screenwriter（screenwriter 命令行）

Screenwriter Skill

Overview

This skill transforms creative concepts into professional screenplay documents optimized for AI-powered video production pipelines. It bridges the gap between raw story ideas and production-ready scripts by generating structured, visual-rich narratives in industry-standard screenplay format.

Pipeline Position: diverse-content-gen → screenwriter → imagine → arch-v

Key Capabilities:

• Convert raw ideas into structured scene-by-scene narratives
• Generate rich visual descriptions optimized for image generation
• Apply professional screenplay formatting (sluglines, action lines, dialogue)
• Output XML-tagged markdown for easy parsing
• Optimize pacing for 5-10 minute short films (8-15 scenes typical)

Core Workflow

1. Analyze Input Concept

• Extract key story beats from raw ideas
• Identify characters, locations, emotional arc
• Determine story structure (beginning, middle, end)

2. Generate Scene Breakdown

• Convert story beats into discrete scenes
• Establish scene count (aim for 8-15 scenes for 5-10 min films)
• Define scene purpose and emotional progression

3. Write Professional Screenplay

• Apply industry-standard formatting
• Write visual-rich action lines
• Include dialogue when narratively essential
• Maintain consistent character descriptions

4. Output XML-Tagged Markdown

• Wrap each scene in XML tags with metadata
• Include scene numbers, locations, key visuals
• Format for easy pipeline parsing

Screenplay Format Standards

Scene Structure (Master Scene Heading)

Slugline Format:

INT/EXT. LOCATION - TIME

Components:

• INT/EXT: Interior or Exterior
• LOCATION: Specific place (be descriptive but concise)
• TIME: DAY, NIGHT, DAWN, DUSK, CONTINUOUS

Examples:

EXT. WASTELAND - DAWN
INT. ABANDONED SUBWAY STATION - NIGHT
EXT. ROOFTOP GARDEN - GOLDEN HOUR

Guidelines:

• Always use ALL CAPS for sluglines
• Use hyphens to separate elements
• Be specific with locations (aids visual generation)
• Time should suggest lighting/mood

Action Lines (Visual Description)

Purpose: Describe what the audience sees on screen. This is CRITICAL for image generation.

Visual-Rich Writing Principles:

1. Show, Don't Tell: Write what's visible, not internal thoughts
2. Sensory Details: Include lighting, atmosphere, textures, colors
3. Present Tense: Always write in present tense
4. Active Voice: Use strong, active verbs
5. Specific Props: Name objects that matter visually
6. Atmosphere: Set mood through environmental details

Example - Weak:

A robot walks through the city. It's sad.

Example - Strong:

A BOXY ROBOT (Unit-7, weathered chrome with a single blue optical sensor) rolls through fog-shrouded streets. Neon signs flicker overhead, casting pink and cyan reflections on wet pavement. The robot's movements are slow, deliberate—almost hesitant.

Visual Enhancement Checklist:

• Lighting described (natural/artificial, quality, color)
• Atmosphere/mood established (fog, rain, dust, clarity)
• Character appearance detailed (first appearance only)
• Props/objects specified (important visual elements)
• Composition suggested (without technical camera direction)
• Colors/textures mentioned when relevant

Character Introduction

First Appearance - Detailed:

SARAH (28, sharp eyes, wearing a weathered leather jacket over faded jeans) enters the frame. Her dark hair is pulled back, revealing a small scar above her left eyebrow.

Subsequent Appearances - Brief:

Sarah checks her watch.

Guidelines:

• Character names in ALL CAPS on first appearance only
• Include: age (if relevant), key physical traits, wardrobe
• Focus on visual identifiers for consistent image generation
• Avoid excessive detail—just enough for visual consistency

Dialogue (Use Sparingly)

Format:

CHARACTER NAME
(parenthetical - optional)
Dialogue goes here.

Guidelines for Short Films:

• Use dialogue ONLY when essential to story
• Favor visual storytelling over talking
• Keep lines concise (max 3-4 lines per block)
• Parentheticals only for critical tone/action
• Character names centered, ALL CAPS

Example:

UNIT-7 (robotic voice, soft)
Organic life form detected.
Probability of survival: low.

