How to Make an AI Video Prompt (8-Element Formula + Examples)

Deepak Joshi
Written byDeepak Joshi
Abhinav Girdhar
Reviewed byAbhinav Girdhar
Read time14 min read
Last updated onNovember 9, 2026
How to Make an AI Video Prompt (8-Element Formula + Examples)

An AI video prompt is the text instruction that tells a video model what to generate, the subject, what it does, the camera, the lighting, the look, and even the audio. Get the prompt right and the model behaves like a crew that follows direction; get it vague and it guesses. This guide gives you a repeatable 8-element formula, copy-paste examples, model-by-model tips, and fixes for when output goes wrong. You can try every prompt here in a free AI Video Generator.

Try the AI Video Generator Free

AI video is moving fast, by 2025 a majority of marketers were using AI somewhere in video creation (Wyzowl), and short-form vertical (9:16) now rivals landscape (16:9) for AI output. Whatever you are making, the prompt is the lever that decides whether a clip looks generic or directed. Think of yourself as the director: the model is a very fast crew that does exactly what you say.

A cinematic AI-generated video still of an astronaut on an alien desert at golden hour, created from a detailed prompt
A detailed prompt produced this cinematic shot, subject, setting, lighting, and camera all specified.
◇

The 8-element AI video prompt formula

node

Almost every strong video prompt is built from the same parts, wired together in order. Chain these eight nodes and you hand the model a complete shot; unplug any of them and it starts guessing.

Subject: who or whatNODE 01

The main focus of the shot. Be specific about who or what it is, including key details of appearance.

out › a young woman with curly red hair and a yellow raincoat
Action & motion: what happensNODE 02

One clear thing the subject does, described as physical movement, not emotion.

out › slowly turns to look over her shoulder
Setting: where and whenNODE 03

The location, time of day, and atmosphere around the subject.

out › in a rain-soaked Tokyo alley at night, neon signs glowing
Camera: shot and movementNODE 04

Direct the camera: shot size, angle, lens, and movement such as pan, dolly, tracking, or aerial.

out › slow dolly-in, low angle, 35mm lens, shallow depth of field
Lighting & moodNODE 05

The light sets the emotion. Name the source, quality, and feeling.

out › moody neon lighting, soft reflections, melancholic and cinematic
Style: the lookNODE 06

The medium or aesthetic, such as photorealistic, film, anime, or a reference look.

out › shot on film, photorealistic, cinematic color grade
Audio: sound & dialogueNODE 07

For models that generate sound, describe ambience, SFX, or dialogue (or ask for none).

out › soft rain ambience, distant traffic, no music
Constraints: what to avoidNODE 08

Limits and exclusions that keep the shot clean and on-brief.

out › single subject in frame, locked-off horizon, no on-screen text

At a glance, here is what each element controls and an example phrase for it.

ElementWhat it controlsExample phrase
SubjectWho or what is on screen"a vintage red convertible"
Action & motionThe single main movement"speeds along a coastal highway"
SettingPlace, time, atmosphere"at golden hour, waves below"
CameraShot, angle, lens, movement"side tracking shot, low angle, 35mm"
Lighting & moodLight source, quality, feeling"warm sunlight, lens flare, cinematic"
StyleMedium and aesthetic"photorealistic film look"
AudioAmbience, SFX, dialogue"engine roar, wind, no music"
ConstraintsLimits and exclusions"smooth stable camera, no text"

Suggested Read: 30+ Prompts to Create Amazing Videos using AI

◇

The universal rules of AI video prompting

node

These cross-model rules matter as much as the formula. Follow them and almost any model behaves better.

Seven rules that always apply
  1. Front-load the intent. The first 20 to 30 words carry the most weight, so lead with the subject and the main action before the fine detail.
  2. One action, one scene per clip. Clips run about 5 to 10 seconds; cramming several events into one makes the model blend them into mush.
  3. Describe physical motion, not emotion. "Shoulders slump, head drops" works; "feels sad" does not. Models render what they can see.
  4. Phrase negatives as positives. Ask for "an empty hallway", not "no people". When a tool has a dedicated negative-prompt field, use that instead, since some models read inline negatives as things to include.
  5. Avoid temporal words. Skip "then", "next", and "after"; each clip is independent, so build sequences as separate shots and stitch them.
  6. Keep readable text and logos out. Models garble on-screen text, so leave titles and logos to post-production.
  7. Iterate one variable at a time. When a result is close, change a single element, the lighting or the camera, and regenerate; expect 5 to 10 tries per shot.
◇

Camera, shot and lens vocabulary

node

Naming the camera like a filmmaker is the fastest way to make a clip feel professional. Pull from this vocabulary.

