) {
const cl = readJson(cfg.abs(cfg.footage?.clips || 'clips.json'), { clips: [] });
const len = new Map(cl.clips.filter(c => c.gap).map(c => [c.id, c.to - c.from]));
const lim = { leadIn: Infinity, tail: Infinity };
for (const id of cfg.order) {
const tm = readJson(cfg.abs(cfg.dirs.timing, id + '.json'), { lines: [] });
for (const g of tm.gaps || []) {
if (!len.has(g.clip)) continue;
const end = g.at + len.get(g.clip);
const before = tm.lines.findLast(l => l.end <= g.at + 1e-3), after = tm.lines.find(l => l.start >= end - 1e-3);
if (before) lim.tail = Math.min(lim.tail, g.at - before.end);
if (after) lim.leadIn = Math.min(lim.leadIn, after.start - end);
}
}
return lim;
}
// 旁白字幕能配的样式键:setCaptionStyle 的 style 认的键,加上 sungColor(KTV 念过部分的颜色,写成 textMotion.karaoke);
// 描边、投影只比、只改给出的字段
const STYLE_KEYS = ['y', 'verticalAlign', 'x', 'width', 'fontSize', 'fontWeight', 'fontFamily', 'fontColor', 'textOutline', 'dropShadow', 'background', 'backgroundColor', 'sungColor'];
const NESTED = new Set(['textOutline', 'dropShadow']);
const isObj = v => v && typeof v === 'object' && !Array.isArray(v);
// 配置里旁白字幕的样式:captions.style,没给时 null。字色或扫色是深色时补上关掉的描边与投影(darkEdges)
export function styleOf(cfg) {
const s = cfg.captions?.style;
if (s == null) return null;
if (!isObj(s) || !Object.keys(s).length) throw new Error(`captions.style: give an object with some of ${STYLE_KEYS.join(', ')}`);
if ('activeColor' in s) throw new Error('captions.style.activeColor is gone: narration captions no longer highlight the current word; give sungColor for a KTV sweep (the sung part turns from fontColor to sungColor)');
const bad = Object.keys(s).filter(k => !STYLE_KEYS.includes(k));
if (bad.length) throw new Error(`captions.style: unknown key ${bad.join(', ')}; use ${STYLE_KEYS.join(', ')}`);
if ('sungColor' in s && !/^#(?:[0-9a-fA-F]{6}|[0-9a-fA-F]{8})$/.test(s.sungColor)) throw new Error(`captions.style.sungColor: give #RRGGBB or #RRGGBBAA, not ${JSON.stringify(s.sungColor)}`);
return { ...s, ...darkEdges(s) };
}
// 和黑色的 WCAG 对比度;认不出的颜色给 null。只看 RGB,不看透明度
export function contrastWithBlack(color) {
const m = typeof color === 'string' && color.trim().match(/^#([0-9a-f]{3}|[0-9a-f]{6}|[0-9a-f]{8})$/i);
if (!m) return null;
const h = m[1].length === 3 ? m[1].replace(/./g, c => c + c) : m[1].slice(0, 6);
const lin = v => (v <= 0.03928 ? v / 12.92 : ((v + 0.055) / 1.055) ** 2.4);
const [r, g, b] = [0, 2, 4].map(i => lin(parseInt(h.slice(i, i + 2), 16) / 255));
return (0.2126 * r + 0.7152 * g + 0.0722 * b + 0.05) / 0.05;
}
// 默认样式的黑描边、黑投影压在深色字上会糊:fontColor 或 sungColor 和黑色的对比度不到 4.5:1 时,配置里没写的
// textOutline、dropShadow 记成关掉。返回要补的键(不用补时为空对象)
export function darkEdges(s) {
const dark = ['fontColor', 'sungColor'].some(k => (contrastWithBlack(s[k]) ?? Infinity) < 4.5);
if (!dark) return {};
const out = {};
for (const k of ['textOutline', 'dropShadow']) if (!(k in s)) out[k] = { on: false };
return out;
}
// 当前词效果的 setCaptionStyle 键:不逐字高亮;给了 sungColor 时开 KTV 扫色(Runtime 把当前词一起换成 none)
export function activeWordPatch(want) {
return want?.sungColor ? { textMotion: { karaoke: { color: want.sungColor } } } : { wordAnimation: { catalogId: 'none' } };
}
// 样式文档的 style 和配置不同的键,写成 setCaptionStyle 的 style 补丁;都一样时为空对象。描边、投影只给配置里的字段,
// setCaptionStyle 只改给出的字段。当前词:没给 sungColor 时要是 none 且没有扫色,给了时要是这个颜色的扫色
export function styleChanges(cur, want) {
const out = {};
for (const [k, v] of Object.entries(want)) {
if (k === 'sungColor') continue;
if (NESTED.has(k) && isObj(v)) {
const base = isObj(cur?.[k]) ? cur[k] : {};
if (!Object.entries(v).every(([kk, vv]) => base[kk] === vv)) out[k] = v;
} else if (cur?.[k] !== v) out[k] = v;
}
const tm = isObj(cur?.textMotion) ? cur.textMotion : {};
const none = isObj(cur?.wordAnimation) && cur.wordAnimation.catalogId === 'none';
const settled = none && !tm.emphasis && (want.sungColor ? tm.karaoke?.color === want.sungColor : !tm.karaoke);
