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/overlay.mjs
View on GitHub#!/usr/bin/env node
// i18n-ignore-file: 给智能体拷走用的独立脚本,日志与报错写英文,给人看的名字与标签从配置来
// 叠在代码画面之上的几层:实拍片段(footage 轨)、来源标签(label 轨)、原声的双语字幕(字幕轨)与水印(watermark 轨)。
// 全部写在时间线上,不剪、不烧、不转码:每个原片链接导入一次,每个镜头是一个视频片段,原片区间在 timeMap、裁剪与取景在
// crop / focalPoint、原声的音量与淡入淡出在 embeddedAudio,转场是时间线转场。都是可选的:没有 clips.json、配置里没有 watermark 就跳过。
// node tools/overlay.mjs [--dry] 打印每个片段与水印在时间线上的起止(默认,不连 Runtime)
// node tools/overlay.mjs --place [seg …] 导入原片(linked)、按段放片段与标签、建原声字幕、放水印;记在 overlay.json,已放过的不重放
// node tools/overlay.mjs --check 对照视频里的片段、标签与字幕和现在算出的计划(assemble 波纹挪过、段落改过之后跑)
// clips.json:{ source, label, clips: [{ id, seg, line, cue?, head?, gap?, follow?, from, to, source?, label?, audio?, subs?, note? }] }
// from / to 是原片的秒数;片段从 seg 段第 line 句(从 0 数)开始放;cue 可省,给了就从这句里 cue 这几个字被说到的时刻开始。
// 同一段里下一个片段开始时截断,不越过段尾。label 是画面一角的来源标签(成片语言写,例如「画面:<来源>」),
// 单个片段可以另给 source 与 label;audio 为 true 时保留原声(当事人原话、现场声),否则静音;给数字是在响度对齐之上再乘的音量(0–1)。
// follow 为 true 的片段不看 line / cue,紧接着 clips.json 里上一个片段的原长之后放(几段素材连成一整块,中间硬切)。
// head 为 true 的片段不看 line / cue,从段首放:段首到第一句之间有零点几秒的空,实拍垫底(under)时这段露出画布底色,
// 每段的第一个镜头带上 head 就盖住了(片段长度要算到段首起)。
// gap 为 true 的片段让旁白停下来单独放:先跑 gaps.mjs 在旁白里留白,片段放在留白的起点(line 是插在哪一句之前)。
// subs:[{ from, to, text, tr, who? }],from / to 是原片秒数,text 成片语言、tr 原话、who 说话人。同一原片的 subs 写成一份转写
// (原话,按 subs 的起止摊开词时间)和一份译文(text),再建成双语字幕层:成片语言在上、原话在下、说话人名字在最上面,样式与旁白字幕分开。
// after:从第 after 句念完的时刻放(负数从末句倒数,true 是末句),用在段尾 hold 里放一句原声;offset(秒)在任何起点之上再挪。
// 只有声音的片段:来源是录音文件(mp3 / wav / m4a / aac / flac / ogg / opus),或视频片段写 video: false。要带 audio,
// 放在单独一条音频轨(names.quote)上,起点精确到子帧(原声与场景里逐词点亮的字对得上),没有画面、裁剪、画中画与转场;
// 来源标签只在片段自己写了 label 时放(顶层的 label 不套用,通常由场景画出处)。asset 可给已在视频里的素材 ID(例如正片的录音),
// 这时不再按 source 导入,source 仍用来量响度。words 为 true 时 quotes.mjs 给它对出逐词时间,build.mjs 写进这一段的 TIMING.clips。
// in / out:{ kind, dur?, easing? } 单侧转场(kind 取 dissolve / wipe / slide / zoom / iris,dur 秒数缺省 0.5,easing 缺省 ease-in-out),
// 片段从下面的代码画面里渐入、渐出;和前后片段首尾相接的那一侧不做(同一条轨上背靠背只能硬切)。原声的淡入淡出跟着转场的长度走,没有转场时 0.15 / 0.3 秒。
// crop:[x, y, w, h] 按原片像素裁掉黑边、色条,按素材记录里的原片宽高换成四边比例;focus:[s, dx, dy] 放大 s 倍、画面中心挪到离正中
// dx%、dy%(正数往左、往上),换成 focalPoint。fit 为 contain 时整幅放进画面、两侧补黑(4:3 的老转播),缺省 cover 铺满裁切。
// pip:{ x, y, w } 画中画,不铺满:x、y 是框中心、w 是框宽(画布宽高的百分比),高按原片(裁过之后)的宽高比;代码画面在下面照常讲。
// 画中画的来源标签贴在框的一角。转场作用在框上。
// 原声的响度:放上之前用 ffmpeg 的 ebur128 只读地量一遍放上去的那段原片(不写文件),换成片段音量 10^((footage.loudness − 实测) / 20),夹在 0–4。
// 来源是图片(png / jpg / webp / bmp / tif)时 from / to 只定时长(写 0 和秒数),整段停在这张图上;不能带 audio。
// 图片推拉(Ken Burns,旧版的 motion)现在没有时间线上的写法:带 motion 的片段报错,换成不动的图或一段视频。
// 配置 footage:{ clips, layer, loudness, label: { size, corner, inset: { x, y }, w, color, background, font }, subs: { y, fontSize, font, speaker: { scale, color }, language } };
// watermark:{ file | text, dur, w, y, x: { left, right }, at: [...], style?, opacity?, fade? }。
// 给 file 是图片水印(画中画);给 text 是文字水印(品牌要一行字时用它):style 是文字样式(fontFamily、fontSize、fontWeight、fontColor 等,
// 与 setStyle 的写法相同),opacity 缺省 0.6,fade 是淡入淡出的秒数(缺省 0.8,0 为硬切);w 是文字框宽(画面宽的百分比)。
// layer 缺省 over(实拍盖在代码画面上面);under 时实拍轨放到代码包那条轨的下面当底,包要透明(bundle.alpha 为 true,场景只画要叠的卡片与标注),
// 实拍之外的时刻露出的是画布底色(空),用 coverage.mjs 看有多少空着。标签轨总在最上面。
// label.corner 取 top-right(缺省)/ top-left / bottom-right / bottom-left,inset 是离画面边的距离(画面宽高的百分比),size 是画布像素。
// subs.language 是原话的语言(缺省 en);译文的语言是配置的 language。
// at 的每一项是 { seg, line, cue?, side } 或 { t, side }(时间线秒数),side 取 left / right;x、y、w 是画面宽高的百分比。
// 单个水印可以另给 x、y(百分比)放在别处:穿插在片中、换着位置放,裁掉一角也去不掉。
import fs from 'node:fs';
import path from 'node:path';
import crypto from 'node:crypto';
import { spawnSync } from 'node:child_process';
import { loadConfig, parseArgs, makeBc, readJson, writeJson, layout, atOf, cueTime, clipTimes, allItems, cropFraction, focalOf, mediaTime, isAudioOnly } from './lib.mjs';
