How to Write AI Video Prompts Without Rewriting Everything

Originally published at openvideos.ai.

Start from the input you already have

The useful prompt formula changes with what you bring in.

• Text-to-video: the prompt must define the whole shot: subject, action, place, framing, camera, light, and timing.
• Image-to-video: the image already carries subject, composition, light, and style. The prompt should mainly describe what moves.

Repeating a long text-to-video description on top of a finished still is a common failure mode. The model gets several competing versions of the same scene.

Formula 1: text-to-video as a seven-part shot brief

Subject + Action + Setting + Shot + Camera movement + Light/style + Timing

You do not need a long sentence for every part. You need enough information to remove the decisions that matter.

Part	Question	Example
Subject	What must stay recognizable?	A lone cyclist in a mustard-yellow rain jacket
Action	What changes in this shot?	Rides steadily across the bridge
Setting	Where and when?	Blue hour after rain, wet road, light mist
Shot	What belongs in the frame?	Wide road-level three-quarter shot
Camera	How does the viewpoint change?	Tracks parallel at a steady distance
Light/style	What is physically visible?	Cool ambience, warm road reflections
Timing	How should motion unfold?	One controlled, continuous shot

Keep short clips focused on one scene. If the idea contains three plot beats, split them into separate shots.

A complete example

A lone cyclist in a mustard-yellow rain jacket rides steadily across an empty suspension bridge at blue hour after rain. Wide road-level three-quarter shot, with the cyclist on the left third and open road ahead. The camera tracks parallel at a steady distance. Wet asphalt reflects distant red and amber traffic lights; cool overcast ambience, light mist, restrained contemporary film color. One continuous shot with controlled speed and smooth tracking throughout.

The prompt is detailed because it resolves seven decisions, not because more words automatically produce a better video.

Formula 2: image-to-video should describe motion

Camera motion + Subject motion + Environmental motion + Timing

Treat the source image as the visual foundation. For the cyclist still, a motion prompt can stay short:

The camera tracks smoothly beside the cyclist. The cyclist pedals at a steady pace while the rain jacket moves lightly in the wind. Fine mist drifts across the bridge and road reflections shimmer subtly. Controlled motion, one continuous shot.

Skip a second biography of face, jacket color, bridge architecture, or lighting. The image already supplies those facts.

Diagnose before you rewrite

After a weak result, name the smallest failed decision first.

What you observe	Likely problem	First revision
Subject looks generic	Missing visible anchors	Add two or three defining details
Too much happens	Competing events	Keep one primary action
Framing feels accidental	No shot size or viewpoint	Add one explicit shot instruction
Motion is confused	Camera and subject fight	Keep one camera move and one subject action
Image barely moves	Prompt re-describes the still	Remove scene description; state motion
Mood is vague	Abstract adjectives	Specify light, color, weather, or texture

Change one component at a time. Keep the rest stable so the next generation answers a real question.

A credit-aware loop

1. Choose the input mode before polishing language.
2. Write the minimum complete prompt for one shot.
3. Inspect the selected model's actual controls.
4. Confirm displayed credits before submit.
5. Save the exact prompt and result.
6. Name the largest mismatch.
7. Change one component.

Sources and limits

This framework synthesizes first-party guidance from Google, Runway, and Luma video-prompt docs. It does not rank models or claim identical results across them.

Full guide with checklist and links: How to Write AI Video Prompts.