product-promo-video-maker

name

product-promo-video-maker

description

Generate a complete product promotional video (HTML analysis page + screen recording + voiceover + subtitles) from any product input. Use when: (1) creating a product promo video with structured analysis, (2) building a marketing video with narration and subtitles, (3) generating a product showcase with selling points and visual effects, (4) producing a distribution-ready MP4 from product info. Input: product URL, description, or files. Output: analysis webpage + final_with_voice.mp4. Not for: simple slideshow videos without structured analysis.

Product Promo Video Maker

Generate a complete, distribution-ready product promotional video from any product input.

Input

Three input types supported:

1. Product URL — e.g. https://enterprise.dji.com/cn/mavic-3-enterprise
2. Product description — plain text with name, specs, features
3. Local files — images + docs dropped by user

Output

deliver/
  ├── final_video.mp4     # Promo video with voiceover + matching timing
  ├── webpage/            # Analysis webpage (open in browser)
  │   ├── index.html
  │   └── images/
  ├── narration.mp3       # Voiceover audio
  └── subtitles.srt     # Subtitle file

Pipeline (9 stages)

INGEST → EXTRACT → ANALYZE → DESIGN → BUILD → CAPTURE → VOICE → COMPOSE → DELIVER

Stage 1: INGEST

Read user input. If URL: use browser to extract text/images. If description: save raw text. If files: index paths.

Output: 01-input/product-info.json

Stage 2: EXTRACT

Use LLM with JSON Schema to extract structured data:

• name, product_name, slogan, brand, category
• core_specs: key-value map
• core_specs_highlight: array of {key, value} for hero stats display
• key_features: array of {name, priority, icon, image}
• pain_points: {before: [], after: []}
• scenarios: array
• value_chain: array of {icon, title, desc}

Output: 02-extracted/product-data.json

Stage 3: ANALYZE

Call the configured analysis framework (default: manufacturing-ai-efficiency-pro).

Generates:

• opportunities: array of 8-field opportunity cards
• painpoint_comparison: before/after lists + metrics
• flowchart_mermaid: Mermaid diagram code
• framework_steps: 9-step detailed breakdown

Output: 03-analysis/opportunities.json

Stage 4: DESIGN

Select template config based on user parameters.

Output: 04-design/config.json + 04-design/images-manifest.json

Stage 5: BUILD

Render index.html using template system. Inline all CSS/JS. Fetch product images.

Output: 05-webpage/index.html + images/

Template variables:

{{product.name}} → product_data["name"]
{{hero.hero_image}} → images_manifest["hero_image"]
{{framework.steps}} → opportunities["framework_steps"]
{{opportunities.list}} → opportunities["opportunities"]
{{valuechain.nodes}} → product_data["value_chain"]

Stage 6: CAPTURE

Use Playwright (scripts/capture.py) to record webpage scrolling through all sections.

Output: 06-capture/raw-video.mp4

Default section timing:

Section	Duration
hero	5s
framework	8s
painpoint	6s
opportunities	22s (scrolls 4 cards)
flowchart	4s
valuechain	4s
brand	4s

Stage 7: VOICE

Generate narration audio using SiliconFlow TTS (scripts/voice.py). System preset voice (alex) with automatic speed variation per section type. Produce subtitle timestamps.

Output: 07-voice/narration.mp3 + 07-voice/subtitles.srt

Default TTS config:

• Provider: SiliconFlow (FunAudioLLM/CosyVoice2-0.5B)
• Voice: alex (沉稳男声, system preset)
• Alternatives: benjamin(低沉), charles(磁性), david(欢快), anna(沉稳女声), bella(激情女声), claire(温柔), diana(欢快女声)
• Speed mapping: hero 0.95x / painpoint 1.05x / opportunity 1.0x / value 0.9x / brand 0.85x
• API key required: siliconflow_api_key in config or SILICONFLOW_API_KEY env var

Stage 8: COMPOSE

Use ffmpeg to merge video + audio. Video speed adjusted to match audio duration.

Output: 08-compose/final_video.mp4

Note: Subtitle burn-in on Windows has ffmpeg path compatibility issues. Current solution: deliver .srt as separate file for editor import or player loading.

Stage 9: DELIVER

Copy all artifacts to clean deliver directory.

Configuration

Parameter	Default	Options
analysis_framework	manufacturing-ai-efficiency-pro	See references/analysis-frameworks.md
output_format	landscape	landscape (1920×1080), portrait (720×1280)
theme	dark-industrial	dark-industrial, robot-theme
tts.provider	siliconflow	siliconflow, none
tts.siliconflow_api_key	(required for voice)	Get from cloud.siliconflow.com
tts.voice	alex	alex, benjamin, charles, david, anna, bella, claire, diana
tts.fallback_provider	edge-tts	edge-tts (local, no key needed but lower quality)
include_subtitles	true	true, false
language	zh-CN	zh-CN, en-US
duration_target	60	Any integer (seconds, used for pacing guidance)

Tool Dependencies

• browser — Product page extraction
• web_search / kimi_search — Image search fallback
• read / write — File operations
• Playwright — Webpage recording (chromium channel with msedge)
• ffmpeg — Video composition (required)
• SiliconFlow API key — TTS voice generation (optional, see "开箱即用" below)