const cfg = loadConfig();
const { ids, flags } = parseArgs(process.argv);
if (flags.render || flags.preview) throw new Error('--render and --preview are gone: footage is placed straight from the source files (--place), nothing is cut ahead');
const FPS = cfg.fps, V = cfg.video, { width: W, height: H } = cfg.canvas;
const fo = {
clips: 'clips.json', loudness: -17, ...cfg.footage,
label: { size: 26, corner: 'top-right', w: 40, color: '#DDDDDD', background: '#0000008C', font: null, ...cfg.footage?.label, inset: { x: 2.4, y: 4, ...cfg.footage?.label?.inset } },
subs: { y: 88, fontSize: null, font: null, language: 'en', ...cfg.footage?.subs, speaker: { scale: 0.62, color: '#BBBBBB', ...cfg.footage?.subs?.speaker } },
};
const wm = cfg.watermark ? { dur: 5, w: cfg.watermark.text ? 20 : 6, y: 70, x: { left: 5, right: 95 }, at: [], opacity: 0.6, fade: 0.8, ...cfg.watermark } : null;
if (wm && !wm.text === !wm.file) throw new Error('watermark needs exactly one of file (an image) or text');
if (fo.layer && !['over', 'under'].includes(fo.layer)) throw new Error(`footage.layer "${fo.layer}" not one of over, under`);
const CORNERS = ['top-right', 'top-left', 'bottom-right', 'bottom-left'];
if (!CORNERS.includes(fo.label.corner)) throw new Error(`footage.label.corner "${fo.label.corner}" not one of ${CORNERS.join(', ')}`);
const under = fo.layer === 'under';
if (under && !cfg.bundle.alpha) console.log('warning: footage.layer is under but bundle.alpha is not true; opaque code scenes will hide the footage (set bundle.alpha and rebuild)');
const cl = readJson(cfg.abs(fo.clips), { clips: [] });
const logFile = cfg.abs('overlay.json');
const log = { video: V, assets: {}, footage: {}, loudness: {}, bites: {}, watermarks: [], ...readJson(logFile, {}) };
if (log.video !== V) throw new Error(`overlay.json belongs to ${log.video}, config says ${V}`);
const save = () => writeJson(logFile, log);
const src = c => c.source || cl.source;
const audioOnly = c => isAudioOnly(c, src(c));
// 只有声音的片段不套用顶层的 label(出处通常由场景画),自己写了才放
const labelOf = c => (audioOnly(c) ? (typeof c.label === 'string' ? c.label : null) : c.label ?? cl.label ?? null);
const KINDS = ['dissolve', 'wipe', 'slide', 'zoom', 'iris'];
const IMAGE = /\.(png|jpe?g|webp|bmp|tiff?)$/i;
const isImage = c => IMAGE.test(src(c));
for (const c of cl.clips) {
for (const side of ['in', 'out']) if (c[side] && !KINDS.includes(c[side].kind)) throw new Error(`${c.id}: ${side}.kind "${c[side].kind}" not one of ${KINDS.join(', ')}`);
if (isImage(c) && c.audio) throw new Error(`${c.id}: ${src(c)} is a still image and has no sound; drop audio`);
if (audioOnly(c) && !c.audio) throw new Error(`${c.id}: ${src(c)} is placed as sound only (an audio file, or video: false); give it audio: true`);
if (audioOnly(c) && (c.pip || c.crop || c.focus || c.in || c.out)) throw new Error(`${c.id}: a sound-only clip has no picture; drop pip, crop, focus, in and out`);
if (c.motion) throw new Error(`${c.id}: motion (Ken Burns) has no timeline form yet; drop it, or use a video clip`);
if (c.crop && !(Array.isArray(c.crop) && c.crop.length === 4)) throw new Error(`${c.id}: crop is [x, y, w, h] in source pixels`);
if (c.focus) focalOf(c.focus);
if (c.pip && (!['x', 'y', 'w'].every(k => Number.isFinite(c.pip[k])) || c.pip.w <= 0)) throw new Error(`${c.id}: pip needs numeric x, y and a positive w (percent of the canvas)`);
for (const s of c.subs || []) if (!(s.to > s.from && s.from >= c.from - 0.05 && s.to <= c.to + 0.05) || !s.text || !s.tr) throw new Error(`${c.id}: each sub needs text, tr and from < to inside the clip (${c.from}–${c.to})`);
}
// 1. 排时间:段起点(与 assemble.mjs 同一套整帧累加)+ 这一句或 cue 的对齐时刻
const lay = layout(cfg);
