Vertical vs widescreen in AI video: how to frame 9:16 and 16:9

A tall frame and a wide frame are not the same picture at two sizes. They are two different jobs. Wide space lets a subject sit off to one side and look across the frame. Tall space stacks things: a face over a hand over a table. If you write one prompt and hope both shapes come out right, you usually get one that works and one that has a subject cut at the neck.

This post is about choosing the shape before you render, composing for it, and the few cases where it is fine to make one clip and crop.

Pick the shape first, from where the clip will live

The shape belongs to the place, not to the idea. A phone feed, a story or a short-video app wants tall. A website banner, a presentation, a TV or a video player on a laptop wants wide. Square sits between and suits a grid post. Decide this before the first draft, because the shape changes the prompt, the camera and the cost of every retry.

Most video models in the catalog let you choose the shape in the settings; the common choices are tall, wide, square and a few in between, and some models have an auto setting that follows your first frame. Not every model offers every shape, so open the model page and read the list before you commit to a plan.

Composing for tall (9:16)

Tall frames have lots of height and almost no width, so work vertically.

• One subject, centred or on a vertical line. A single figure, a bottle, a tower, a waterfall. Two subjects side by side get squeezed.
• Use depth instead of width. Put something in the foreground, the subject in the middle and a far light behind. Layers fill a tall frame that sideways spread cannot.
• Keep the face and the key action in the middle band. Apps put buttons, captions and a name along the bottom and the right edge. Leave the lower fifth and the outer edges free of anything you need people to see.
• Favour moves that follow the height. A tilt up, a rise, a slow push-in. A pan across a tall frame shows very little new picture.

Composing for wide (16:9)

Wide frames reward space on both sides of the subject. They are the natural shape for scenery, for travel moves and for two things interacting.

• Place the subject off centre. A third in from one side, with empty room in the direction it faces or moves. A subject that looks into a blank wall feels trapped.
• Describe the whole stage. Say what is left, right and behind. A wide frame with no description of the sides gets filled with whatever the model likes.
• Let the camera travel sideways. A slow pan, a track along a street, a drift past a row of lamps. This is where wide beats tall. The words for each are in nine camera moves for AI video.
• Mind the middle. If you may cut a tall version later, keep the real action near the centre third, as the next section explains.

What to keep in the safe area

A safe area is the part of the frame you can count on being seen whatever the player does. For tall clips, the centre of the picture and the upper two thirds are the safest, because interface elements crowd the bottom and the sides. For wide clips the risk is a crop, not a button: a wide clip shown in a tall space loses the left and right, so the middle third is the part that survives.

• Put faces, products and the one thing you want people to read in the centre third, whatever the shape.
• Leave a clean strip at the edge where you plan to add words in an editor; the studio does not draw text.
• Do not put anything important in the corners. Corners are the first thing a crop, a rounded screen or an app overlay removes.
• If a prompt mentions something at the far edge, expect it to be partly cut or drifting. Move it inward.

Prompting each shape

Say the shape in words as well as in the setting. A short opening line such as "vertical frame" or "wide frame" costs nothing and keeps the description consistent with what the model will draw. Then write the layout in the direction the shape allows: top to bottom for tall, left to right for wide.

• Tall: name what is at the top, the middle and the bottom of the frame, and give one vertical move.
• Wide: name what is left, centre and right, and give one sideways or forward move.
• Both: one subject, one move, one light. The advice in lens and focus words works in either shape.

When to generate once and crop

Cropping is cheaper than rendering twice, and sometimes it is enough. It works when the subject is a single thing in the middle of a calm, simple background and nothing moves toward the edges. A slow push-in on a bottle in the centre of a wide frame survives a tall crop. A pan across a street does not, because the pan is the whole idea.

• Crop from wide to tall only if the subject stays in the centre third for the whole clip.
• Never stretch a tall clip into wide. A tall picture has no spare picture to put at the sides, so you end up with bars or a blurred copy of itself. Render wide instead.
• Crop loses sharpness. A tall crop out of a wide frame uses only part of the pixels, so it looks softer than a clip rendered tall. If sharpness matters, render the shape you need.
• Render twice when the clip has a sideways move, two subjects or any action near the edges.

A cheap way to test is to draft both shapes with a fast, low-cost setting first. Draft, lock, finish shows how to keep the best wording and spend on the final render only once. If you start from a picture, make the picture in the shape you want first, as the teaser ad post suggests, then send it to an image-to-video model in the studio.

What the studio cannot do

It cannot reframe a finished clip for you. There is no smart crop, no extend-the-sides tool and no automatic switch from wide to tall. If you want a different shape, you render again or crop in a video editor. And the shapes on offer depend on the model: some give only a few options, so the one you want may need a different model.