Designing a Short‑Video Script Skill: Storyboard Layout, Voiceover Copy, and Rhythm Control

Pain Points of Manual Short‑Video Script Writing

Short‑video scripts differ from text copy because they must coordinate three layers—visuals, sound, and text—simultaneously. A director at an MCN agency reported vague scene descriptions, missing hooks in the first five seconds, and flat pacing that caused many scripts to be rejected.

Core Challenge: Semi‑Structured Visual Description

Scene descriptions sit between pure natural language and pure parameter lists. Simple prose like "shoot a warm scene" leaves the cinematographer unsure, while pure parameters such as "mid‑shot/fixed/3000K" lose emotional tone. The solution is a "parameter + atmosphere" dual‑segment format.

Rhythm Alignment

A 60‑second video typically follows: 0‑3 s hook, 5‑15 s core content, 45‑55 s emotional lift, final 5 s call‑to‑action. Voiceover timing must match storyboard cuts; otherwise the script fails.

Video‑Type Specific Structures

Three video types require distinct pacing and visual styles:

Knowledge‑science : steady pace, infographics, soft BGM (90‑105 BPM).

Product‑grass‑root : fast pace (2‑3× knowledge), dynamic cuts, upbeat BGM (110‑125 BPM).

Drama short : narrative arcs, piano/strings BGM (60‑75 BPM).

Skill Design: Seven Modules

Module 1 – Metadata

name: short-video-script
description: |
   Short‑video shooting script generator. Input topic and video type,
   output full storyboard (visual + duration + voiceover), BGM suggestion,
   subtitle style for Douyin/Kuaishou/Bilibili vertical videos.
priority: 8
tags: [content creation, short video, script, storyboard]
version: 1.3.0

Module 2 – Trigger Conditions

triggers:
  keywords:
    - "short video script"
    - "storyboard"
    - "voiceover"
    - "shooting plan"
    - "video script"
  intent_patterns:
    - pattern: "help me write.*video.*script"
      confidence: 0.75
    - pattern: ".*storyboard.*design"
      confidence: 0.7
  negation_words:
    - "video title"
    - "video description"
    - "cover copy"
    - "long video"
    - "Bilibili column"

Module 3 – Input/Output Specification

inputs:
  required:
    - name: topic
      type: string
      description: "video topic or product"
    - name: video_type
      type: enum
      values: [knowledge, product, drama]
    - name: duration
      type: enum
      values: [15s, 30s, 60s]
      default: 30s
    - name: platform
      type: enum
      values: [Douyin, Kuaishou, Bilibili]
  optional:
    - name: key_points
      type: array
      description: "must‑cover information points"
outputs:
  - name: storyboard
    type: structured_array
    item_schema:
      - name: visual_description
        type: semi_structured_text
        format: "[subject] in [scene], [lighting], [composition]. Mood: [one sentence]"
        example: "Projector on bedside table, warm side light, product centered left. Mood: lazy weekend cinema."
      - name: voiceover
        type: string
        max_length: 50
      - name: bgm_suggestion
        type: structured_object
        fields: [style, tempo_bpm_range, cue_points]
      - name: subtitle_style
        type: structured_object
        fields: [font, size_ratio, position, animation, highlight_color]

Module 4 – Execution Instructions (Five‑Step Flow)

### Step 1: Script Structure Planning
# Choose template based on video_type
knowledge → hook 3s → problem 5s → core 15s → case 5s → summary 2s
product   → hook 3s → showcase 5s → selling 12s → scenario 7s → CTA 3s
drama     → suspense 3s → conflict 8s → twist 5s → climax 10s → reversal 4s

### Step 2: Storyboard Generation
Iterate paragraphs → assign shot type (avoid repeats) → assign camera move → generate visual_description in "parameter + atmosphere" format → generate voiceover (≤50 chars per shot) → generate on‑screen text

### Step 3: Three‑Layer Alignment ★
Time alignment: voiceover_word_count ÷ 4 chars / s ≤ shot_duration
Emotion alignment: voiceover emotional turn points = shot cut points
Info alignment: each key_point appears in ≥1 shot

### Step 4‑5: BGM Suggestion & Subtitle Style
Match BGM style/BPM/cue_points to video_type and tone
Match subtitle animation and highlight color to platform and BGM mood

Module 5 – Examples

Three example inputs (knowledge, product, drama) each produce a distinct storyboard structure, shot composition, BGM style, and subtitle animation.

