video-create/.claude/skills/video-from-script/scripts/口播_assemble.js

#!/usr/bin/env node
/**
 * 口播批量剪辑 — CapCut 草稿组装 v2.1
 *
 * 策略:
 *   1. ffmpeg 按 edit 字幕时间轴裁剪视频片段 + concat 拼接 + 1.1x 倍速 → 单文件
 *   2. 上传裁剪后视频到 OSS → 剪映里显示为一条连续轨道
 *   3. 长字幕按标点拆分为 2-3s 短段，避免长时间遮挡画面
 *
 * 注: 尝试过 slot 级 source_timerange 方案(v3)，但 CapCut Mate API 不支持，
 *     本地改写 draft_content.json 会被云同步覆盖/删除。故回到 ffmpeg 预裁剪。
 *
 * 用法:
 *   node 口播_assemble.js --video <视频路径> --transcript <_edit.txt路径> [--name <草稿名>] [--speed 1.1]
 */

const path = require('path')
const fs = require('fs')
const { execSync, execFileSync } = require('child_process')

const { api, US, getConfig } = require('./lib/capcut-api')
const { addVideos, addSubtitles } = require('./lib/capcut-tracks')
const { syncDraft, registerDraft } = require('./sync-to-jianying')
const ossUpload = require('./oss-upload')

// ============================================================================
// 解析参数
// ============================================================================

function parseArgs() {
  const args = process.argv.slice(2)
  const result = {}
  for (let i = 0; i < args.length; i++) {
    if (args[i].startsWith('--')) {
      result[args[i].slice(2)] = args[i + 1] || true
      i++
    }
  }
  return result
}

// ============================================================================
// 解析字幕文件 → [{begin_ms, end_ms, text}]
// ============================================================================

function parseTranscript(filePath) {
  const lines = fs.readFileSync(filePath, 'utf-8').trim().split('\n')
  const sentences = []

  const re = /\[(\d+):(\d+)\.(\d+)\s*-->\s*(\d+):(\d+)\.(\d+)\]\s*(.*)/

  for (const line of lines) {
    const m = line.match(re)
    if (!m) continue
    const [, m1, s1, ms1, m2, s2, ms2, text] = m
    const begin_ms = (parseInt(m1) * 60 + parseInt(s1)) * 1000 + parseInt(ms1.padEnd(3, '0').slice(0, 3))
    const end_ms   = (parseInt(m2) * 60 + parseInt(s2)) * 1000 + parseInt(ms2.padEnd(3, '0').slice(0, 3))
    if (text.trim()) {
      sentences.push({ begin_ms, end_ms, text: text.trim() })
    }
  }

  return sentences
}

// ============================================================================
// 字幕拆分：长句按标点拆成 2-3s 短段
// ============================================================================

function splitSubtitleChunks(text, totalDurationMs) {
  if (!text) return []

  const rawParts = text.split(/(?<=[，。！？,\.!\?、；：])/).filter(Boolean)
  if (rawParts.length === 0) return [{ text, ratio: 1 }]

  const charCounts = rawParts.map(p => p.length)
  const totalChars = charCounts.reduce((a, b) => a + b, 0)
  if (totalChars === 0) return [{ text, ratio: 1 }]

  const merged = []
  let buffer = { text: '', chars: 0 }
  for (let i = 0; i < rawParts.length; i++) {
    buffer.text += rawParts[i]
    buffer.chars += charCounts[i]
    const estDuration = (buffer.chars / totalChars) * totalDurationMs
    if (estDuration >= 1500 || i === rawParts.length - 1) {
      merged.push({
        text: buffer.text,
        ratio: buffer.chars / totalChars,
      })
      buffer = { text: '', chars: 0 }
    }
  }
  if (buffer.chars > 0) {
    if (merged.length > 0) {
      merged[merged.length - 1].text += buffer.text
      merged[merged.length - 1].ratio += buffer.chars / totalChars
    } else {
      merged.push({ text: buffer.text, ratio: buffer.chars / totalChars })
    }
  }

  return merged.length > 0 ? merged : [{ text, ratio: 1 }]
}

// ============================================================================
// 构建 ffmpeg trim+concat+speed 命令
// ============================================================================

function buildTrimCommand(inputPath, outputPath, segments, speed) {
  const ffmpeg = '/usr/local/bin/ffmpeg'
  const filters = []
  const concatParts = []
  const n = segments.length

  for (let i = 0; i < n; i++) {
    const seg = segments[i]
    const t0 = seg.begin_ms / 1000
    const t1 = seg.end_ms / 1000
    filters.push(`[0:v]trim=${t0}:${t1},setpts=PTS-STARTPTS[vt${i}];`)
    filters.push(`[0:a]atrim=${t0}:${t1},asetpts=PTS-STARTPTS[at${i}];`)
    concatParts.push(`[vt${i}][at${i}]`)
  }

  const concatInputs = concatParts.join('')
  const filterComplex = (
    filters.join('') +
    `${concatInputs}concat=n=${n}:v=1:a=1[vraw][araw];` +
    `[vraw]setpts=${(1/speed).toFixed(6)}*PTS[v];` +
    `[araw]atempo=${speed}[a]`
  )

  return [
    ffmpeg, '-i', inputPath,
    '-filter_complex', filterComplex,
    '-map', '[v]', '-map', '[a]',
    '-c:v', 'libx264', '-crf', '23', '-preset', 'fast',
    '-c:a', 'aac', '-b:a', '128k',
    outputPath, '-y', '-loglevel', 'error',
  ]
}