const seg = Object.fromEntries(lay.map(s => [s.id, s]));
const timing = id => readJson(cfg.abs(cfg.dirs.timing, `${id}.json`)) ?? (() => { throw new Error(`timing/${id}.json missing; run align.mjs ${id}`); })();
const at = (segId, line, cue) => {
const s = seg[segId];
if (!s) throw new Error(`segment ${segId} not in order`);
return s.from / FPS + cueTime(timing(segId), line, cue);
};
const plan = [];
for (const s of lay) {
const mine = cl.clips.filter(c => c.seg === s.id), t0 = mine.length ? clipTimes(timing(s.id), mine) : {};
const cs = mine.map(c => ({ ...c, at: s.from / FPS + t0[c.id] })).sort((a, b) => a.at - b.at);
const lead = cs.find(c => !c.pip);
if (under && lead && lead.at - s.from / FPS >= 1 / FPS) console.log(`warning: ${s.id} has no footage for its first ${(lead.at - s.from / FPS).toFixed(2)}s (the canvas shows); give ${lead.id} head: true`);
cs.forEach((c, i) => {
const end = (s.from + s.frames) / FPS, stop = Math.min(cs[i + 1]?.at ?? Infinity, end);
if (c.at + c.to - c.from > end + 0.5 / FPS && stop === end) console.log(`warning: ${c.id} runs ${(c.at + c.to - c.from - end).toFixed(2)}s past the end of ${s.id}; cut at the segment end (shorten to, or move the clip earlier)`);
// 只有声音的片段按精确时刻放(整帧 + 子帧),画面片段按最近的整帧
const f0 = audioOnly(c) ? Math.floor(c.at * FPS + 1e-9) : Math.round(c.at * FPS), f1 = Math.min(Math.round((c.at + c.to - c.from) * FPS), Math.round(stop * FPS));
if (f1 > f0) plan.push({ ...c, f0, frames: f1 - f0, play: Math.min(c.to - c.from, stop - c.at) });
});
}
// 转场:首尾和别的片段相接的一侧不做;长度不超过片段的一半
plan.sort((a, b) => a.f0 - b.f0);
plan.forEach((c, i) => {
const prev = plan[i - 1], next = plan[i + 1];
if (audioOnly(c)) { c.tin = c.tout = null; return; }
const tr = (t, touching) => t && !touching ? { kind: t.kind, easing: t.easing || 'ease-in-out', frames: Math.min(Math.round((t.dur ?? 0.5) * FPS), Math.floor(c.frames / 2)) } : null;
c.tin = tr(c.in, prev && prev.f0 + prev.frames >= c.f0);
c.tout = tr(c.out, next && c.f0 + c.frames >= next.f0);
});
const stray = cl.clips.filter(c => !seg[c.seg]).map(c => c.id);
if (stray.length) throw new Error(`clips in segments not in order: ${stray.join(', ')}`);
const marks = (wm?.at || []).map(m => {
if (m.x == null && !wm.x[m.side]) throw new Error(`watermark side "${m.side}" not in watermark.x (or give the mark its own x)`);
const f = Math.round((m.t ?? at(m.seg, m.line, m.cue)) * FPS);
return { side: m.side, f, t: +(f / FPS).toFixed(4), x: m.x ?? wm.x[m.side], y: m.y ?? wm.y };
});
for (const c of plan) console.log(`${c.id.padEnd(16)} ${(c.f0 / FPS).toFixed(2).padStart(8)}s +${(c.frames / FPS).toFixed(2)}s src ${c.from}${c.gap ? ' gap' : ''}${c.audio ? ` audio${typeof c.audio === 'number' ? ' ' + c.audio : ''}` : ''}${c.subs ? ` subs:${c.subs.length}` : ''}${c.pip ? ` pip ${c.pip.x},${c.pip.y} w${c.pip.w}` : ''}${c.crop ? ` crop ${c.crop.join(',')}` : ''}${c.focus ? ` focus ${c.focus.join(',')}` : ''}${isImage(c) ? ' still' : ''}${audioOnly(c) ? ' sound only' : ''}${c.tin ? ` in ${c.tin.kind} ${(c.tin.frames / FPS).toFixed(2)}` : ''}${c.tout ? ` out ${c.tout.kind} ${(c.tout.frames / FPS).toFixed(2)}` : ''}`);
for (const m of marks) console.log(`watermark ${String(m.side ?? '').padEnd(5)} ${m.t}s +${wm.dur}s at ${m.x},${m.y}`);
if (!flags.place && !flags.check) process.exit(0);
// 原声的响度:只读地量放上去的那一段(ebur128 的积分响度),按来源与起止缓存在 overlay.json
const spanKey = c => `${src(c)}|${c.from}|${(audioOnly(c) ? c.play : c.frames / FPS).toFixed(3)}`;
function ebur(c) {
const key = spanKey(c);
if (log.loudness[key] !== undefined) return log.loudness[key];
const r = spawnSync('ffmpeg', ['-hide_banner', '-nostats', '-ss', String(c.from), '-t', (audioOnly(c) ? c.play : c.frames / FPS).toFixed(3), '-i', src(c), '-vn', '-af', 'ebur128', '-f', 'null', '-'], { encoding: 'utf8', maxBuffer: 64 << 20 });
