The AI Video Prompt Engineering Master Handbook
A platform-neutral production guide for Text-to-Video and Image-to-Video. Stop AI glitches, master camera kinematics, control hand anatomy, and direct cinema-grade multi-clip sequences in Botock, Sora, Kling, and Runway Gen-3.

Specialized Deep-Dive Hubs (Cluster Guides)
3 Modular GuidesFor granular prompt lists, specialized physics tables, and direct copy-paste syntax, explore each dedicated deep-dive guide:

Camera Movements & Angles Guide
Pans, tilts, dollies, tracking shots, vertigo zoom, Dutch angles, and speed calibration tables.

Actions & Body Gestures Guide
Fix 6-finger flaws, anchor hand grips, eyeline shifts, posture transitions, and fight scene action beats.

Cinematic Lighting & Optics Guide
Anamorphic bokeh, focal length selection (24mm vs 85mm), golden hour rim lighting, and volumetric fog.
The 9 Core Blocks of an AI Video Prompt
Every high-fidelity prompt in modern generative models (Botock, Kling 1.5, Sora, Luma Dream Machine, Runway Gen-3) is constructed from nine distinct semantic building blocks. Leaving any block undefined forces the model to hallucinate random values.
Shot Size & Framing
Wide shot, full shot, medium close-up, macro.
Subject Identity & Anchor
Exact age, gender, clothing texture, hair color.
Starting Coordinate Pose
Seated at frame-left, facing frame-right.
The Single Action Beat
Only ONE atomic physical change during clip.
Movement Vector & Speed
Direction, displacement, velocity (e.g. slow, 1 m/s).
Camera Kinematics & Lens
Tripod static, dolly in, tracking shot, 50mm lens.
Lighting & Environment
Golden hour rim light, volumetric fog, rainy reflections.
Audio / Dialogue / Silence
Exact line in quotes, room tone, "No music".
Exclusion Constraints
No camera motion, no talking, no smiling.
The Subject Map: Audience Perspective vs. Character Perspective
The #1 reason AI models place characters or actions on the wrong side is ambiguous directional language. Use the standard spatial conventions below:
| Directive Term | Precise Spatial Meaning | Handbook Writing Rule |
|---|---|---|
| frame-left / frame-right | Left and right as the AUDIENCE views the display screen. | Always use "frame-left" to anchor positions on screen. |
| character's right hand | The character's biological anatomical hand. (Appears on frame-left if facing camera). | Write "the character's right hand" for body mechanics. |
| moves toward frame-right | Linear motion vector across the screen. | Never say "moves right". Always qualify with "frame-right". |
| Person A, Person B | Neutral semantic identifiers to prevent role blending. | Give each character a distinct identifier and unique wardrobe color. |
Motion Intensity Calibration Table
AI video generators tend to over-animate emotions and gestures. To control subtle acting, calibrate with exact intensity vocabulary:
| Intensity Keyword | Target Motion Scope | Cinematic Effect |
|---|---|---|
| Imperceptible | A tiny shift restricted only to the iris and pupil. | Subconscious tension, suppressed panic, deep concentration. |
| Subtle | Quiet, realistic human acting; slight corner-of-mouth lift. | Photorealistic natural drama; avoids cartoonish grins. |
| Moderate | Clear to any viewer; deliberate nodding or head turn. | Standard conversational action and reaction shots. |
| Restrained | High emotional intensity held under strict muscular control. | Grief, jaw clenching, anger held back. |
Dialogue, Voice & Sound Design Syntax
Modern video generation platforms integrate audio generation directly. Without strict sound constraints, models generate generic synth music and unintelligible gibberish chatter:
- • One speaker only: Never request two people talking in a 5-second clip.
- • Exact quoted lines: Place spoken dialogue in exact quotes:
"We have to leave now." - • Vocal delivery: Specify pitch, speed, and timbre:
whispered in a low husky voice.
- • No music: Always state
AUDIO: No music. Ambient room tone only.to stop cheap synth loops. - • No dialogue: Prevents silent characters from moving their lips like fish.
- • Foley precision: Name exact physical impacts:
heavy leather boots crunching on wet gravel.
Production Template Library (T1 - T25)
25 TemplatesPlug-and-play prompt structures taken straight from Chapter 23 of the handbook. Replace the bracketed variables with your narrative details:
CAMERA: Locked-off tripod, zero movement, zero tilt, zero pan.
ACTION: [Character] shifts eyes from [Position A] to [Position B] over 3 seconds and stops. Nothing else changes.
NEGATIVE: No mouth movement. No blinking. No body posture shift.
ACTION: Person A looks toward frame-left (never looking into lens) and delivers one line: "[Exact spoken sentence]".
EXPRESSION: Subtle, restrained determination. Lips finish speaking at 4 seconds and remain closed.
AUDIO: Natural human voice, crisp room ambience. No soundtrack, no background music.
ACTION: Person A's right hand enters from frame-right, index finger and thumb grip the brass fountain pen, and lift it 3 cm vertically.
ANATOMY: Exactly 5 fingers visible. Clean knuckles, firm grip around the pen shaft.
NEGATIVE: No morphing fingers, no floating pen, no camera motion.
ACTION: Person A walks forward in a straight line through the wet street. Boots plant firmly on asphalt with physical weight transfer.
ENVIRONMENT: Steady light rain, asphalt reflecting street lamps, volumetric haze.
NEGATIVE: No moonwalking, no sliding feet, no camera wobble.
Complete 25-Chapter Handbook Index
Browse all 25 chapters from the official handbook. Click on chapters linked to our specialized cluster deep-dive guides: