# Daily Workflow 5 — `video_summarizer`: Video URL / Local Video File -> Native Subtitles or FFmpeg + ASR (.srt) -> Key Viewpoints & Timestamped Chapters + Mindmap + SRT Subtitles + PDF/HTML Summary.
#
#   yanshi workflow run examples/generators/video_summarizer.yml \
#     --input source "https://www.youtube.com/watch?v=..." --output-dir out
name: video_summarizer
description: "输入在线视频链接或上传本地音视频文件，优先提取内封/平台字幕并自动回退 FFmpeg 音轨抽取 + ASR 语音转写生成标准 .srt 字幕，提炼核心观点、时间戳章节纪要、思维导图与精要 PDF"

inputs:
  source:
    description: "在线视频 URL（支持 yt-dlp 站点与直链 MP4/WebM）或本地视频/音频文件（支持 Studio 一键选择本地文件）"
    default: "https://media.w3.org/2010/05/sintel/trailer.mp4"
  subtitle_langs:
    description: "优先抓取的原生字幕语言列表（逗号分隔）"
    default: "zh-Hans,zh,en"
  asr_model:
    description: "无现成字幕时使用的语音识别模型（如 gemini-2.5-flash-asr、glm-asr、whisper-1）"
    default: "gemini-2.5-flash-asr"
  summary_lang:
    description: "观点总结与双语润色字幕的目标语言"
    default: "简体中文"
  polish_bilingual_srt:
    description: "是否额外生成一份按时间轴校对润色的目标语言 .srt 字幕文件"
    default: true
  export_pdf:
    description: "同时导出排版精美的视频总结报告 PDF"
    default: true

nodes:
  - id: ingest
    type: video_ingest
    parameters:
      prefer_subtitles: true
      max_duration_seconds: 5400

  - id: asr_opts
    type: template
    parameters:
      output_format: json
      template: '{"asr_model": {{ asr_model | json_encode() }}, "polish_srt": {{ polish_bilingual_srt }}, "want_pdf": {{ export_pdf }}}'

  - id: transcribe
    type: asr
    dependencies: ["ingest", "asr_opts"]
    input_mapping:
      model: "{{ nodes.asr_opts.outputs.asr_model }}"
      audio_source: "{{ nodes.ingest.outputs.media_or_subtitles }}"
    parameters:
      response_format: srt

  - id: summary_prompt
    type: template
    dependencies: ["ingest", "transcribe"]
    input_mapping:
      video_title: "{{ nodes.ingest.outputs.title }}"
      duration_secs: "{{ nodes.ingest.outputs.duration_seconds }}"
      raw_srt: "{{ nodes.transcribe.outputs.srt }}"
      plain_text: "{{ nodes.transcribe.outputs.text }}"
    parameters:
      template: |
        请针对视频《{{ video_title }}》（时长约 {{ duration_secs | default(value=0) | round }} 秒）的字幕与转写文稿，使用「{{ summary_lang }}」深度提炼结构化视频精要：
        1. `video_title`：规范化后的视频标题；
        2. `executive_summary`：200-350 字高密度全景摘要，交代核心主题、背景脉络与最终结论；
        3. `key_viewpoints`：提炼 4-8 条核心观点，每条包含 `viewpoint`、`supporting_detail`（具体论据/案例/数据）与 `timestamp_hint`（对应出现的大致时间点，如 `01:25`）；
        4. `chapters`：按时间推进划分的章节大纲（`time_range`、`heading`、`summary`）；
        5. `memorable_quotes`：视频中最具启发性的金句或关键结论原话精译；
        6. `actionable_takeaways`：面向听众的可执行启发与落地建议；
        7. `mindmap_tree`：概括全片脉络的思维导图树（根节点为视频主题，3 层以内分支，标签简明凝练）。