// ============================================================================
// 获取视频时长（毫秒）
// ============================================================================

function getVideoDurationMs(videoPath) {
  try {
    const out = execSync(
      `/usr/local/bin/ffmpeg -i "${videoPath.replace(/"/g, '\\"')}" 2>&1 | grep Duration || true`
    ).toString()
    const m = out.match(/Duration:\s*(\d+):(\d+):(\d+)\.(\d+)/)
    if (!m) throw new Error('Duration not found')
    return (parseInt(m[1]) * 3600 + parseInt(m[2]) * 60 + parseInt(m[3])) * 1000
            + Math.round(parseInt(m[4].padEnd(3, '0').slice(0, 3)))
  } catch (e) {
    throw new Error(`无法读取视频时长: ${e.message}`)
  }
}

// ============================================================================
// 获取视频分辨率
// ============================================================================

function getVideoResolution(videoPath) {
  try {
    const out = execSync(
      `/usr/local/bin/ffmpeg -i "${videoPath.replace(/"/g, '\\"')}" 2>&1 | grep 'Video:' || true`
    ).toString()
    const m = out.match(/(\d{3,4})x(\d{3,4})/)
    if (m) return { width: parseInt(m[1]), height: parseInt(m[2]) }
  } catch {}
  return { width: 1080, height: 1920 }
}

// ============================================================================
// 获取下一个全局序号
// ============================================================================

function getNextGlobalIndex() {
  const config = getConfig()
  const draftDir = config.jianyingDraftPath
  if (!fs.existsSync(draftDir)) return 1

  const folders = fs.readdirSync(draftDir)
  let maxIdx = 0
  for (const f of folders) {
    const m = f.match(/_(\d{2})_/)
    if (m) maxIdx = Math.max(maxIdx, parseInt(m[1]))
  }
  return maxIdx + 1
}

// ============================================================================
// 主流程
// ============================================================================

async function main() {
  const args = parseArgs()

  if (!args.video || !args.transcript) {
    console.error('用法: node 口播_assemble.js --video <视频> --transcript <字幕文件> [--name <草稿名>] [--speed 1.1]')
    process.exit(1)
  }

  const videoPath = path.resolve(args.video)
  const transcriptPath = path.resolve(args.transcript)
  const speed = parseFloat(args.speed) || 1.1

  if (!fs.existsSync(videoPath)) { console.error(`视频不存在: ${videoPath}`); process.exit(1) }
  if (!fs.existsSync(transcriptPath)) { console.error(`字幕文件不存在: ${transcriptPath}`); process.exit(1) }

  // 账号字幕样式
  const accountDir = path.resolve(__dirname, '..', '..', '..', 'accounts', '口播批量剪辑')
  const accountConfig = fs.existsSync(path.join(accountDir, 'account.json'))
    ? JSON.parse(fs.readFileSync(path.join(accountDir, 'account.json'), 'utf-8'))
    : {}
  const subtitleStyle = accountConfig.capcut?.subtitleStyle || {}

  // 1. 解析字幕
  console.log('[1/6] 解析字幕...')
  const sentences = parseTranscript(transcriptPath)
  console.log(`   ${sentences.length} 句字幕`)
  if (sentences.length === 0) {
    console.error('字幕文件为空或格式不对')
    process.exit(1)
  }

  console.log('   --- 保留时间段（已删除气口/停顿）---')
  for (let i = 0; i < sentences.length; i++) {
    const s = sentences[i]
    const segDur = (s.end_ms - s.begin_ms) / 1000
    console.log(`   [${i+1}] ${(s.begin_ms/1000).toFixed(2)}s → ${(s.end_ms/1000).toFixed(2)}s (${segDur.toFixed(1)}s) ${s.text.slice(0,20)}...`)
  }

  // 2. 读取视频信息
  console.log('[2/6] 读取视频信息...')
  const { width: vw, height: vh } = getVideoResolution(videoPath)
  const width = 1080
  const height = 1920
  console.log(`   原始分辨率: ${vw}x${vh}`)

  // 3. 压缩到1080p + 按字幕裁剪 + 倍速
  console.log(`[3/6] 裁剪视频（${sentences.length}段）+ ${speed}x 倍速...`)

