Create News Video

Create News Video Skill

Generate a Vietnamese 9:16 motion-graphic news video from a URL or .txt file.

Input

Single argument: a news article URL (starts with http:// or https://) OR a path to a .txt file.

Workflow (MUST follow these steps in order)

Step 1: Detect input type

• Starts with http:// or https:// → URL mode
• Otherwise → file mode

Step 2: Fetch content

URL mode:

• Use WebFetch with prompt:

  Trích xuất từ trang này:
  - title (string): tiêu đề bài báo
  - content (string): nội dung chính, ~500-1500 từ
  - ogImage (string|null): URL ảnh og:image (meta og:image hoặc ảnh đầu bài)
  - domain (string): domain của URL (vd "vnexpress.net")
  Trả về JSON với 4 field trên.

• If WebFetch fails (paywall, JS-rendered, 4xx) → tell user to save content to a .txt file and pass that instead. Stop.

File mode:

• Use Read to read the .txt file
• Title = first non-empty line (strip whitespace, max 80 chars)
• Content = remaining lines joined
• ogImage = null
• domain = "local"

Step 3: Create slug + output directory

• slug = lowercase ASCII (strip Vietnamese diacritics, đ→d), replace non-alphanumeric with -, trim dashes, max 40 chars
• timestamp = current local time as YYYYMMDD-HHmm
• outputDir = output/<slug>-<timestamp>/
• Use Bash: mkdir -p <outputDir>

Step 4: Generate script.json

Following the schema in docs/superpowers/specs/2026-04-29-auto-news-video-design.md Section 4. Key rules:

Script content (Vietnamese):

• Total voiceText: ~150–200 words → ~55–65s spoken at speed 1.0
• Number of scenes: 5–8 (1 hook + 3–6 body + 1 outro)
• Each scene voiceText is 1-3 short sentences, văn nói (spoken style, not formal)
• No emoji, no markdown in voiceText

⚠️ CRITICAL: Vietnamese TTS Phonetic Rules

The voiceText field is read aloud by LucyLab/ElevenLabs Vietnamese TTS. Numbers and symbols are read literally — if you write "5.5", TTS may say "năm rưỡi" (five and a half — WRONG for version numbers). Always spell out numbers in Vietnamese phonetic form in voiceText. The templateData fields (visual text on screen) can keep the original "5.5" / "82.7%" formatting.

Mandatory rules for voiceText:

Notation choices:

• For decimal point use chấm (more spoken/natural) or phẩy (formal). Both work; pick consistent.
• For comma separator, use phẩy (e.g. "1,000" → "một nghìn")
• For ratio "3:1" → say ba trên một or ba so với một

English brand names — keep as-is, TTS handles them OK:

• Apple, Google, OpenAI, Microsoft, TikTok, YouTube ✅

English acronyms — write phonetically if TTS misreads:

• AI → usually OK, sometimes write ây ai for clarity
• API → write ây pi ai if matter
• GPT → usually OK; if not, write gí pi tí
• iOS → write ai ô ét if matter

Symbols to AVOID in voiceText:

• → & % $ # + = (TTS may say literal name or skip)
• ! ? at end of sentence is OK — they create natural intonation
• Emoji: NEVER (TTS pronounces or skips inconsistently)
• URLs: NEVER (TTS reads dot/slash literally)

End each voiceText sentence with . or ? for natural pause/intonation.

Examples — full scene:

WRONG (will sound bad):

{ "voiceText": "GPT 5.5 đạt 82.7% trên Terminal-Bench, vượt GPT 5.4 (75.1%)." }

→ TTS reads: "GPT năm rưỡi đạt tám mươi hai chấm bảy phần trăm trên Terminal-Bench..."

RIGHT (natural):

{ "voiceText": "GPT năm chấm năm đạt tám mươi hai phẩy bảy phần trăm trên Terminal Bench, vượt phiên bản năm chấm bốn ở mức bảy mươi lăm phẩy một." }

Note: templateData (text on screen) CAN use original formatting — the visual is separate from spoken:

{
  "voiceText": "GPT năm chấm năm đạt tám mươi hai phẩy bảy phần trăm.",
  "templateData": {
    "template": "stat-hero",
    "value": "82.7%",                 ← Visual: keep readable formatting
    "label": "Terminal-Bench"
  }
}

Hook (most important — gets first 3 seconds of viewer attention):

• Must contain a claim, statistic, or curious question
• NEVER generic ("Hôm nay chúng ta sẽ nói về..." is wrong)
• ALWAYS include at least 1 effect: flash-white-3f or particle-burst

Visual rules:

• For image scenes: background.src = "$source.image" (literal — CLI substitutes)
• Vary kenBurns across scenes (don't use zoom-in for every scene)
• Vary text animation (don't use slide-up for every line)
• Each line ≤ 25 characters
• Each scene 1-3 lines

Outro (always fixed format):

{
  "id": "outro",
  "type": "outro",
  "voiceText": "Theo dõi Công nghệ 24h để xem bản tin mới mỗi ngày.",
  "visual": {
    "background": { "type": "gradient", "preset": "outro-purple" },
    "text": {
      "position": "center",
      "style": "outro-card",
      "lines": [
        { "content": "Xem bản tin mới mỗi ngày", "emphasis": "primary", "animation": "fade-in" },
        { "content": "Công nghệ 24h",            "emphasis": "channel", "animation": "scale-pop" },
        { "content": "Nguồn: <DOMAIN>",          "emphasis": "muted",   "animation": "fade-in-late" }
      ]
    }
  }
}

Replace <DOMAIN> with the actual domain string. Note: outro line 1 is shortened to fit 25-char schema rule (full CTA "Theo dõi để xem bản tin mới mỗi ngày" is 36 chars).

Step 5: Self-validate before writing

Check:

• Total word count ~150-200
• Every line.content ≤ 25 chars
• 5-8 scenes total
• scenes[0].type === "hook"
• last scene type === "outro"
• All enum values valid (see spec Section 4.2)

If invalid, fix yourself silently. Up to 2 self-correction passes. After that, write anyway — the CLI's Zod validation will produce a precise error message that the user can act on.

Step 6: Write script.json

Use the Write tool (not Bash) to write the validated JSON to <outputDir>/script.json.

Step 7: Run the pipeline

Use Bash, foreground (not background), stream output:

npm run pipeline -- <outputDir>/script.json

If exit code != 0:

• Report the error message clearly
• Tell user the output dir path so they can inspect intermediate files

Step 8: Report success

If successful, report to user with markdown links:

✓ Video:  [video.mp4](output/<slug>-<timestamp>/video.mp4)
✓ Audio:  [voice.mp3](output/<slug>-<timestamp>/voice.mp3) — for CapCut
✓ Script: [script.txt](output/<slug>-<timestamp>/script.txt) — for CapCut auto-caption
Tổng thời lượng: XX.Xs

Examples