        带时间轴的原始 SRT 字幕：
        {{ raw_srt | truncate(length=42000) }}

        纯文本参考：
        {{ plain_text | truncate(length=12000) }}

  - id: summarize_video
    type: llm
    dependencies: ["summary_prompt"]
    input_mapping:
      prompt: "{{ nodes.summary_prompt.outputs.output }}"
    parameters:
      max_tokens: 8192
      system: "你是专业的内容主编与知识萃取专家，严格依据视频真实字幕内容提炼观点与时间点，仅输出合法 JSON。"
      json_schema:
        type: object
        required:
          - video_title
          - executive_summary
          - key_viewpoints
          - chapters
          - actionable_takeaways
          - mindmap_tree
        properties:
          video_title: { type: string }
          executive_summary: { type: string }
          key_viewpoints:
            type: array
            minItems: 1
            maxItems: 10
            items:
              type: object
              required: [viewpoint, supporting_detail, timestamp_hint]
              properties:
                viewpoint: { type: string }
                supporting_detail: { type: string }
                timestamp_hint: { type: string }
          chapters:
            type: array
            minItems: 1
            maxItems: 10
            items:
              type: object
              required: [time_range, heading, summary]
              properties:
                time_range: { type: string }
                heading: { type: string }
                summary: { type: string }
          memorable_quotes:
            type: array
            items: { type: string }
          actionable_takeaways:
            type: array
            items: { type: string }
          mindmap_tree:
            type: object
            required: [label, children]
            properties:
              label: { type: string }
              children:
                type: array
                items:
                  type: object
                  required: [label]
                  properties:
                    label: { type: string }
                    children:
                      type: array
                      items:
                        type: object
                        required: [label]
                        properties:
                          label: { type: string }
                          children:
                            type: array
                            items:
                              type: object
                              required: [label]
                              properties:
                                label: { type: string }

  - id: polish_srt_prompt
    type: template
    dependencies: ["transcribe", "asr_opts"]
    run_if: "{{ nodes.asr_opts.outputs.polish_srt }}"
    input_mapping:
      raw_srt: "{{ nodes.transcribe.outputs.srt }}"
    parameters:
      template: |
        请将以下原始 SRT 字幕校对、断句优化并翻译/润色为标准「{{ summary_lang }}」SRT 字幕格式（严格保持 `序号` + `HH:MM:SS,mmm --> HH:MM:SS,mmm` + `字幕正文` + 空行的标准 SubRip 格式，不要包裹代码块标记）：

        {{ raw_srt | truncate(length=28000) }}

  - id: polish_srt_llm
    type: llm
    dependencies: ["polish_srt_prompt", "asr_opts"]
    run_if: "{{ nodes.asr_opts.outputs.polish_srt }}"
    input_mapping:
      prompt: "{{ nodes.polish_srt_prompt.outputs.output }}"
    parameters:
      max_tokens: 8192
      system: "你只输出严格合法的标准 SubRip (.srt) 字幕纯文本，绝不输出任何额外解释或 Markdown 代码围栏。"

  - id: mindmap_tree_json
    type: template
    dependencies: ["summarize_video"]
    input_mapping:
      summary: "{{ nodes.summarize_video.outputs.output }}"
    parameters:
      output_format: json
      template: '{"tree": {{ summary.mindmap_tree | json_encode() }}}'

  - id: mindmap_svg
    type: render_mindmap
    dependencies: ["mindmap_tree_json"]
    input_mapping:
      tree: "{{ nodes.mindmap_tree_json.outputs.tree }}"

  - id: mindmap_png
    type: render_svg
    dependencies: ["mindmap_svg"]
    input_mapping:
      svg: "{{ nodes.mindmap_svg.outputs.mindmap_svg_output }}"
    parameters:
      width: 1600

  - id: mindmap_html
    type: render_mindmap
    dependencies: ["mindmap_tree_json"]
    input_mapping:
      tree: "{{ nodes.mindmap_tree_json.outputs.tree }}"
    parameters:
      format: html

  - id: assemble_report
    type: template
    dependencies: ["ingest", "transcribe", "summarize_video", "mindmap_png", "polish_srt_llm"]
    input_mapping:
      has_native_subs: "{{ nodes.ingest.outputs.has_subtitles }}"
      duration_secs: "{{ nodes.ingest.outputs.duration_seconds }}"
      summary: "{{ nodes.summarize_video.outputs.output }}"
      mm_png: "{{ nodes.mindmap_png.outputs.mindmap_png_output }}"
      polished_srt: "{{ nodes.polish_srt_llm.outputs.output }}"
      raw_srt: "{{ nodes.transcribe.outputs.srt }}"
    parameters:
      template: |
        # {{ summary.video_title }} — 视频核心观点与字幕精要