Module 6 – Constraint Rules

Hard rules:

Each shot 2‑8 s (merge <2 s, split >8 s)

No consecutive shots with the same shot type

voiceover_word_count ÷ 4 ≤ shot_duration

visual_description must contain subject, scene, lighting, composition

First 3 s must contain a hook (question/conflict/contradiction/suspense)

Each key_point appears in at least one shot

Avoid absolute language (best/first/100%)

Soft rules:

Product videos use push/pan moves; knowledge videos use fixed/tilt

BGM cue points ≈ 1 : 1.5 of shot count

Subtitle highlight color matches BGM mood temperature

Module 7 – Version History

version: 1.3.0
changelog:
  - 1.0.0 # initial, only knowledge type, visual_description as plain text
  - 1.1.0 # added product and drama types; visual_description switched to dual‑segment
  - 1.2.0 # added three‑layer alignment and voiceover‑speed hard rule
  - 1.3.0 # added BGM cue alignment, subtitle animation per type, adjacent shot diversity

Real‑World Demonstration: Input‑Output Walkthrough

Input example for a portable juicer (product type, 30 s, Douyin, target fitness audience, key points: battery life, cleaning method, capacity). The generated storyboard includes six shots with timestamps, shot type, camera move, visual description, voiceover, and subtitle text.

- Shot 1 (00:00‑00:03, mid‑shot, move: pan)
  Visual: "Girl at gym entrance, natural light, subject on right. Mood: post‑work fatigue."
  Voiceover: "Finished workout, want fresh juice, a cup costs 28 ¥."
  Subtitle: "One cup = 28 ¥?"
- Shot 2 (00:03‑00:08, close‑up, move: push)
  Visual: "Juicer taken from bag, warm desk light, product centered. Mood: compact surprise."
  Voiceover: "This portable juicer, pack it and juice anywhere."
  Subtitle: "Portable juicer"
- Shot 3 (00:08‑00:15, close‑up, move: static)
  Visual: "Fruit placed in cup, top light, product left‑center. Mood: simple爽快."
  Voiceover: "Add fruit, press, juice in 30 s."
  Subtitle: "30 s juice"
- Shot 4 (00:15‑00:20, close‑up, move: push)
  Visual: "Juice poured, side backlight, liquid centered. Mood: fresh tempting."
  Voiceover: "One charge makes 12 cups."
  Subtitle: "12 cups per charge"
- Shot 5 (00:20‑00:25, mid‑shot, move: pull)
  Visual: "Tap rinses cup, natural light, product centered. Mood: easy cleaning."
  Voiceover: "Rinse in 10 s, done."
  Subtitle: "One rinse, clean"
- Shot 6 (00:25‑00:30, mid‑shot, move: static)
  Visual: "Girl exits gym with juicer, backlight, subject centered. Mood: energetic ending."
  Voiceover: "Fitness joy is that simple."
  Subtitle: "Order now ↓"

Efficiency Comparison

Total time : manual 2‑3 h → Skill 20‑30 s

Visual description executability : photographer asks 2‑3 follow‑ups → parameter + atmosphere format executes directly

Voiceover‑shot alignment : manual estimation often mismatched → automatic three‑layer check guarantees sync

Quality stability : manual pass rate ~50 % → Skill pass rate >80 %

Type adaptation : single template for all types → three distinct templates per video type

Pitfall Guide and Extensions

Trap 1: Dual‑End Visual Description

Pure prose lacks executability; pure parameters lose mood. Solution: use semi_structured_text with a parameter segment and a one‑sentence mood segment.

Trap 2: Voiceover Length vs. Shot Duration

30‑s video should contain ≤120 words (4 words/s). Add hard rule voiceover_word_count ÷ 4 ≤ shot_duration to auto‑prompt trimming.

Trap 3: Incomplete Example Coverage

Missing a video type in examples degrades that type’s output and contaminates others. Ensure examples cover every enum value of video_type with distinct pacing and BGM.

Extension ideas: combine this script Skill with a cross‑platform content‑adaptation Skill to turn the same script into a Xiaohongshu graphic note or a WeChat article, or link it to a PPT‑generation Skill where the storyboard becomes slide outlines.