  // 先压缩到 1080p
  const tmpCompressed = `/tmp/口播_compressed_${Date.now()}.mp4`
  execSync(
    `/usr/local/bin/ffmpeg -i "${videoPath.replace(/"/g, '\\"')}" -vf "scale=1920:1080:force_original_aspect_ratio=decrease" -c:v libx264 -crf 23 -preset fast -c:a aac -b:a 128k "${tmpCompressed}" -y -loglevel error`,
    { stdio: 'pipe' }
  )
  console.log(`   压缩后: ${(fs.statSync(tmpCompressed).size / 1024 / 1024).toFixed(1)} MB`)

  // trim + concat + speed
  const tmpTrimmed = `/tmp/口播_trimmed_${Date.now()}.mp4`
  const trimArgs = buildTrimCommand(tmpCompressed, tmpTrimmed, sentences, speed)
  execFileSync(trimArgs[0], trimArgs.slice(1), { stdio: 'pipe', maxBuffer: 10 * 1024 * 1024 })
  console.log(`   剪辑后: ${(fs.statSync(tmpTrimmed).size / 1024 / 1024).toFixed(1)} MB`)

  const trimmedDurationMs = getVideoDurationMs(tmpTrimmed)
  const totalSourceMs = sentences.reduce((a, s) => a + (s.end_ms - s.begin_ms), 0)
  console.log(`   剪辑后时长: ${(trimmedDurationMs / 1000).toFixed(1)}s (原 ${totalSourceMs / 1000}s ÷ ${speed})`)

  fs.unlinkSync(tmpCompressed)

  // 4. 上传到 OSS
  console.log('[4/6] 上传到 OSS...')
  const { url: videoUrl } = await ossUpload.uploadFile(tmpTrimmed)
  fs.unlinkSync(tmpTrimmed)
  console.log(`   ${videoUrl.slice(0, 60)}...`)

  // 5. 创建草稿 + 添加视频 + 字幕
  console.log('[5/6] 创建草稿 + 添加视频 + 字幕...')
  const draftRes = await api('create_draft', { width, height })
  const draftUrl = draftRes.draft_url
  const draftId = new URL(draftUrl).searchParams.get('draft_id')
  console.log(`   draft_id: ${draftId}`)

  // 单段视频（已裁剪拼接+倍速）
  const durationUs = trimmedDurationMs * 1000
  const videoItems = [{ videoUrl, id: 'main' }]
  const videoTimeline = [{ start: 0, end: durationUs, duration: durationUs }]

  await addVideos(draftUrl, path.dirname(videoPath), videoItems, videoTimeline, width, height)

  // 字幕 timeline（按倍速后时间轴，逐句拆分短段）
  const subtitleItems = []
  const subtitleTimeline = []
  let cumUs = 0

  for (const s of sentences) {
    const segDurationMs = (s.end_ms - s.begin_ms) / speed
    const segDurationUs = Math.round(segDurationMs * 1000)
    const chunks = splitSubtitleChunks(s.text, segDurationMs)

    let chunkCumMs = 0
    for (const chunk of chunks) {
      const chunkDurationMs = chunk.ratio * segDurationMs
      const chunkDurationUs = Math.round(chunkDurationMs * 1000)
      const chunkStartUs = cumUs + Math.round(chunkCumMs * 1000)

      subtitleItems.push({ text: chunk.text })
      subtitleTimeline.push({
        start: chunkStartUs,
        end: chunkStartUs + chunkDurationUs,
        duration: chunkDurationUs,
      })
      chunkCumMs += chunkDurationMs
    }

    cumUs += segDurationUs
  }

  console.log(`   字幕拆分为 ${subtitleItems.length} 段`)
  await addSubtitles(draftUrl, subtitleItems, subtitleTimeline, subtitleStyle, false)

  // 6. 同步到本地剪映
  console.log('[6/6] 同步到剪映...')
  const today = new Date()
  const mmdd = String(today.getMonth() + 1).padStart(2, '0') + String(today.getDate()).padStart(2, '0')
  const idx = String(getNextGlobalIndex()).padStart(2, '0')
  const shortTitle = args.name || path.basename(videoPath, path.extname(videoPath)).replace(/[\s()]/g, '').slice(0, 6)
  const draftName = `口播批量剪辑_${mmdd}_${idx}_${shortTitle}`

  await syncDraft(draftUrl, { name: draftName })
  registerDraft(draftId, draftName, durationUs)

  // 重命名草稿文件夹（Mac 版剪映显示名 = 文件夹名）
  const config = getConfig()
  const oldDir = path.join(config.jianyingDraftPath, draftId)
  const newDir = path.join(config.jianyingDraftPath, draftName)
  if (fs.existsSync(oldDir) && !fs.existsSync(newDir)) {
    fs.renameSync(oldDir, newDir)
    console.log(`   草稿文件夹已重命名: ${draftName}`)
  }

  console.log(`\n完成！`)
  console.log(`  草稿名称: ${draftName}`)
  console.log(`  草稿地址: ${draftUrl}`)
  console.log(`  倍速: ${speed}x | 视频: 1段连续 | 字幕: ${subtitleItems.length}段`)
  console.log(`\n下一步: 封面生成`)
}

main().catch(err => {
  console.error('组装失败:', err.message)
  process.exit(1)
})