Vertical AI Video for Reels and TikTok: Prompting for 9:16

The first thing that surprises people about vertical AI video is how much of the frame the subject takes up. Nine by sixteen is tall and narrow, so a person standing three metres away in a landscape shot becomes a small figure in a letterbox that no longer exists. When you re-render the same prompt at 9:16, the model either crops in automatically and decapitates your subject, or it keeps the whole scene and leaves two thirds of the frame as empty floor and ceiling. Neither is what you wanted, and both are fixable in the prompt rather than in an editor.

The second surprise is camera language. Sideways movement is the horizontal format's best trick: a long pan across a landscape or a lateral tracking shot past a shopfront reads beautifully in sixteen by nine. In nine by sixteen the same move spends the clip pushing the subject out of frame. Vertical rewards movement along the axis instead, which means forward, backward, up and down. A slow push in reads as intimacy. A pull back reads as reveal. A tilt up from a pair of shoes to a face is a complete shot. A pan is usually a mistake.

Why the aspect ratio rewrites the prompt

There are three changes worth making on almost every vertical prompt, and they are all about where the subject sits. Describe the subject as filling the frame from roughly the waist up, or head to toe if the crop allows it, so the model does not shrink the person to fit a wide scene it invented. Move the camera instruction from lateral to axial. And say explicitly that the background should stay simple, because a tall frame is mostly background above and below whatever is in the middle.

Build the hook in the first three seconds

Vertical feeds are ruthless and the first three seconds carry almost all the weight. Which means the prompt should not describe a slow build. Start the clip already in motion, with the subject already where you want them: hands already working, the product already in frame, the face already turned toward the lens. The common mistake is writing an establishing beat, a wide shot, a movement toward the camera, and then the thing you actually wanted to show. By the time that arrives, the thumb has moved. Write the third second first, then ask whether the first second earns it.

End in a loop, then batch a week

A clip that loops cleanly gets watched twice, and a second view is a real gain on a vertical feed. Looping is a prompt problem: ask for a motion that returns to where it started, or for an action that completes and settles. A candle that flickers and holds, a pour that finishes, a leaf that sways and stops. Actions that end mid-gesture look broken when the video restarts, so keep the last frame close to the first and the seam disappears. Batching is what makes vertical video sustainable rather than a weekend project. Pick one product or one idea, write five variations of the same shot with different motion and different environments, and run them in one sitting. Reuse the same still across all five so the look stays consistent and movement is the only variable, then cut the week from what survived. Batch in the morning, review in the afternoon, and resist the urge to re-roll a single clip overnight, because a batch of five tells you far more than one perfect take. A handful of renders at 60 to 600 credits each is still cheaper than one shoot day, and the clip you discard costs a fraction of what a bad shoot day costs.