return settled ? out : { ...out, ...activeWordPatch(want) };
}
export function applyTiming(cfg, bc, caps) {
const lim = timingLimits(cfg), want = { ...TIMING, ...cfg.captions?.displayTiming }, style = styleOf(cfg);
for (const k of ['leadIn', 'tail']) {
const max = Math.max(0, Math.floor((lim[k] - 1 / cfg.fps) * 20) / 20);
if (want[k] > max) { console.log(`caption ${k} ${want[k]}s -> ${max}s: a sound-bite gap leaves ${lim[k].toFixed(2)}s`); want[k] = max; }
}
const segs = cfg.order.filter(id => caps[id]?.items?.length);
if (!segs.length) return;
const head = () => bc(['videos', 'inspect', cfg.video, '--limit', '1']);
const apply = (label, ops) => bc(['edits', 'apply', cfg.video, '--expected-revision', String(head().revision), '--label', label, '--operations', JSON.stringify(ops)]);
// 原声字幕(overlay.mjs 建的)有自己的样式文档,不是旁白的
const bites = new Set(Object.values(readJson(cfg.abs('overlay.json'), {}).bites || {}).map(b => b.styleDocumentId));
// 不是本脚本建的字幕在用的样式(前插项目里正片的字幕)也不是旁白的
const ours = new Set(segs.flatMap(id => caps[id].items));
const foreign = new Set(allItems(bc, cfg.video).items.filter(it => it.type === 'caption' && !ours.has(it.id) && it.styleDocumentId).map(it => it.styleDocumentId));
const styles = (head().documents || []).filter(d => d.kind === 'caption-style' && !bites.has(d.id) && !foreign.has(d.id));
let doc = styles.find(d => d.id === caps._style?.documentId)?.id ?? (styles.length === 1 ? styles[0].id : null);
if (!doc && styles.length) throw new Error(`video has ${styles.length} caption styles (${styles.map(d => d.id).join(', ')}); put the narration one in captions.json as _style.documentId`);
if (!doc) {
// 以默认样式加上配置的 style 与当前词效果建一份,序列上还没有样式的字幕一起挂上
const { sungColor: _sung, ...first } = style ?? {};
const r = apply('Caption style for narration', [{ type: 'setCaptionStyle', itemId: caps[segs[0]].items[0], style: { ...first, ...activeWordPatch(style) } }]);
doc = r.createdIds?.find(x => x.startsWith('doc_'));
if (!doc) throw new Error(`no style document in receipt: ${JSON.stringify(r).slice(0, 400)}`);
for (const id of segs) caps[id].style = doc;
console.log('caption style', doc, 'created', JSON.stringify(style ?? {}));
}
// 先把各段挂上这份样式:下面的 setCaptionStyle 合进字幕实例自己的样式文档,没挂上的会另建一份
const loose = segs.filter(id => caps[id].style !== doc);
if (loose.length) {
apply(`Attach caption style ${loose.join(' ')}`, loose.flatMap(id => caps[id].items.map(itemId => ({ type: 'setCaptionStyle', itemId, styleDocumentId: doc }))));
for (const id of loose) caps[id].style = doc;
console.log(loose.join(' '), 'captions use style', doc);
}
const cur = bc(['documents', 'read', cfg.video, '--document-id', doc]).body?.style ?? {};
const patch = styleChanges(cur, style ?? {});
if (Object.keys(patch).length) {
apply('Caption style for narration', [{ type: 'setCaptionStyle', itemId: caps[segs[0]].items[0], style: patch }]);
console.log('caption style', doc, JSON.stringify(patch));
}
const t = cur.displayTiming ?? {};
if (t.leadIn !== want.leadIn || t.tail !== want.tail) {
apply('Caption timing for narration', [{ type: 'patchDocument', documentId: doc, changes: [{ path: ['style', 'displayTiming'], value: { ...t, ...want } }] }]);
console.log('caption timing', doc, JSON.stringify(want));
}
caps._style = { documentId: doc, displayTiming: want, ...(style ? { style } : {}) };
}
export function scribe(lines, language) {
const ws = [];
let floor = 0;
for (const l of lines) {
const lw = words(l, floor);
const lastWord = lw.findLast(w => w.type === 'word');
if (lastWord) floor = lastWord.end;
if (ws.length && lw.length) ws.push({ type: 'spacing', text: ' ' });
for (const w of lw) ws.push(w.type === 'word' ? { ...w, speaker_id: 'narrator' } : w);
}
return { language_code: language.split('-')[0].toLowerCase(), words: ws };
}
if (process.argv[1] === fileURLToPath(import.meta.url)) {
const cfg = loadConfig();
const { flags } = parseArgs(process.argv, ['only']);
const only = listFlag(flags.only);