const hit = [...(r.stderr || '').matchAll(/I:\s+(-?[\d.]+|-inf)\s+LUFS/g)].at(-1);
if (!hit) throw new Error(`${c.id}: loudness of ${src(c)} not measured (ffmpeg exit ${r.status}): ${(r.stderr || '').slice(-400)}`);
const lufs = hit[1] === '-inf' ? null : +hit[1];
log.loudness[key] = lufs;
save();
return lufs;
}
const volumeOf = c => {
const lufs = ebur(c), gain = typeof c.audio === 'number' ? c.audio : 1;
if (lufs === null) return { volume: gain, lufs };
const raw = Math.pow(10, (fo.loudness - lufs) / 20) * gain;
if (raw > 4) console.log(`warning: ${c.id} measures ${lufs} LUFS; reaching ${fo.loudness} needs ×${raw.toFixed(2)}, clamped to ×4`);
return { volume: +Math.min(Math.max(raw, 0), 4).toFixed(4), lufs };
};
const bc = makeBc(cfg);
const inspect = (limit = 1) => bc(['videos', 'inspect', V, '--limit', String(limit)]);
const subsHash = list => 'sha256:' + crypto.createHash('sha256').update(JSON.stringify(list.map(({ c, s }) => [c.id, s.from, s.to, s.who ?? null, s.text, s.tr]))).digest('hex').slice(0, 24);
const bitesBySource = () => {
const out = new Map();
for (const c of plan) for (const s of c.subs || []) {
if (!out.has(src(c))) out.set(src(c), []);
out.get(src(c)).push({ c, s });
}
for (const list of out.values()) list.sort((a, b) => a.s.from - b.s.from);
return out;
};
const labelText = c => labelOf(c);
// 2. 核对:记录里的片段、标签与字幕还在不在、起点与现在算出的是否一致
if (flags.check) {
const items = new Map(allItems(bc, V).items.map(it => [it.id, it]));
const bad = [];
for (const c of plan) {
const r = log.footage[c.id], it = r && items.get(r.itemId);
if (!r) { bad.push(`${c.id}: not placed`); continue; }
if (!it) bad.push(`${c.id}: item ${r.itemId} gone`);
else if (it.startFrame !== c.f0) bad.push(`${c.id}: at frame ${it.startFrame}, plan says ${c.f0} (${((c.f0 - it.startFrame) / FPS).toFixed(2)} s off)`);
else if (r.frames !== c.frames) bad.push(`${c.id}: ${r.frames} frames long, plan says ${c.frames}`);
if (r.from !== undefined && r.from !== c.from) bad.push(`${c.id}: plays the source from ${r.from}s, clips.json says ${c.from}s`);
if (c.audio && log.loudness[spanKey(c)] === undefined) bad.push(`${c.id}: loudness not measured for the current span`);
const label = labelText(c);
if (label && !r.labelItemId) bad.push(`${c.id}: no source label`);
else if (r.labelItemId && !items.get(r.labelItemId)) bad.push(`${c.id}: label item ${r.labelItemId} gone`);
else if (r.labelItemId && r.label !== label) bad.push(`${c.id}: label says "${r.label}", clips.json says "${label}"`);
}
for (const [id, r] of Object.entries(log.footage)) if (!plan.some(c => c.id === id)) bad.push(`${id}: placed but no longer in clips.json (item ${r.itemId})`);
for (const [file, list] of bitesBySource()) {
const b = log.bites[file];
if (!b) bad.push(`subs of ${path.basename(file)}: no captions`);
else {
if (b.hash !== subsHash(list)) bad.push(`subs of ${path.basename(file)}: changed since the captions were made (delete the caption items ${b.itemIds.join(', ')} and its entry in overlay.json bites, then --place)`);
for (const id of b.itemIds) if (!items.get(id)) bad.push(`subs of ${path.basename(file)}: caption item ${id} gone`);
}
}
for (const r of log.watermarks) {
const it = items.get(r.itemId);
if (!it) bad.push(`watermark ${r.t}s: item ${r.itemId} gone`);
else if (!marks.some(m => m.f === it.startFrame)) bad.push(`watermark ${r.t}s: at frame ${it.startFrame}, not where watermark.at puts one now`);
}
for (const m of marks) if (!log.watermarks.some(r => Math.round(r.t * FPS) === m.f)) bad.push(`watermark ${m.t}s: not placed`);
console.log(bad.length ? 'CHECK FAILED\n' + bad.join('\n') : `check ok: ${plan.length} clips, ${Object.keys(log.bites).length} sound-bite caption layers, ${log.watermarks.length} watermarks`);
