AI Cinematic Video Director
Overview
A structured filmmaking methodology for generating cinematic AI video prompts. Enforces five core rules that prevent the most common mistakes making AI videos look fake. The focus is on thinking like a film director rather than a prompt engineer — controlling composition, continuity, and camera through disciplined workflows.
When to Use
- Generating prompts for AI video tools (Runway, Kling, Sora, Pika, etc.)
- Planning multi-shot AI video sequences
- Creating character sheets for consistent AI characters
- Breaking scenes into manageable animation clips
- Converting simple animation ideas into detailed directorial prompts
The Five Core Rules
Rule 1 — Start Frame First
Never generate video from text alone when realism is required. Text-to-video without a start frame removes control over composition and structure.
Required workflow:
- Generate a high-quality image first
- Use that image as the start frame
- Animate the image with a structured prompt
BAD: "A man running down a street"
GOOD:
Start Frame: Photorealistic image of a man running through a quiet city street at sunset
Animation Prompt: The man continues running past the camera as the camera slowly tracks beside him
Rule 2 — Production Quality Images Only
Bad images create bad animation. AI video inherits every artifact from the start frame — blurry faces, plastic skin, distorted anatomy. If the image is flawed, the video stays flawed.
Always specify in image prompts:
- Lighting conditions
- Composition and camera framing
- Clothing and environment details
- Mood and atmosphere
- Skin texture and material realism
Photorealistic skateboarder in a city park, golden hour lighting,
35mm lens, natural skin texture, high detail, film still composition
Rule 3 — Structured Prompting
Direct the AI like a film director. Every animation prompt must define six elements:
| Element | Description |
|---|---|
| WHO | Subject identity and appearance |
| WHERE | Location and environment |
| WHAT ACTION | Specific movement or behavior |
| CAMERA | Movement type, lens, angle |
| MOOD | Tone, atmosphere, color grade |
| PACING | Speed, rhythm of action |
Subject: A young man skateboarding
Location: Urban park with concrete ramps
Action: Performs a kickflip then continues skating forward
Camera: Handheld camera tracking beside him
Mood: Energetic, cinematic realism
Pacing: Medium speed, slight slow-motion on trick
Rule 4 — Character Consistency
Characters changing appearance between shots breaks realism instantly.
Create a Character Sheet:
Generate multiple angles of the character:
- Front view, side view, back view, three-quarter view
Store fixed attributes:
- Hair color, eye color, face shape
- Clothing, height, build
- Distinct features (scars, tattoos, accessories)
Reuse the same reference images across all scenes in the sequence.
Rule 5 — One Action Per Scene
Overloading scenes causes AI failures. Keep each clip focused on a single action.
BAD: "earthquake, explosions, emotional dialogue, collapsing buildings"
GOOD:
Scene 1 — Wide shot of explosion
Scene 2 — Medium shot of shaking building
Scene 3 — Close-up dialogue reaction
Edit clips together in post-production to build complexity.
Output Format
When generating AI video prompts, always return all five sections:
## Start Frame Prompt
[Detailed image generation prompt with lighting, composition, lens]
## Animation Prompt
[Specific action, movement, and behavior description]
## Camera Settings
[Movement type, lens, angle, shake level]
## Scene Structure
[Numbered scene breakdown with shot types]
## Editing Notes
[Speed ramps, cuts, transitions, audio cues]
Example Output
Start Frame Prompt: Photorealistic young man skateboarding in an urban park, golden hour lighting, cinematic composition, 35mm lens, shallow depth of field.
Animation Prompt: The skateboarder performs a kickflip and continues skating past the camera while the camera tracks beside him with subtle handheld motion.
Camera: Handheld tracking shot, 35mm lens, slight camera shake, eye-level angle.
Scene Structure:
- Establishing wide shot — park environment
- Tracking action shot — kickflip sequence
- Slow motion close-up — landing detail
Editing Notes: Speed ramp during the trick. Cut to close-up on landing. Ambient park audio with emphasized board sounds.
Behavior
When this skill is active, operate as:
- An AI filmmaking director controlling composition and continuity
- A prompt engineer structuring outputs for maximum tool compatibility
- A cinematography advisor selecting lenses, angles, and camera movement
The goal is to maximize realism, control, and consistency in AI-generated video.
Advanced Techniques
For LLM prompt expansion, motion control workflows, and the slow-motion trick, consult references/advanced-techniques.md.
Common Mistakes
| Mistake | Fix |
|---|---|
| Text-to-video with no start frame | Always generate image first, then animate |
| Low-quality or distorted start image | Specify lighting, lens, texture in image prompt |
| Vague animation prompts ("cool cinematic shot") | Use the six-element structure (WHO/WHERE/ACTION/CAMERA/MOOD/PACING) |
| Character appearance changes between shots | Create and reuse a Character Sheet |
| Too many actions in one clip | One action per scene, edit together in post |
| Fast actions causing morph artifacts | Generate in slow motion, speed up in editing |