Camera moves, grouped
  • Static: locked-off, tripod, no movement.
  • Linear: push-in, pull-back, dolly, tracking, boom or crane up and down.
  • Rotational: pan, tilt, whip pan, orbit or arc around the subject.
  • Complex: dolly-zoom (vertigo), crane moves that combine directions.
  • Specialty: FPV or drone, SnorriCam, handheld for energy.
Shot types and lenses
  • Shot sizes: extreme close-up (ECU), close-up (CU), medium, wide, extreme wide shot.
  • Lenses: 35mm or 50mm for natural framing; anamorphic for cinematic flares, spherical for clean bokeh.
  • Focus: shallow depth of field to isolate the subject, deep focus to keep everything sharp.
◇

Lighting, look and pacing

node

The same subject feels completely different under different light and grade. Reach for named recipes.

Light, look and speed
  • Lighting: golden hour, soft window light, dramatic side or rim light, neon practical key, high-key or low-key.
  • Look and grade: film grain, teal-and-orange cinematic grade, pastel storybook palette, film-stock emulation.
  • Pacing and speed: slow motion, timelapse, hyperlapse, speed ramping, motion blur for energy.
◇

Weak prompt vs strong prompt

node

Here is the pipeline in action. The weak prompts leave almost everything unplugged; the strong ones wire in every node.

Input: weak prompt

a woman in a city

Input: strong prompt

A young woman with curly red hair and a yellow raincoat slowly turns to look over her shoulder in a rain-soaked Tokyo alley at night, neon signs glowing. Slow dolly-in, low angle, 35mm lens, shallow depth of field, moody neon lighting, photorealistic and cinematic, soft rain ambience, gentle smooth motion, 5-second clip.

Input: weak prompt

a car driving

Input: strong prompt

A vintage red convertible speeds along a winding coastal highway at golden hour, waves crashing on the rocks below. Side tracking shot, low angle, warm sunlight with lens flare, photorealistic cinematic film look, engine roar and wind, dynamic fast motion, 6-second clip.

◇

Templates you can copy

node

Keep this wiring handy and fill in the sockets. The first is the universal fill-in; the others are shortcuts for common jobs.

Universal template

[Subject with details] [one clear action] in [setting, time, atmosphere]. [Camera shot, angle, lens, movement], [lighting and mood], [visual style], , [motion, pacing, duration], [constraints].

Short-form social (9:16)

[Subject] [punchy action] in [eye-catching setting], vertical 9:16, dynamic handheld move, bright high-key lighting, vibrant and fast, upbeat ambience, 6-second clip.

Product / ecommerce

[Product] on [clean surface/background], slow orbit or push-in, studio softbox lighting, crisp photorealistic, shallow depth of field, subtle ambience, smooth 360 motion, 5-second clip, no text.

Cinematic

[Character] [subtle action] in [evocative location] at [time of day], slow dolly-in, anamorphic 35mm, teal-and-orange grade, film grain, moody score bed, deliberate slow motion, 8-second clip.

Also Read: SeeDance 2.0 Prompts Collection

◇

Text-to-video vs image-to-video prompts

node

There are two ways to prompt, and they ask for different things. In text-to-video you describe the whole scene from scratch; in image-to-video you upload a still and describe only the motion and camera to apply to it.

Text-to-video (describe everything)

A lone hiker stands on a cliff edge at sunrise, arms raised, wind in their jacket. Wide shot, slow push-in, warm backlight, cinematic, gentle motion, 6s.

Image-to-video (describe only the motion)

[uploaded photo] Animate: slow push-in, the hiker's jacket and hair ripple in the wind, drifting clouds behind, subtle parallax, keep the composition, 5s.

In our stack, image-to-video runs on Runway Gen-4.5; in the AI Video Generator, upload your still as the start frame (JPG, PNG, or WEBP), then prompt the motion.

◇

Build a longer video as a timed shot list

node

Because each clip is independent, you make a longer video by generating short shots separately and stitching them. Here is a 13-second product ad as a shot list.

