AI Video Prompts Guide: Describe Motion over Time

Direct answer: A useful AI video prompt describes how one shot changes over time. Name the subject, one main action, the setting, one camera instruction, the visual treatment, and what should remain restrained. For a starting image, focus on motion instead of repeating everything already visible.

Fact-checked September 16, 2026. This page owns motion-brief structure. Use the text-to-video guide for a prompt-only workflow and the image-to-video guide when an uploaded or saved first frame should guide the shot.

Build one coherent shot

Use this motion-brief structure:

Subject, main action over time, in setting, camera behavior, light or visual treatment, restraint.

Example:

A ceramic cup on a walnut desk, steam curling slowly upward as morning light grows warmer, gentle camera push-in, natural reflections and shallow depth of field, keep the cup centered and the background calm.

Prompt layer	Question	Useful example
Subject	What anchors the shot?	A ceramic cup on a walnut desk
Main action	What changes over time?	Steam curls slowly upward
Environment	What else moves?	Morning light grows warmer
Camera	How does the view move?	Gentle push-in
Treatment	What should it feel like?	Natural reflections, shallow depth of field
Restraint	What should remain stable?	Keep the cup centered and background calm

Text-to-video and image-to-video need different emphasis

Starting point	Spend prompt detail on	Avoid
Text only	Subject, composition, setting, action, camera, and look	Several scenes or conflicting camera moves
One starting image	Subject motion, environmental motion, camera, and restraint	Rewriting every visible detail or contradicting the source

Text-to-video leaves the opening composition to the model. Image-to-video uses one uploaded or saved image as guidance, but it does not guarantee exact preservation of faces, products, logos, text, or crops.

Prompt examples by motion type

Controlled product motion

Matte-black headphones on a stone pedestal. A narrow studio highlight moves slowly across the ear cups while the camera slides a few centimeters from left to right. Keep the product shape, logo position, and background visually stable. Restrained premium movement.

For a full use-case workflow, read how to turn a product photo into a video.

Human-scale action

A chef places one herb garnish on a plated dish, then withdraws their hand. Gentle handheld push-in, bright natural kitchen light, crisp appetizing detail, one continuous action.

Environmental motion

Rain falls across a quiet neon-lit side street at night while reflections ripple on wet pavement. Locked wide camera, one cyclist crosses the frame, restrained realistic motion.

Motion from an illustration

The illustrated character blinks once as hair and coat move subtly in the wind. Very slow camera push-in, background clouds drift gently, preserve the calm drawn style.

Diagnose the result before rewriting everything

Result	Likely prompt issue	Next change
Too much happens	Competing actions or scene changes	Keep one action and one location
Camera feels erratic	Multiple camera instructions	Choose one move or lock the camera
Clip feels static	Motion is implied, not named	Add a concrete verb and direction
Starting image drifts	Motion request is too aggressive or contradictory	Reduce subject and camera motion
Product text changes	Fine-detail fidelity is being assumed	Simplify motion and inspect every frame
Framing is wrong	Delivery ratio was not considered	Prepare the source for 16:9, 9:16, or 1:1

Change one variable per retry. Each new creative attempt has its own quote; you pay only for a saved video, while provider or delivery failures release the reservation. A diagnosable prompt is still a cost-control tool because an unwanted but completed result is a charged run.

Match the brief to exposed controls

magicdoor.ai offers Wan 3, Kling 3.0, and FLUX 3. Each accepts text plus one optional uploaded or saved starting image and 720p/1080p output. Text-only runs expose 16:9/9:16/1:1. For a starting image, Wan 3 and Kling 3.0 follow the source framing, while FLUX 3 keeps the framing selector available.

Model	Duration	Generated audio	Five-second entry point
Wan 3	5-30 seconds	No generated-audio claim	$0.25 at 720p
Kling 3.0	5-15 seconds	Optional	$0.84 silent at 720p
FLUX 3	5-20 seconds	Optional	$0.30 in 720p draft

These are capability and price boundaries, not quality rankings. Choose 720p or 1080p based on the stage and destination, using the resolution decision guide for the cost tradeoff.

Access and first run

Only active or trialing subscribers can generate. Usage draws from included credit or account balance, and the full quote appears before generation. Up to three jobs can reserve funds concurrently. Finished videos remain private until deleted.

Anonymous visitors to /videos are redirected to sign in. Prepare one motion brief, then start with a five-second run that is easy to evaluate.

Create a video

FAQ

What should an AI video prompt include?

Include one subject, one main action, the setting, camera behavior, visual treatment, and a useful restraint. For image-to-video, spend more of the prompt on what changes over time than on re-describing the visible image.

How long should an AI video prompt be?

Long enough to define one coherent shot, but not a script for several scenes. A compact motion brief with one action and one camera instruction is usually easier to diagnose than a long list of competing events.

How should I troubleshoot an AI video prompt?

Start with a five-second 720p run, review the clip, and change one variable per retry. Simplify competing actions, make motion verbs concrete, and separate subject, environment, and camera movement.

Sources

Accessed September 16, 2026.

magicdoor.ai pricing for subscription and included-credit details.
The current magicdoor.ai video model registry and quote function for exposed inputs, controls, limits, and customer prices.
Wan 3, Kling 3.0, and FLUX 3 official model pages as supporting capability context. Magicdoor code governs exposed controls and customer prices.