How to write a text-to-video prompt for one clear shot

1. Give the shot a subject and a single action

Start with a concrete subject: an adult cyclist, a red paper boat, or a lighthouse. Then add an observable action. ‘A paper boat drifts past a stone’ is easier to assess than ‘a beautiful, inspiring journey.’ You can check whether the action happens and whether the boat remains recognizable.

A short clip has limited time. If you ask the subject to enter, speak, run away, and return, the model may compress or skip events. Begin with one movement and use separate clips for other story beats. You can assemble those shots later in an editor.

2. Choose a setting and one camera instruction

Describe the environment only to the level needed for the shot. Weather, time of day, and the surface under the subject often matter more than a long list of adjectives. Add one camera instruction, such as ‘fixed camera’ or ‘slow push-in.’ A stationary camera is a useful baseline when you want to judge the subject’s movement.

Camera movement and subject movement are different instructions. ‘The boat drifts left while the camera stays still’ gives both a direction. Asking for a tracking shot, a zoom, and an orbit at the same time makes it difficult to know which movement is the priority.

A small red paper boat drifts slowly past a smooth stone in a shallow stream. Soft overcast daylight, gentle ripples, fixed low camera, one continuous shot, no cuts, no text.

3. Match the prompt to the actual controls

Set duration, aspect ratio, and resolution in the generator controls rather than relying on words in the prompt. The selected model determines which options are available. Writing ‘4K’ does not create a 4K output when the chosen model or tier does not support it.

Audio also depends on the model. A prompt asking for music or dialogue cannot enable audio on a model that lacks that capability. Review the available controls and the credit quote before submitting. New accounts receive 5 trial credits, which are below current video task costs.

4. Review the result against a small checklist

Watch the clip once for the action, again for the camera, and once more for object continuity. Does the boat keep its shape? Does the stone stay in place? Do the ripples agree with the direction of movement? This separates a prompt problem from a visual artifact.

For the next attempt, change one instruction. If the camera moves unexpectedly, strengthen the stationary-camera instruction while leaving the scene unchanged. If there are too many objects, remove background detail. A new attempt is another generation task with its own cost; there is no guarantee that a longer prompt fixes the result.

5. Know when to start from an image instead

Text to video is useful when the scene can be invented. If a specific product, face, or composition matters, use a permitted source image in image to video. The image gives the model a starting arrangement, while your prompt focuses on motion.

A template can save typing, but it still needs review. Load a camera-motion preset, replace the subject and setting, then check the selected model and quote. Preview clips illustrate a direction; they do not promise that your result will reproduce the example.