Introduction

Prompt engineering for AI video is the discipline of structuring text instructions to maximize model output quality across subject, action, camera, lighting, environment, and style dimensions. This guide covers the six part prompt structure that works across Veo, Runway, and Sora, the cinematic vocabulary you need, timing and pacing techniques, and twelve production ready templates.

The Six Part Prompt Structure

A production grade AI video prompt has six components in order. Subject describes who or what is in the frame. Action describes what they are doing. Camera describes the shot type and movement. Lighting describes the quality and direction of light. Environment describes the location and time. Style describes the overall look and lens character.

Why Structure Beats Length

Longer prompts do not produce better results. Structured prompts do. A 60 word prompt with clear section boundaries consistently outperforms a 200 word stream of consciousness prompt because the model can map each dimension to the relevant attention layer.

Camera Language That Actually Works

Use specific cinematic terms: wide establishing shot, medium two shot, over the shoulder, Dutch tilt, dolly in, crane up, whip pan, rack focus. Generic terms like zoom or pan produce generic output. Specific terms anchor the model to learned cinematographic patterns.

Lighting Vocabulary

Strong lighting prompts name the source and quality: golden hour backlight with rim light on subject, overcast soft key from camera left, neon practicals with blue shadow fill, single hard source from above creating deep contrast. Vague prompts like good lighting do nothing.

Timing and Pacing Control

For models that accept duration cues use phrases like slow sustained movement over four seconds, quick two beat action with hold on final frame, continuous camera drift from left to right across full duration. Avoid contradictory timing cues in a single prompt.

Negative Prompting and What to Omit

Avoid listing things you do not want. Most video models respond poorly to negative prompts and sometimes introduce the negated element. Instead describe the positive state you want. If you do not want blur, say tack sharp focus throughout.

Twelve Production Templates

Copy ready templates for the twelve most common shots: product hero, UGC testimonial, fashion editorial, cinematic establishing, dialogue two shot, action beat, transition element, lifestyle b roll, interview framing, macro detail, aerial wide, and brand logo reveal. Each template follows the six part structure with interchangeable variables.

FAQ

Do the same prompts work across Veo Runway and Sora?

The structure transfers but specific vocabulary tunings differ. Veo responds best to precise camera language, Runway to stylized visual descriptors, Sora to narrative framing.

How long should a video prompt be?

Sixty to one hundred twenty words is the sweet spot. Beyond that gains diminish and contradiction risk rises.

Should I use prompt weighting syntax?

Only for models that explicitly support it. Most video models in 2026 do not, and adding syntax can degrade output.