Reframing

Changing the canvas alone leaves subjects half out of shot. Reframe deliberately.

Workflow

Set the canvas: setProjectSettings with the target aspectRatio ("9:16", "1:1", "4:5", "16:9").
Frame people: one reframeAuto with follow:"faces" per clip covers the frame and centres every shot on its measured face. Do not keyframe shots one by one. Add zoom (at least 1) only when the brief wants a tighter frame.
Follow a moving subject: reframeAuto with keyframes (a per-frame track), or the subject-tracking workflow for objects.
Named layouts: applyLayout frames a clip as "full", "split-top" / "split-bottom", "framed" (rounded card) or "caption-safe" (raised, bottom strip clear for captions), with cover-fit and no bars. Pass mediaWidth/mediaHeight when known.
Check preview frames across each shot: nobody half out of shot, faces clear of captions and platform UI.

Never letterbox a landscape shot into 9:16: a 16:9 picture fills only 32% of the frame, wasting about 68%. A blurred plate is a last resort, not a style.
reframeAuto follow:"faces" needs measured face boxes (semanticProfile.visualContext). Without them, inspect or analyse the source first.
Faces in portrait sources were once squashed by face detection; trust preview frames over the numbers.
Reframe before captions and titles; overlays placed for the old canvas land in the wrong place.