Script & Story Development with AI

The AI Writers' Room

Professional screenwriting follows a funnel: concept to logline to treatment to outline to screenplay. Each stage compresses or expands the story. LLMs excel at every stage when you provide the right constraints. The key principle: **LLMs are not your writer. You are the writer. LLMs are your writers' room** -- they brainstorm, challenge, expand, and refine your ideas. The director's vision must come from you. Start every project with a concept document. Feed this to your LLM: ``` CONCEPT BRIEF Genre: Sci-fi noir Duration: 3 minutes Tone: Melancholic, contemplative Theme: Memory is unreliable Setting: Near-future Tokyo, perpetual rain Protagonist: Retired memory detective Core conflict: She discovers her own memories are fabricated Visual reference: Blade Runner meets Lost in Translation ``` From this brief, ask the LLM to generate ten loglines. Pick the strongest and iterate. A good logline for AI cinema is under 30 words and implies visual spectacle.

Three-Act Structure for Short Films

The classic three-act structure compresses beautifully into short form: **Act 1 (20% of runtime):** Establish world, character, and normal state. For a 3-minute film, this is 36 seconds. One or two shots that immediately communicate setting, mood, and protagonist. **Act 2 (60% of runtime):** Introduce conflict, escalate tension, build toward crisis. About 108 seconds. This is where your story lives. Three to five scenes with clear visual progression. **Act 3 (20% of runtime):** Resolution or revelation. 36 seconds. The twist, the emotional payoff, the final image that lingers. Here is a practical scene breakdown template: ``` SCENE BREAKDOWN - "Rain Memory" Runtime: 3:00 ACT 1 (0:00-0:36) Scene 1 (0:00-0:18): Establishing shot - Tokyo skyline, rain, neon. Slow push-in. Mood: isolation. Scene 2 (0:18-0:36): INT. Apartment - Keiko stares at a holographic photo. She touches it. It glitches. ACT 2 (0:36-2:24) Scene 3 (0:36-1:00): Keiko walks rain-soaked streets. Voice-over: "Every memory I extracted was real. Mine weren't." Scene 4 (1:00-1:30): FLASHBACK - A memory of a beach. Colors oversaturated. Something is wrong. Scene 5 (1:30-2:00): Keiko at her old office. Discovers her case file. Her face in the "subject" photo. Scene 6 (2:00-2:24): She runs. Rain. Neon reflections. The city feels like it is watching. ACT 3 (2:24-3:00) Scene 7 (2:24-2:48): Keiko stops running. Looks up. Rain on her face. Acceptance. Scene 8 (2:48-3:00): Wide shot - she is one of hundreds of identical figures in the rain. Pull out to black. ``` Each scene description doubles as the seed for your video generation prompt. Write scenes with visual precision from the start.

Writing Prompt-Ready Scene Descriptions

Traditional screenwriting separates action lines from technical direction. In AI cinema, you merge them. Every scene description must encode: 1. **Subject**: Who/what is in frame, with specific physical details 2. **Action**: What is happening, described as motion 3. **Setting**: Where, with atmospheric details 4. **Camera**: Shot type, movement, lens characteristics 5. **Mood**: Lighting, color palette, emotional tone Compare these approaches: ``` TRADITIONAL SCREENPLAY: INT. APARTMENT - NIGHT Keiko sits at her desk. Rain streams down the window. She picks up an old photograph and studies it. PROMPT-READY SCREENPLAY: INT. APARTMENT - NIGHT Medium shot, static camera. A Japanese woman in her 40s (short black hair, dark circles under eyes, grey sweater) sits at a cluttered desk lit by a single warm desk lamp. Rain streaks down a floor-to-ceiling window behind her, city lights blurred bokeh. She picks up a physical photograph with both hands, bringing it closer to her face. Her expression shifts from neutral to confused. Color palette: warm amber interior vs cool blue exterior. Aspect ratio: 2.39:1 anamorphic. ``` The second version can be fed almost directly to Kling or Runway with minor formatting changes. This is the core skill: writing prose that is simultaneously a screenplay and a generation prompt.

LLM Prompt Patterns for Screenwriting

Use these proven prompt patterns to get professional-quality output from any LLM: **The Genre Expert Pattern:** ``` You are a screenwriter who specializes in [genre]. Your influences are [Director A], [Director B], and [Director C]. Write in a style that combines [quality A] with [quality B]. ``` **The Constraint Pattern:** ``` Write a scene with these constraints: - Maximum 3 characters - Single location - Must convey [emotion] without dialogue - Visual storytelling only - Under 45 seconds of screen time ``` **The Iteration Pattern:** ``` Here is my scene. Rewrite it three ways: 1. More visually specific (add camera, lighting, color details) 2. More emotionally resonant (deepen the subtext) 3. More producible (simplify for AI video generation) ``` **The Dialogue Pattern (for voice-over):** ``` Write voice-over narration for this scene. Rules: Under 30 words. Present tense. First person. Tone: [specific adjective]. The character is [emotional state]. The narration must contrast with what we see on screen. ``` Chain these patterns: generate with the Genre Expert, refine with the Constraint Pattern, polish with the Iteration Pattern, add voice-over with the Dialogue Pattern.

Building Your Script Package

Before moving to storyboarding (Lesson 3), your script package should contain: 1. **Logline** (1 sentence) 2. **Treatment** (1 paragraph summary) 3. **Scene breakdown** (numbered scenes with timestamps) 4. **Prompt-ready scene descriptions** (all five encoding elements) 5. **Character reference sheet** (physical details, wardrobe, mannerisms) 6. **Voice-over script** (if applicable, timed to scenes) 7. **Visual reference list** (films, photographers, color palettes) Store everything in a single markdown file. This becomes your production bible -- every subsequent step references it.