Do not write a story synopsis; write what happens in a shot

01Start with seven slots

A prompt becomes executable when it specifies subject and scene, one action, camera, light and style, time and sound. You need not fill every slot, but know that an empty slot is a choice handed to the model. For image-to-video, the reference already supplies subject, composition and style, so text should focus on motion and camera.

02Use five practical camera words

Wide shot establishes person and place; medium shot shows action and body language; close-up emphasizes expression, hands or product detail; push in or pull back approaches a point or reveals a setting; pan or orbit shows spatial relation but can distort a complex subject. Begin with one clear movement, not three combined movements.

03Give one shot one main action

Prompt failure is often too many changes at once: entering a café, sitting, opening a laptop, looking out and smiling while camera circles behind. Split it into entering, placing the laptop, then looking toward the window. Each can be generated and replaced separately. Every additional action creates another chance for drift, occlusion or identity change.

04Use a reusable shot template

Template: [shot size and angle], subject in [scene] [one action]. [camera movement]. [light and style]. [elements that remain unchanged]. [ambient sound or dialogue]. Example: medium shot, low angle, a woman in a dark grey coat slowly looks at a neon sign after rain; camera gently pushes forward; wet road reflects brick-red light; clothes, hair and buildings stay consistent; rain and distant vehicles only, no dialogue.

05Storyboard distributes information before generation

For fifteen seconds write three rows: shot duration, visual task and sound task. 0–5 establishes setting and ambience; 5–10 completes the main action and marks it with sound; 10–15 lands on result or brand frame. Even a text storyboard is easier to revise than one huge prompt.

Next: choose a video model by the shot.

Continue to video models