How to Write AI Video Prompts: The Ultimate Guide for Creators (2026)

AI video tools have transformed what's possible for solo creators. But the gap between a mediocre output and a stunning clip comes down almost entirely to how you write the prompt. This guide covers universal principles that work across every major tool, plus tool-specific tips for VEO 3, Kling, Runway, Pika, and Midjourney.

The 7 Elements of a Great AI Video Prompt

Every excellent AI video prompt includes some combination of these seven elements:

Subject

Who or what is in the shot. Be specific — not 'a person' but 'a young man in his 30s with curly hair'.

Action

What the subject is doing. Use active, specific verbs: walks briskly, pours carefully, gazes into distance.

Setting

Where the scene takes place. Include environment, time of day, and any notable details.

Lighting

How the scene is lit. This is the most underrated element — lighting defines the mood.

Camera

Where the camera is and how it moves. Static, tracking, dolly push, aerial, close-up, wide, etc.

Style

The visual aesthetic: cinematic, anime, cartoon, minimal, realistic, etc.

Technical

Duration, aspect ratio (9:16 for Shorts), and any tool-specific parameters.

Universal Prompt Principles

Be specific, not generic

AI models interpret vague prompts by averaging their training data. "A nice sunset" produces a mediocre sunset. "A fiery orange and crimson sunset over the Pacific Ocean, sun partially below the horizon, long shadows stretching across wet sand" produces something cinematic.

❌ Weak

"A person running in the city"

✓ Strong

"A woman in athletic gear sprints through empty Manhattan streets at dawn, camera tracks low beside her, morning mist, golden light between skyscrapers"

One subject per scene

Every AI video tool struggles when you give it too many subjects doing different things. Focus each prompt on one primary subject. If you need multiple subjects, write it as a wide establishing shot or keep them in the background.

Lead with the most important element

Most AI models weight the beginning of prompts more heavily. Put your subject and primary action first, then add context and technical details at the end.

Tool-Specific Prompt Formats

VEO 3

VEO 3 uses pure natural language. Write as if describing the scene to a cinematographer. It responds best to specific cinematic language.

"[Subject doing action] in [setting with atmosphere], [lighting description], [camera movement], [duration in seconds], 9:16 vertical format"

Key tip: Always end with "9:16 vertical format" — VEO 3 defaults to landscape.

Kling 2.0

Kling uses a semi-structured format with square bracket notation and parameters.

[Subject] [Action] [Setting and atmosphere] --duration [seconds] --ar 9:16 --style [realistic/anime/cartoon]

Key tip: The --style flag dramatically changes output. Test both "realistic" and your target style.

Runway Gen-4

Runway is motion-obsessed. Lead with camera movement and reference Runway's native camera terms.

[Camera movement description] of [subject and action], [detailed setting], [lighting and atmosphere], [mood]

Key terms: dolly in/out, pan left/right, tilt up/down, orbit, handheld, smooth tracking.

Pika 2.0

Pika responds well to comma-separated descriptive tags combined with parameters.

[subject], [action], [setting], [lighting style], [camera angle], [mood] -ar 9:16 -s [style]

Midjourney (stills for video)

Midjourney generates stills, not video. Use it for reference frames or thumbnails in your video workflow.

[Scene description], [lighting style], [camera angle] --ar 9:16 --style raw --v 6.1

The 30-Second Prompt Test

Before generating, read your prompt out loud and ask: "If I described this to a filmmaker on set, would they know exactly what to shoot?"

If any of these questions are unanswered, your prompt needs work:

Who/what is in the shot?
What are they doing?
What does the environment look like?
What is the light source and quality?
Where is the camera and how does it move?
How long is the clip?
What is the aspect ratio?

Stop Writing Prompts Manually

For a 60-second Short with 10 scenes, you're writing 10 individual prompts — each tailored to the specific AI tool you're using. That's 30-45 minutes of work before you've generated a single frame.

ScriptFlow AI generates complete scene-by-scene scripts with tool-specific prompts pre-formatted for whichever AI tool you select. Enter your concept once, select your tool, and get all 10 prompts in 10 seconds.

Generate Tool-Specific Prompts Automatically

VEO 3, Kling, Runway, Pika — choose your tool and get formatted prompts for every scene

Try ScriptFlow AI Free →