        > **时长**：约 {{ duration_secs | default(value=0) | round }} 秒 · **字幕来源**：{% if has_native_subs %}原生内嵌/平台字幕直提{% else %}FFmpeg 音轨抽取 + ASR 语音转写 (`{{ asr_model }}`){% endif %}

        ## 1. 全片核心摘要

        {{ summary.executive_summary }}

        ## 2. 知识脉络思维导图

        ![视频脉络思维导图]({{ mm_png }})

        ## 3. 核心观点与论据提炼

        | 时间点 | 核心观点 | 关键论据 / 细节支撑 |
        | --- | --- | --- |
        {% for v in summary.key_viewpoints %}| `{{ v.timestamp_hint }}` | **{{ v.viewpoint }}** | {{ v.supporting_detail }} |
        {% endfor %}

        ## 4. 时间轴章节纪要

        {% for ch in summary.chapters %}
        ### `{{ ch.time_range }}` {{ ch.heading }}

        {{ ch.summary }}
        {% endfor %}

        {% if summary.memorable_quotes and summary.memorable_quotes | length > 0 %}
        ## 5. 关键金句摘录

        {% for q in summary.memorable_quotes %}> "{{ q }}"

        {% endfor %}
        {% endif %}

        {% if summary.actionable_takeaways and summary.actionable_takeaways | length > 0 %}
        ## 6. 实践启发与行动清单

        {% for a in summary.actionable_takeaways %}- {{ a }}
        {% endfor %}
        {% endif %}

        ## 7. 完整时间轴字幕（SRT）

        ```srt
        {% if polished_srt %}{{ polished_srt }}{% else %}{{ raw_srt }}{% endif %}
        ```

  - id: summary_md
    type: render_markdown
    dependencies: ["assemble_report"]
    input_mapping:
      markdown: "{{ nodes.assemble_report.outputs.output }}"
    parameters:
      format: md

  - id: summary_html
    type: render_markdown
    dependencies: ["assemble_report"]
    input_mapping:
      markdown: "{{ nodes.assemble_report.outputs.output }}"

  - id: summary_pdf
    type: render_document
    dependencies: ["assemble_report", "asr_opts"]
    run_if: "{{ nodes.asr_opts.outputs.want_pdf }}"
    input_mapping:
      markdown: "{{ nodes.assemble_report.outputs.output }}"
    parameters:
      format: pdf
      toc: true

  - id: deliverable
    type: template
    dependencies:
      - transcribe
      - summarize_video
      - mindmap_svg
      - mindmap_html
      - summary_md
      - summary_html
      - summary_pdf
    input_mapping:
      srt_path: "{{ nodes.transcribe.outputs.srt_file }}"
      summary: "{{ nodes.summarize_video.outputs.output }}"
      mm_svg: "{{ nodes.mindmap_svg.outputs.mindmap_svg_output }}"
      mm_html: "{{ nodes.mindmap_html.outputs.mindmap_html_output }}"
      md: "{{ nodes.summary_md.outputs.summary_md_output }}"
      html: "{{ nodes.summary_html.outputs.summary_html_output }}"
      pdf: "{{ nodes.summary_pdf.outputs.summary_pdf_output }}"
    parameters:
      output_format: json
      template: |
        {"deliverable": {
          "kind": "video_summarizer",
          "title": {{ summary.video_title | json_encode() }},
          "content": {{ summary | json_encode() }},
          "citations": [],
          "files": [
            {"node_id": "transcribe", "name": {{ srt_path | split(pat="/") | last | json_encode() }}, "mime_type": "application/x-subrip", "role": "primary"},
            {"node_id": "summary_html", "name": {{ html | split(pat="/") | last | json_encode() }}, "mime_type": "text/html", "role": "preview"},
            {"node_id": "summary_md", "name": {{ md | split(pat="/") | last | json_encode() }}, "mime_type": "text/markdown", "role": "asset"},
            {"node_id": "mindmap_svg", "name": {{ mm_svg | split(pat="/") | last | json_encode() }}, "mime_type": "image/svg+xml", "role": "asset"},
            {"node_id": "mindmap_html", "name": {{ mm_html | split(pat="/") | last | json_encode() }}, "mime_type": "text/html", "role": "preview"}{% if pdf %},
            {"node_id": "summary_pdf", "name": {{ pdf | split(pat="/") | last | json_encode() }}, "mime_type": "application/pdf", "role": "primary"}{% endif %}
          ]
        }}
