How to Write Better AI Video Prompts: Examples for Beginners

Start with one shot. Describe the subject, one visible action, where it happens, and what the camera does. Then add a detail about the light or mood if it helps explain what the viewer should see. Try the text-only examples in Text to Video. The product and reference-clip examples below use their own input modes.

Write for the camera

“A cinematic bakery” describes a mood but leaves the shot open. “A baker slides a tray of croissants onto a counter. Hold a waist-high close shot as warm window light catches the steam” specifies an action, framing and light. If you have a short clip, give the action enough time to read. For a sequence, plan a separate clip for each shot and cut them together in an editor.

Three prompts to adapt

Vertical social opening. “A cyclist rolls a bike past a yellow café window. The camera holds a steady medium shot. Morning light reflects off the glass.” Choose 9:16 in the studio for a vertical post. If you need a caption, keep some open space in the composition and add the exact text in an editor.

Product shot from a photo. “The camera slides slowly to the right while the blue bottle stays upright on the table. Keep the bottle centered and the background plain.” Use Image to Video with a clear photo when you have a real product. Check the shape, logo and label throughout the clip. The photo-to-video guide explains how to prepare the input.

Camera movement from an existing clip. “Follow the reference’s slow forward camera movement. Set the new shot in an empty greenhouse with soft light through the glass.” Upload a compatible MP4 in Video to Video. If your reference includes both subject action and camera movement, say which one matters most. The new clip can differ in framing and timing.

Change one thing when you revise

1. The subject does too much. Remove extra actions. Try “the baker places the tray on the counter” before adding a turn, a smile or a second person.
2. The camera moves too much. Replace “dynamic camera” with one direction and speed, such as “slow push-in,” or ask for a stationary camera.
3. The output is hard to crop. Choose the intended aspect ratio before generating and tell the model where to keep the subject. Leave space around hands or feet if the action needs it.

Check the model, duration, resolution and credit quote before each new generation. If you want a preset scene with the prompt already set up, browse the video templates.