process.exit(bad.length ? 1 : 0);
}
// 3. 放上时间线
const apply = (label, ops) => bc(['edits', 'apply', V, '--expected-revision', String(inspect().revision), '--label', label, '--operations', JSON.stringify(ops)]);
const trackId = (name, kind = 'visual') => {
const find = () => inspect().tracks.find(x => x.kind === kind && x.name === name);
let t = find();
if (!t) { apply(`Add ${name} track`, [{ type: 'addTrack', kind, name }]); t = find(); }
return t.id;
};
const labelName = `${cfg.names.footage} · label`;
const todoClips = plan.filter(c => !log.footage[c.id] && (!ids.length || ids.includes(c.seg)));
// 3a. 每个原片链接导入一次;宽高与版本取素材记录
const assetOf = (file, { needSize = true, assetId = null } = {}) => {
if (assetId) {
const rec = inspect().assets?.find(a => a.id === assetId);
if (!rec?.revision) throw new Error(`asset ${assetId} not in videos inspect`);
return { assetId, ...(rec.size ? sizeOf(rec) : { revision: rec.revision }) };
}
const known = log.assets[file];
const listed = inspect().assets?.find(a => a.id === known?.assetId);
if (listed?.revision && (listed.size || !needSize)) return { ...known, ...(listed.size ? sizeOf(listed) : { revision: listed.revision }) };
if (!known) {
if (!fs.existsSync(file)) throw new Error(`source ${file} not found`);
const a = bc(['assets', 'import', V, '--path', file, '--storage', 'linked', '--name', path.basename(file)]);
log.assets[file] = { assetId: a.assetId };
save();
console.log('linked', path.basename(file), a.assetId);
}
const rec = inspect().assets?.find(a => a.id === log.assets[file].assetId);
if (!rec?.revision) throw new Error(`asset ${log.assets[file].assetId} for ${file} not in videos inspect; was it removed?`);
if (!rec.size && needSize) throw new Error(`asset ${rec.id} (${path.basename(file)}) has no picture size in its record; crop and pip need it`);
return { ...log.assets[file], ...(rec.size ? sizeOf(rec) : { revision: rec.revision }) };
};
const sizeOf = rec => { const [w, h] = rec.size.split('x').map(Number); return { revision: rec.revision, w, h }; };
if (todoClips.length) {
const T = todoClips.some(c => !audioOnly(c)) ? trackId(cfg.names.footage) : null;
const QA = todoClips.some(audioOnly) ? trackId(cfg.names.quote, 'audio') : null;
// footage.layer 为 under:实拍轨挪到代码包那条轨下面(包要透明,bundle.alpha);tracks 按叠放次序从下到上列
if (under && T) {
const vs = inspect().tracks.filter(x => x.kind === 'visual');
const base = cfg.tracks?.visual || vs.find(x => ![cfg.names.footage, cfg.names.watermark, labelName].includes(x.name))?.id;
if (base && vs.findIndex(x => x.id === T) > vs.findIndex(x => x.id === base)) {
apply(`Move ${cfg.names.footage} track under the code scenes`, [{ type: 'moveTrack', trackId: T, target: base, position: 'below' }]);
console.log('footage track moved under', base);
}
}
const needLabels = todoClips.some(labelText);
const L = needLabels ? trackId(labelName) : null;
if (L) {
const vs = inspect().tracks.filter(x => x.kind === 'visual');
const top = vs.at(-1);
if (top && top.id !== L) apply(`Move ${labelName} track to the top`, [{ type: 'moveTrack', trackId: L, target: top.id, position: 'above' }]);
}
const fontSize = px => +(px * 540 / Math.min(W, H)).toFixed(2);
for (const segId of [...new Set(todoClips.map(c => c.seg))]) {
const batch = todoClips.filter(c => c.seg === segId);
const items = [], meta = [];
for (const c of batch) {
if (audioOnly(c)) {
// 只有声音:一个音频片段,起点 = 整帧 + 子帧,长度按原片秒数(不越过下一个片段与段尾)
const a = assetOf(src(c), { needSize: false, assetId: c.asset || null });
const audio = volumeOf(c), name = `${cfg.names.quote} ${c.id}`;
const fadeIn = c.fadeIn ?? 0.08, fadeOut = c.fadeOut ?? 0.45;
items.push({
type: 'audio', trackId: QA, name, assetRef: { id: a.assetId, revision: a.revision },