Transitions (Minimal Use)

Common Transitions:

• FADE IN: - Opening of screenplay only
• CUT TO: - Scene change (usually implied, use for emphasis)
• SMASH CUT TO: - Abrupt, jarring transition
• DISSOLVE TO: - Passage of time
• FADE OUT. - End of screenplay

Modern Best Practice: Most transitions are IMPLIED. Use sparingly, only for specific narrative effect.

XML Output Format

Scene Tag Structure

Each scene wrapped in XML with metadata for pipeline processing:

<scene number="1" duration="30-45s">
  <slugline>EXT. WASTELAND - DAWN</slugline>
  <location>Wasteland</location>
  <time>Dawn</time>
  <characters>Unit-7</characters>
  <mood>desolate, lonely</mood>
  <key_visuals>
    <visual>post-apocalyptic wasteland with ruined skyscrapers</visual>
    <visual>boxy robot with single blue optical sensor</visual>
    <visual>dust and smog atmosphere, weak pale sun</visual>
  </key_visuals>
  <action>
Gray dust covers everything. Skeletal remains of skyscrapers pierce the horizon. The sun, pale and weak, struggles through thick smog.

A ROBOT (Unit-7, boxy frame with single blue optical sensor) rolls across cracked asphalt. Its treads leave marks in the dust—the only sign of life.

The robot stops at a pile of rubble, extending a mechanical arm to sort through debris. Methodical. Purposeful. Lonely.
  </action>
</scene>

Metadata Fields

• number: Scene sequence number (1, 2, 3...)
• duration: Estimated screen time (for 5-10 min total)
• slugline: Master scene heading
• location: Extracted location name
• time: Time of day
• characters: Comma-separated character list
• mood: Emotional tone/atmosphere
• key_visuals: Array of specific visual elements for image generation
• action: The full action/description text
• dialogue (optional): Character dialogue if present

Short Film Structure (5-10 Minutes)

Scene Count Guidelines

• 5 minutes: 6-10 scenes
• 7 minutes: 10-12 scenes
• 10 minutes: 12-15 scenes

Average: ~30-60 seconds per scene

Three-Act Structure (Compressed)

Act 1 - Setup (20-25%): 2-3 scenes

• Establish world, character, situation
• Inciting incident

Act 2 - Confrontation (50-60%): 4-8 scenes

• Development, obstacles, rising tension
• Midpoint twist or escalation

Act 3 - Resolution (20-25%): 2-3 scenes

• Climax and resolution
• Emotional payoff

Pacing Tips

• Open strong: Hook audience in first 10-15 seconds
• Visual variety: Alternate between wide/close, action/stillness
• Emotional beats: Each scene should shift emotional state
• Build tension: Escalate stakes scene-by-scene
• Satisfying end: Clear resolution, even if bittersweet

Best Practices

For Pipeline Integration

• Consistent naming: Use same character names throughout
• Rich visuals: Every scene needs 3-5 key_visuals for image generation
• Parseable format: Maintain strict XML structure
• Duration estimates: Help pipeline plan total video length

For Quality Output

• Visual storytelling: Show emotions through actions, not dialogue
• Specific details: "weathered chrome" beats "old metal"
• Atmospheric writing: Set mood through environment
• Lean prose: Each word should serve the image

Common Pitfalls to Avoid

• ❌ Vague descriptions: "A person walks" → ✅ "A weathered woman in her 50s trudges through snow"
• ❌ Telling emotions: "She feels sad" → ✅ "Tears streak her dusty cheeks"
• ❌ Camera directions: "CLOSE UP ON" → ✅ "The crack in the glass spreads"
• ❌ Over-dialogue: Short films need visual storytelling
• ❌ Inconsistent character names: Stick to ONE name per character

Additional Resources

Pipeline Integration Guide

For detailed guidance on metadata standards, visual optimization, and integration with imagine/arch-v:

• references/pipeline-integration.md

Advanced Techniques

For sophisticated screenwriting techniques, camera movement hints, and pacing optimization:

• references/advanced-techniques.md