AI Video line · stop 14 of 14 · 20 min · members
Composing for a tall frame from the start, so the result does not look like a wide shot with the sides missing.
Free with an account
Membership is free: an account opens all 86 script pages. The Lab, Studio Canvas and the paid guides need the $99 pass, paid once. Already signed in on this browser? The page opens by itself.
The tell
Viewers cannot name why, and they can see it.
A wide frame cropped to a tall one has the subject in the middle, empty space above and below, and nothing happening at the top or bottom of the frame. It reads as a fragment of something else, because it is.
Native vertical is composed differently: the subject is closer, the frame is layered from top to bottom, and the height is used rather than tolerated.
The difference is not resolution or quality. It is what the composition assumes about the shape it lives in, and that assumption is made at generation time.
Rule one
A tall frame has less horizontal room, so the subject must fill more of it.
The same subject at the same distance occupies a much smaller proportion of a tall frame than a wide one. Left alone, this produces the characteristic empty vertical.
Move in. What is a medium shot in landscape is a wide shot in portrait, and the framing that reads as close in a feed would look uncomfortably tight in a wide frame.
State the distance in the prompt rather than hoping. 'Subject at one metre, filling the upper two thirds' produces a tall composition; the same prompt without it produces a cropped-looking one.
Rule two
Use the height for something rather than leaving it empty.
A tall frame supports foreground, subject and background stacked vertically in a way a wide frame does not. Something in the lower part of the frame, the subject in the middle, something above — the shape becomes an advantage.
This is why vertical works well for interiors, streets and anything with depth: the format naturally shows the ground and the ceiling, and a composition that uses both feels made for the frame.
The alternative — subject centred with even emptiness above and below — is what a crop produces, which is exactly the impression to avoid.
Rule three
The platform will cover them, and it does not ask.
Feed interfaces occupy the top and bottom of the frame with captions, usernames, buttons and progress indicators. Anything you place there is either hidden or fighting for attention with a button.
Compose so the essential content sits in the middle portion, with the outer bands carrying atmosphere rather than information.
Check by overlaying a mock interface on your frames before delivering. It takes seconds and it catches the subject whose face is behind a caption.
Movement
A pan reveals very little in a narrow frame; a tilt reveals a lot.
Horizontal camera movement in a tall frame moves through a narrow window and shows little for the distance travelled. Vertical movement — a tilt, a rise, a fall — works with the shape.
Subject movement follows the same logic. Someone walking towards the camera reads strongly; someone crossing laterally exits almost immediately.
When adapting a shot idea from landscape, rotate the movement as well as the frame. Keeping the horizontal move is one of the most common reasons an adapted shot feels wrong.
Generating for it
The model composes for the shape it is given.
Set the output dimensions to the tall ratio at generation, never crop afterwards. The model will compose for the frame it is asked for, and that composition is the thing you are trying to get.
Then state the vertical structure explicitly: what is in the lower third, where the subject sits, what fills the upper area. Without it you will get a centred subject and empty bands.
Two extra clauses in the prompt, and the frame stops looking like it used to be something else.
1:1 · two hours