开箱即用 / Out-of-the-Box Usage

模式 A：完整语音版（推荐）

1. 注册 SiliconFlow 获取 API key（免费额度足够测试）
2. 复制 config.template.json 为 config.json，填入 siliconflow_api_key
3. 运行 pipeline → 自动出带语音 + 字幕的视频

模式 B：无 API key 版（纯视频 + 字幕文本）

• 不填 key → 跳过语音合成，出视频 + subtitles.srt + narration.txt
• 用户可：① 自己配音后导入剪辑软件 ② 用其他 TTS 工具生成音频后合成

必需依赖：

• ffmpeg（配置 ffmpeg_path 和 ffprobe_path）
• Python 3.10+ + requests 库
• Playwright（用于录屏）

可选依赖：

• SiliconFlow API key（模式 A）
• edge-tts pip 包（旧版 fallback，不推荐）

File Structure

product-promo-video-maker/
├── SKILL.md                          # This file
├── assets/
│   ├── templates/
│   │   ├── base.html                 # HTML skeleton
│   │   ├── styles.css               # CSS (supports dark-industrial + robot-theme)
│   │   ├── scripts.js               # Particle canvas + scroll animations
│   │   └── sections/
│   │       ├── hero.html            # Product showcase + stats
│   │       ├── framework.html       # 9-step timeline with details
│   │       ├── painpoint.html       # Before/after comparison
│   │       ├── opportunities.html   # 8-field opportunity cards
│   │       ├── flowchart.html       # Mermaid container
│   │       ├── valuechain.html      # Value chain nodes
│   │       ├── emotion.html         # SVG emotion curve
│   │       └── brand.html           # Brand closing
│   └── themes/
│       ├── dark-industrial.css      # Default dark theme (blue accent)
│       └── robot-theme.css          # Robot/tech theme (orange accent)
├── scripts/
│   ├── pipeline.py                  # Main 9-stage orchestrator
│   ├── render.py                     # Template rendering engine
│   ├── capture.py                    # Playwright webpage recording
│   └── voice.py                      # TTS + subtitle generation
└── references/
    ├── analysis-frameworks.md         # Available frameworks + schemas
    ├── pipeline-guide.md             # Stage-by-stage execution details
    ├── template-guide.md             # Template rendering guide
    ├── template-variables.md         # Complete variable reference
    └── subtitle-guide.md            # Subtitle generation + burn-in notes

Usage Example

User: "帮我做一个 DJI Mavic 3 行业版的宣传视频"
→ Skill triggers
→ Stage 1: Extract product info from URL
→ Stage 3: Run manufacturing-ai-efficiency-pro framework → 7 opportunity cards
→ Stage 5: Build analysis webpage with dark-industrial theme
→ Stage 6: Record 60s scrolling video via Playwright
→ Stage 7: Generate narration + subtitles (zh-CN-XiaoxiaoNeural)
→ Stage 8: Compose final video with speed-matched timing
→ Stage 9: Deliver to deliver/dji-mavic3-20260512/

Known Limitations

1. Subtitle burn-in on Windows: ffmpeg subtitles filter requires SRT in same directory. Use force_style parameter with care (avoid & in shell).
2. Image quality depends on source: Product page images may be low-res. Recommend using official store/press kit images when available.
3. TTS prosody: SiliconFlow system presets (alex/benjamin/etc) are stable and clean but don't clone user's personal voice. For personal voice cloning: requires FishAudio API key or local GPU deployment (F5-TTS / GPT-SoVITS / IndexTTS).
4. Framework coverage: Only manufacturing-ai-efficiency-pro fully implemented. SaaS / consumer frameworks have schemas but not full prompts.

Development History

• v1.0 (2026-05-12) — Initial release based on DJI Mavic 3 and Unitree As2 demos

  • Validated end-to-end pipeline with 2 real products
  • Support: landscape output, Edge TTS, dark-industrial + robot themes
  • Template system: 8 sections, Mustache-style variable syntax

• v1.1 (2026-05-14) — TTS migration: Edge TTS → SiliconFlow system presets

  • Replaced Edge TTS with SiliconFlow FunAudioLLM/CosyVoice2-0.5B system presets
  • Added automatic speed variation: painpoint 1.05x, brand 0.85x, hero 0.95x, etc.
  • Added subtitle burn-in support (ffmpeg subtitles filter)
  • Added no-key fallback mode: skips TTS, outputs video + SRT + narration.txt
  • Added config.template.json for easy configuration
  • Removed rhythm parasitism issue by using system presets (no voice cloning)
  • Known limitation: personal voice cloning requires FishAudio API or local GPU

References

• references/pipeline-guide.md — Stage-by-stage execution details + checkpoint logic
• references/analysis-frameworks.md — Available frameworks, their schemas, and LLM prompts
• references/template-guide.md — How to render templates, customize themes, add new sections
• references/template-variables.md — Complete variable reference for all sections
• references/subtitle-guide.md — Subtitle generation, burn-in workarounds, editor import