ShotLengthPrompt (abbreviated)
12sECU of the product on a matte surface, slow push-in, soft window light
23sHands open the box, top-down shot, warm light, shallow focus
33sProduct rotates on a turntable, slow orbit move, studio lighting
43sLifestyle wide shot, person uses the product outdoors, golden hour
52sClean end frame on the product, static shot, soft gradient background

Generate each shot on its own, keep the lighting and style wording consistent across all five, then edit them together.

Related Read: The Complete Guide to Text-to-Video Generation

◇

Copy-paste prompt library by video type

node

Adapt one of these ready-made prompts, each already uses the 8-element formula.

Ready-to-run prompts
  • Cinematic: A lone figure in a long coat walks away down a misty road at dawn, slow tracking shot from behind, soft backlight, teal-and-orange grade, film grain, ambient wind, gentle motion, 8s.
  • Product / ecommerce: A sleek pair of headphones rotates on a reflective pedestal, slow orbit, studio softbox lighting, crisp photorealistic, shallow depth of field, soft ambience, smooth 360 motion, 5s, no text.
  • Social / viral (9:16): A chef flips a sizzling pan of vegetables in a bright kitchen, vertical 9:16, quick handheld push-in, high-key lighting, vibrant, sizzle sound, fast energetic motion, 6s.
  • Character performance: A detective leans forward and narrows their eyes under a single desk lamp, medium close-up, slow push-in, low-key noir lighting, film look, quiet room tone, subtle motion, 6s.
  • Transition: A hand swipes across the lens and reveals a new city skyline, whip-pan transition, motion blur, bright daylight, energetic, whoosh sound, fast motion, 3s.
  • Nature / timelapse: Clouds race over a mountain valley as light sweeps the slopes, wide locked-off shot, timelapse, golden to blue hour, cinematic, ambient wind, 6s.
  • Sports: A sprinter explodes out of the blocks on a track, low tracking shot, shallow focus, harsh daylight, dynamic, crowd ambience, fast motion with slight slow-mo at the end, 5s.
  • Horror / suspense: An empty hallway stretches into darkness as a door slowly creaks open, static wide shot, flickering practical light, desaturated, low rumble ambience, very slow motion, 6s.
  • UGC-style ad: A person smiles and holds a product up to a phone camera in a sunlit room, selfie framing, handheld, natural window light, authentic, casual room tone, gentle motion, 6s.
  • SaaS / app demo: A clean animated UI glides across a soft gradient background as cards slide into place, straight-on shot, smooth motion-graphic style, bright and minimal, subtle ui clicks, 5s.
  • Faceless / educational: A top-down view of hands sketching a diagram in a notebook, static overhead shot, soft daylight, clean and calm, pencil-on-paper sound, steady motion, 7s.
  • B-roll / establishing: An aerial drone shot sweeps over a coastal city at sunrise, slow forward push, warm golden light, haze on the horizon, cinematic and expansive, ambient wind, 6s.
A cinematic AI-generated video still of a neon cyberpunk street in the rain, created from a detailed prompt
A different mood from the same pipeline: swap the subject, setting, lighting, and style, and keep the wiring.
◇

Match your prompt to the model

node

Different video models reward slightly different phrasing. The pipeline stays the same; the emphasis shifts. Pixazo’s AI Video Generator runs Seedance 2.5, Veo 3.1 Fast, Gemini Omni, and Runway Gen-4.5; here is how the major models across the market differ, with each name linking to its page on Pixazo.

  • Narrative models, Veo and Sora: respond to rich, natural-language scene descriptions and audio cues like “ambient rain and distant traffic”.
  • Motion-first models, Kling and Hailuo: love explicit camera and movement terms and clear constraints, so lean into shot type, speed, and direction.
  • Style-driven, Runway: does well with tight, keyword-led prompts and strong aesthetic references, and it also handles image-to-video.
  • Subject models, Wan and Seedance: want a clear subject, one action, and a defined camera move; keep it clean and uncluttered.

Not sure which to use? Our AI video model comparison breaks down the strengths of each. Inside Pixazo’s AI Video Generator you can run the same prompt on Seedance 2.5, Veo 3.1 Fast, Gemini Omni, or Runway Gen-4.5 and compare the results.

How long should an AI video prompt be? Most models reward a focused 40 to 120 words, front-loaded.

