How to Write Better AI Video Prompts

Most weak AI video prompts fail before the model does. They ask for a mood and a resolution, and they never describe a shot. The fix is not a longer paragraph. It is a smaller scene.

Start with the shot, not the story

A prompt is a shot list of one. If you describe a character waking up, leaving home, and boarding a train, you have written a film. Most generators will smear those events together. Decide the single moment a viewer would see if they looked up.

An order that keeps prompts short

Use this sequence and stop when the shot is clear. The same order is illustrated on the text-to-video page.

1. Subject: who or what is in frame
2. Action: the one thing that changes
3. Place: only the details the camera would notice
4. Camera: locked, push-in, lateral, aerial, or handheld
5. Light and lens: time of day, contrast, film or clean digital
6. Exclusions: no titles, no logos, no extra people

Rewrite a vague prompt

Weak: “A cool cinematic video of a woman in a city at night, emotional, 4K, masterpiece.” That asks for a compliment and a resolution. It does not ask for a shot.

Stronger: “A woman in a black wool coat waits under a station clock as a train arrives. She does not board. Slow push-in, sodium light, shallow depth of field, fine rain, no readable signage.”

The second prompt can still fail. It can at least fail in a direction you can correct. More finished examples are in the prompt library on the homepage.

Camera words that earn their place

“Cinematic” is a compliment, not an instruction. These are instructions:

• Locked-off: the frame does not move. Useful for products.
• Slow push-in: a gradual move closer. Useful for portraits and tension.
• Lateral track: the camera moves with the subject. Useful for walking.
• Aerial, high and slow: landscapes. Ask for little else in the same prompt.
• Handheld, gentle: fashion and documentary energy. “Gentle” matters, or the frame shakes itself apart.

Change one thing, then judge

If the light is right and the action is wrong, do not rewrite the light. Keep a note of the prompt that almost worked. Ten unrelated ideas feel productive and teach less than four controlled variations. When you compare tools, paste the same prompt. The model guide explains what else to hold constant.

Leave room for the picture

Image-to-video prompts should not restate the whole photograph. Say what moves. “The coat hem lifts in a light wind. The camera drifts slightly left. The face stays calm.” If you already have a still, the image-to-video page is the better brief. If you have no picture yet, begin with text to video.