fromFrame: c.f0, subframeOffset: mediaTime(Math.max(0, c.at - c.f0 / FPS)), playDuration: mediaTime(c.play),
timeMap: { kind: 'linear', sourceIn: mediaTime(c.from), rate: { num: 1, den: 1 } },
mix: { volume: audio.volume, fadeIn: mediaTime(fadeIn), fadeOut: mediaTime(fadeOut) },
});
const label = labelText(c);
if (label) items.push(labelItem(c, label, L, null, null, fontSize));
meta.push({ c, name, label, audio, assetId: a.assetId });
continue;
}
const a = assetOf(src(c));
const image = isImage(c);
const crop = c.crop ? cropFraction(c.crop, a.w, a.h) : null;
const audio = c.audio ? volumeOf(c) : null;
const fadeIn = c.tin ? c.tin.frames / FPS : 0.15, fadeOut = c.tout ? c.tout.frames / FPS : 0.3;
const name = `${cfg.names.footage} ${c.id}`;
items.push({
type: image ? 'image' : 'video', trackId: T, name, role: 'broll',
assetRef: { id: a.assetId, revision: a.revision },
span: { fromFrame: c.f0, durationFrames: c.frames },
place: c.pip ? { x: c.pip.x, y: c.pip.y, w: c.pip.w } : {},
mode: c.pip ? 'pip' : 'fullscreen', fit: c.fit ?? 'cover',
...(crop ? { crop } : {}),
...(c.focus ? { focalPoint: focalOf(c.focus) } : {}),
...(image ? {} : {
timeMap: { kind: 'linear', sourceIn: mediaTime(c.from), rate: { num: 1, den: 1 } },
embeddedAudio: audio ? { enabled: true, volume: audio.volume, fadeIn: mediaTime(fadeIn), fadeOut: mediaTime(fadeOut) } : { enabled: false, volume: 1 },
}),
});
const label = labelText(c);
if (label) items.push(labelItem(c, label, L, a, crop, fontSize));
meta.push({ c, name, label, audio, assetId: a.assetId });
}
const r = apply(`${cfg.names.footage} ${segId}`, [{ type: 'insertItems', items }]);
// 回执的 createdIds 不一定和 items 同序:按名字与起点在视频里认回片段
const created = new Set(r.createdIds || []);
const placed = allItems(bc, V, Math.max(0, Math.min(...batch.map(c => c.f0)) / FPS - 1)).items.filter(it => created.has(it.id));
const find = (name, f0) => placed.find(it => it.name === name && it.startFrame === f0)?.id;
const trans = [];
for (const m of meta) {
const itemId = find(m.name, m.c.f0);
if (!itemId) throw new Error(`${m.c.id}: placed item not found in the receipt (${JSON.stringify(r).slice(0, 400)})`);
const labelItemId = m.label ? find(`${labelName} ${m.c.id}`, m.c.f0) : null;
log.footage[m.c.id] = { itemId, assetId: m.assetId, f0: m.c.f0, frames: m.c.frames, from: m.c.from, ...(m.audio ? { volume: m.audio.volume, lufs: m.audio.lufs } : {}), ...(m.label ? { labelItemId, label: m.label } : {}) };
for (const [k, side] of [['tin', 'rightItemId'], ['tout', 'leftItemId']]) if (m.c[k]) trans.push({
type: 'setTransition', [side]: itemId, kind: m.c[k].kind, params: {}, easing: m.c[k].easing,
duration: { unit: 'frames', value: m.c[k].frames }, alignment: 'exact-frame',
});
}
save();
if (trans.length) apply(`${cfg.names.footage} ${segId} transitions`, trans);
console.log('placed', segId, batch.map(c => c.id).join(' '));
}
}
// 来源标签:一个文字片段,和镜头同起同止,贴在画面(画中画时是框)的一角
function labelItem(c, text, trackId, a, crop, fontSize) {
const { corner, inset, w } = fo.label;
const right = corner.endsWith('right'), top = corner.startsWith('top');
let box = { l: 0, t: 0, r: 100, b: 100 };
if (c.pip) {
const cw = a.w * (1 - (crop?.left ?? 0) - (crop?.right ?? 0)), ch = a.h * (1 - (crop?.top ?? 0) - (crop?.bottom ?? 0));
const hPct = c.pip.w * W / 100 * ch / cw / H * 100;
box = { l: c.pip.x - c.pip.w / 2, r: c.pip.x + c.pip.w / 2, t: c.pip.y - hPct / 2, b: c.pip.y + hPct / 2 };
}
const ix = c.pip ? inset.x / 2 : inset.x, iy = c.pip ? inset.y / 2 : inset.y;
const lw = Math.min(w, box.r - box.l);
const x = right ? box.r - ix - lw / 2 : box.l + ix + lw / 2;
const y = top ? box.t + iy : box.b - iy;
return {
type: 'text', trackId, name: `${labelName} ${c.id}`, role: 'screentext', text,
span: { fromFrame: c.f0, durationFrames: c.frames },