const log = cfg.abs('captions.json');
const caps = readJson(log, {});
const bc = flags.dry ? null : makeBc(cfg);
const segs = readJson(cfg.abs('timeline.json'), {}).segments || {};
const style = styleOf(cfg);
const plain = style ? Object.keys(darkEdges(cfg.captions.style)) : [];
if (plain.length) console.log(`caption text is dark, so ${plain.join(' and ')} off: the default black edges smear it (set them in captions.style to keep an edge)`);
if (flags.dry && style) console.log('caption style', JSON.stringify(style));
// 1. 每段的字与旁白片段,和记录对一下,找出要重建的段
const todo = [];
for (const id of cfg.order) {
if (only.length && !only.includes(id)) continue;
const tf = cfg.abs(cfg.dirs.timing, id + '.json');
if (!fs.existsSync(tf)) throw new Error(`${tf} missing; run align.mjs ${id}`);
const tm = JSON.parse(fs.readFileSync(tf, 'utf8'));
const body = scribe(tm.lines, cfg.language);
const wc = body.words.filter(w => w.type === 'word');
if (flags.dry) {
console.log(id, wc.length, 'words', wc.slice(0, 10).map(w => `${w.text}@${w.start}-${w.end}`).join(' '));
continue;
}
const seg = segs[id];
if (!seg || !seg.audioAsset) throw new Error(`${id} narration not on the timeline yet; run assemble.mjs first`);
const hash = textHash(JSON.stringify(body));
const cap = caps[id];
if (cap && !cap.audioItem) { cap.audioItem = seg.audioItem; cap.wordsHash ??= hash; } // 旧记录:当作和现在一致
todo.push({ id, body, wc, seg, hash, stale: !!cap && (cap.audioItem !== seg.audioItem || cap.wordsHash !== hash) });
}
// 2. 要重建的段,旧字幕片段一起删掉(先记下字幕原来在哪条轨上)
const home = Object.values(caps).find(c => c.track)?.track;
const stale = todo.filter(t => t.stale);
if (stale.length) {
const v = allItems(bc, cfg.video);
const live = new Set(v.items.map(it => it.id));
const still = stale.flatMap(t => (caps[t.id].items || []).filter(itemId => live.has(itemId)));
if (still.length) bc(['edits', 'apply', cfg.video, '--expected-revision', String(v.revision), '--label', `Remove stale captions ${stale.map(t => t.id).join(' ')}`, '--operations', JSON.stringify([{ type: 'deleteItems', itemIds: still }])]);
console.log(stale.map(t => t.id).join(' '), 'stale captions removed', still.length);
for (const t of stale) delete caps[t.id];
writeJson(log, caps);
}
// 3. 逐段建字幕
for (const { id, body, wc, seg, hash } of todo) {
const cap = caps[id];
if (cap) { console.log(id, 'already', cap.caption); continue; }
const tmp = cfg.abs(cfg.dirs.work, `scribe-${id}.json`);
writeJson(tmp, body);
const d = bc(['documents', 'put', cfg.video, '--format', 'elevenlabs', '--kind', 'speech', '--language', cfg.language, '--source-asset', seg.audioAsset, '--name', `${cfg.names.narration} ${id}`, '--body', '@' + tmp]);
const c = bc(['captions', 'create', cfg.video, '--document-id', d.documentId]);
caps[id] = { speech: d.documentId, caption: c.documentId, track: c.trackId, items: c.itemIds, cues: c.cueCount, enabled: c.enabled, audioItem: seg.audioItem, wordsHash: hash };
if (home && c.trackId !== home) console.log(`${id}: captions landed on ${c.trackId}, not ${home}; something on that track overlaps them, move them back with moveItems`);
const from = seg.from ?? Math.round(parseFloat(seg.at) * cfg.fps);
if (from === 0 && wc.length && wc[0].start > 0) {
const v = allItems(bc, cfg.video);
const cover = v.items.find(it => (c.itemIds || []).includes(it.id) && it.startFrame === 0);
if (cover) {
const at = Math.ceil(wc[0].start * cfg.fps) / cfg.fps;
bc(['edits', 'apply', cfg.video, '--expected-revision', String(v.revision), '--label', `Keep captions off the cover ${id}`, '--operations', JSON.stringify([{ type: 'trimItem', itemId: cover.id, edge: 'start', at }])]);
caps[id].coverTrim = at;
console.log(id, 'caption starts at', at, 's, off the cover frame');
}
}
writeJson(log, caps);
console.log(JSON.stringify({ id, ...caps[id] }));
}
if (!flags.dry) { applyTiming(cfg, bc, caps); writeJson(log, caps); }
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |