Framing, aspect ratio and where the eye lands

Aspect ratio is passed to the model before it renders, so it shapes what gets generated rather than trimming it afterwards. The same prompt at 2:3 and at 16:9 produces genuinely different images: the portrait fills the height with the subject, the widescreen version invents environment on either side to fill the space.

• 2:3 — single figures, portraits, anything where height is the subject.
• 1:1 — objects, icons, tight character studies with no environment to justify.
• 3:2 — the default photographic frame; a subject with room to breathe.
• 16:9 and 21:9 — landscapes and establishing shots. Expect the model to invent scenery.
• 9:16 — vertical video and phone-first stills.

Without a stated shot distance most models default to a mid shot, roughly waist up, subject centred. If you want anything else you have to ask. Shot distance and camera height are two short clauses that do more for a composition than any amount of style vocabulary.

The camera control in the video composer works by adding a sentence to your prompt, which means you can write the same instruction yourself and be more precise than the preset list allows.

Describe one movement per clip. A prompt that asks for a slow push in, a pan left and a rack focus in five seconds will usually produce none of them cleanly.