place: { x: +x.toFixed(3), y: +y.toFixed(3), w: +lw.toFixed(3) },
verticalAlign: top ? 'top' : 'bottom',
style: {
fontSize: fontSize(fo.label.size), fontColor: fo.label.color, textAlign: right ? 'right' : 'left',
...(fo.label.background ? { backgroundColor: fo.label.background, backgroundPadding: Math.round(fontSize(fo.label.size) * 0.45) } : {}),
...(fo.label.font ? { fontFamily: fo.label.font } : {}),
},
};
}
// 3b. 原声字幕:同一原片的 subs 写成一份转写(原话)和一份译文(成片语言),建成双语字幕层,样式与旁白字幕分开
const work = cfg.abs(cfg.dirs.work, 'bites');
for (const [file, list] of bitesBySource()) {
if (log.bites[file]) continue;
if (!list.every(({ c }) => log.footage[c.id])) continue; // 这一原片的片段还没放完(--place 只放了部分段)
const assetId = log.assets[file].assetId;
const speakers = [...new Set(list.map(({ s }) => s.who).filter(Boolean))].map((name, i) => ({ id: `spk-${i + 1}`, name }));
const spk = who => speakers.find(x => x.name === who)?.id;
const words = [], paragraphBreaks = [];
list.forEach(({ s }, k) => {
const toks = s.tr.trim().split(/\s+/);
const total = toks.reduce((n, t) => n + t.length + 1, 0);
let t = s.from * 1000, step = (s.to - s.from) * 1000 / total;
toks.forEach((text, j) => {
const len = (text.length + 1) * step, id = `b${k}_${j}`;
if (j === 0) paragraphBreaks.push(id);
words.push({ id, start: Math.round(t), end: Math.round(t + len - Math.min(step, len / 4)), text, ...(spk(s.who) ? { speaker: spk(s.who) } : {}) });
t += len;
});
});
const body = { schema: 'baocut.speech/1', clock: 'source-asset', timescale: 1000, engine: { kind: 'sound-bite-subs' }, speakers, words, sentences: null, chapters: [], paragraphBreaks };
fs.mkdirSync(work, { recursive: true });
const name = path.basename(file).replace(/\.[^.]+$/, '');
const speechFile = path.join(work, `${name}.speech.json`);
writeJson(speechFile, body);
const sp = bc(['documents', 'put', V, '--kind', 'speech', '--name', `${cfg.names.footage} ${name}`, '--language', fo.subs.language, '--source-asset', assetId, '--body', '@' + speechFile]);
const speechDoc = sp.documentId;
const basis = bc(['documents', 'read', V, '--document-id', speechDoc]).translationBasis;
if (!basis?.sentences?.length) throw new Error(`speech ${speechDoc}: no sentences in translationBasis`);
// 一条 sub 的原话里有句末标点时会切成几句:成片语言按自己的句末标点分成同样多份,分不齐时按原话的词数比例切
const units = [];
list.forEach(({ s }, k) => {
const mine = basis.sentences.filter(x => x.wordIds[0].startsWith(`b${k}_`));
const parts = splitText(s.text, mine.map(x => x.wordIds.length));
mine.forEach((x, i) => units.push({ id: `t-${x.id}`, sourceSentenceId: x.id, sourceFingerprint: x.fingerprint, naturalText: parts[i], alignment: null, status: 'draft' }));
});
const trFile = path.join(work, `${name}.translation.json`);
writeJson(trFile, { schema: 'baocut.translation/2', language: cfg.language, sourceBasis: basis.sourceBasis, units });
const tr = bc(['documents', 'put', V, '--kind', 'translation', '--name', `${cfg.names.footage} ${name} (${cfg.language})`, '--language', cfg.language, '--source-document', speechDoc, '--body', '@' + trFile]);
// 词时间是按各条起止摊开的,不逐字高亮(默认样式会把念到的词染成青色)
const style = { y: fo.subs.y, verticalAlign: 'bottom', speakerLabel: { on: true, scale: fo.subs.speaker.scale, color: fo.subs.speaker.color },
...(fo.subs.fontSize ? { fontSize: fo.subs.fontSize } : {}), ...(fo.subs.font ? { fontFamily: fo.subs.font } : {}), wordAnimation: { catalogId: 'none' } };
// 先给原话单建一层原文字幕(带自己的样式):转写没有原文层时,双语会配上画面上已有的旁白字幕、改掉旁白的样式
const orig = bc(['captions', 'create', V, '--document-id', speechDoc, '--style', JSON.stringify(style)]);
if (!orig.styleDocumentId) throw new Error(`captions for ${speechDoc}: no style document in receipt`);
const cap = bc(['captions', 'create', V, '--document-id', tr.documentId, '--bilingual']);
if (cap.pairedOriginal?.documentId !== orig.documentId || cap.styleDocumentId !== orig.styleDocumentId) {