Prompt lengthBest forNotes
Short (15-40 words)quick tests, simple shotsfast to iterate; may leave the model guessing on look
Medium (40-90 words)most shotsthe sweet spot: full formula without contradictions
Long (90-150 words)complex cinematic shotsfront-load the key parts; extra words dilute weight

Suggested Read: GPT Image 2.5 Prompt Guide: How to Write Prompts

◇

Suggested Read: MiniMax H3 Image Prompt Guide

Troubleshooting: fix common AI video problems

node

When output goes wrong, it is usually a prompt fix, not a model fault. Match the problem to the phrasing that fixes it.

ProblemFix with this phrasing
Distorted hands or faces"hands relaxed at sides, natural anatomy"; use fewer simultaneous actions
Drifting or shaky camera"locked-off tripod shot, smooth stable framing"
Generic, flat lookadd a lens, a lighting recipe, and a colour grade
Garbled on-screen textremove text from the prompt; add it in post
Flicker or morphingshorten the clip, one subject, simpler background
Too fast or too busyone action per clip; "slow, deliberate movement"
◇

Suggested Read: How to Dub a Video Into Another Language With AI

How do you write a prompt that gets the video you want?

node

Direct it like a shot. Start with a specific subject, give it one clear action, place it in a vivid setting, then add the camera move, the lighting, the style, the audio, and the constraints. Front-load the important words, keep the clip short, phrase negatives as positives, and when a result is close, change just one node and regenerate. The eight-element pipeline turns a vague idea into instructions a model can actually follow. Open a free AI Video Generator, paste a template above, and start swapping in your own details.

Create Your AI Video Free

Related Read: Google Veo 3.1 Prompts Collection

Frequently asked questions

What is an AI video prompt?

An AI video prompt is a text instruction that tells a video model what to generate, covering the subject, the action, the setting, the camera, the lighting, the style, the audio, and any constraints. The clearer and more specific the prompt, the closer the output matches your vision.

How do you write a good AI video prompt?

Use the 8-element formula: name the subject, one clear action, the setting, the camera shot and movement, the lighting and mood, the visual style, the audio, and any constraints. Front-load the most important words, keep it to one action per clip, and iterate one element at a time.

What is the best structure for an AI video prompt?

A reliable structure is: [subject] [one action] in [setting]. [camera shot and movement], [lighting and mood], [visual style], , [pacing and duration], [constraints]. Layering these eight parts in that order is what separates cinematic clips from generic ones.

How long should an AI video prompt be?

Aim for about 40 to 120 words, front-loading the subject and main action in the first 20 to 30 words. Short prompts leave the look to chance, while very long prompts dilute the weight of each phrase and can contradict themselves.

What are the key elements of a video prompt?

Subject, action and motion, setting, camera, lighting and mood, style, audio, and constraints. Most strong prompts include all eight; the camera and lighting elements do the most to make a clip feel professional.

What is the difference between text-to-video and image-to-video prompts?

A text-to-video prompt describes the entire scene from scratch. An image-to-video prompt starts from an uploaded still and describes only the motion and camera to apply to it, so you keep the exact look of your image and just bring it to life.

Do negative prompts work for AI video?

Yes, where the model supports them. Use a dedicated negative-prompt field to exclude things like “blurry, distorted, extra limbs, text, watermark”. If there is no such field, phrase what you want positively instead, because some models read inline negatives as instructions to include.

How do I fix blurry or distorted AI video?

Simplify to one main action, add specific camera and lighting direction, shorten the clip, and keep one subject with a simpler background. For bad hands or faces, add “natural anatomy, hands relaxed” and reduce simultaneous movement, then regenerate.

Can an AI video prompt add sound or dialogue?

With models that generate audio, yes. Describe the ambience, sound effects, or dialogue you want, for example “soft rain ambience, distant traffic, no music”, or ask for silence. Models without audio simply ignore those terms.

How do you make a longer or multi-scene AI video?

Build it as a shot list. Generate each short clip separately with consistent lighting and style wording, avoid temporal words like “then”, and edit the clips together, since each generation is independent and runs only a few seconds.

Deepak Joshi

Deepak Joshi

Author · Pixazo

Deepak writes about generative AI models, APIs, and the workflows teams use to ship them. Reviewed by Abhinav Girdhar.

Related articles

With Pixazo’s platform we deliver enterprise-class security and compliance to you and your customers through every interaction.

Create anything.

Copyright © 2026 Pixazo by Appy Pie LLP. All rights reserved.