throw new Error(`bilingual captions for ${name} paired with ${cap.pairedOriginal?.documentId}, not the sound-bite layer ${orig.documentId}`);
}
// 原声字幕跟着原话走:显示的提前与拖尾都收短,不越过片段(displayTiming 不是 setCaptionStyle 的键,用 patchDocument 写)
apply(`${cfg.names.footage} ${name} caption timing`, [{ type: 'patchDocument', documentId: orig.styleDocumentId, changes: [{ path: ['style', 'displayTiming'], value: { leadIn: 0, tail: 0.1 } }] }]);
log.bites[file] = { hash: subsHash(list), speechDoc, translationDoc: tr.documentId, styleDocumentId: orig.styleDocumentId, itemIds: [...(cap.itemIds || []), ...(orig.itemIds || [])] };
save();
console.log('sound-bite captions', name, `${list.length} subs, ${units.length} sentences`, log.bites[file].itemIds.join(' '));
}
function splitText(text, counts) {
if (counts.length === 1) return [text];
let parts = text.split(/(?<=[。!?!?])/).map(s => s.trim()).filter(Boolean);
if (parts.length === counts.length) return parts;
const chars = [...text], total = counts.reduce((a, b) => a + b, 0), out = [];
let from = 0, acc = 0;
counts.forEach((n, i) => {
acc += n;
const to = i === counts.length - 1 ? chars.length : Math.max(from + 1, Math.round(chars.length * acc / total));
out.push(chars.slice(from, to).join('').trim() || '…');
from = to;
});
return out;
}
// 3c. 水印
const todo = marks.filter(m => !log.watermarks.some(r => Math.round(r.t * FPS) === m.f));
if (wm?.text && todo.length) {
// 文字水印:一笔放下全部文字片段,再一笔设不透明度与两侧的淡入淡出(转场要片段 ID)
const S = trackId(cfg.names.watermark);
const dur = Math.round(wm.dur * FPS);
const style = { textAlign: 'center', lineHeight: 1.2, textOutline: { on: false }, dropShadow: { on: false }, ...wm.style };
const r = apply(`${cfg.names.watermark} × ${todo.length}`, [{ type: 'insertItems', items: todo.map(m => ({
type: 'text', trackId: S, name: cfg.names.watermark, text: wm.text, style, verticalAlign: 'center',
span: { fromFrame: m.f, durationFrames: dur }, place: { x: m.x, y: m.y, w: wm.w },
})) }]);
const created = r.createdIds || [];
if (created.length !== todo.length) throw new Error(`expected ${todo.length} watermark items, receipt has ${created.length}: ${JSON.stringify(r).slice(0, 600)}`);
const fade = Math.min(wm.fade, wm.dur / 2);
apply(`${cfg.names.watermark} opacity and fades`, created.flatMap(itemId => [
{ type: 'setStyle', itemId, opacity: wm.opacity },
...(fade > 0 ? [
{ type: 'setTransition', rightItemId: itemId, kind: 'dissolve', duration: fade, easing: 'linear' },
{ type: 'setTransition', leftItemId: itemId, kind: 'dissolve', duration: fade, easing: 'linear' },
] : []),
]));
todo.forEach((m, i) => { log.watermarks.push({ t: m.t, side: m.side, itemId: created[i] }); console.log('watermark', m.side ?? '', m.t, created[i]); });
save();
} else if (wm && todo.length) {
const S = trackId(cfg.names.watermark);
const a = inspect().assets?.find(x => x.name === cfg.names.watermark) ?? bc(['assets', 'import', V, '--path', cfg.abs(wm.file), '--storage', 'managed', '--name', cfg.names.watermark]);
const assetId = a.assetId ?? a.id;
for (const m of todo) {
const r = apply(`${cfg.names.watermark} ${m.t}s`, [{ type: 'addItem', asset: assetId, at: m.t, trackId: S, name: `${cfg.names.watermark} ${m.side}` }]);
const itemId = r.createdIds?.[0];
if (!itemId) throw new Error(`no item id in receipt: ${JSON.stringify(r).slice(0, 600)}`);
apply(`${cfg.names.watermark} ${m.t}s layout`, [
{ type: 'trimItem', itemId, edge: 'end', at: m.t + wm.dur },
{ type: 'setStyle', itemId, mode: 'pip' },
{ type: 'setTransform', itemId, x: m.x, y: m.y, w: wm.w },
]);
log.watermarks.push({ t: m.t, side: m.side, itemId });
save();
console.log('watermark', m.side, m.t, itemId);
}
}
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 |