A 9:16 crop of HD is 608 pixels wide

Not 1080 — 608, because the window is the target shape at the source's height and everything else follows from that. It is placed by clamping: the subject's centre minus half the strip, then pushed back inside the frame. A subject near the edge gets a window shoved off centre, still containing them and no longer centred on them. Put two people 1,510 pixels apart and no single x holds them both — which is arithmetic, not a preference, and it is the reason to pan rather than a matter of taste.

The engine is free and needs no account. Give it the frame, the target shape and a box round each subject, and every window is placed in source pixels: where it wants to be, where the clamp allows, and the gap between them. It also states the thing nobody computes — a 608-pixel strip stretched into a 1080-wide delivery is a 1.78× upscale, real softness that nothing downstream recovers — and it merges the speaker turns too short to cut to, so the pan count you see is the one that will actually run. Every lane ships a worked example that costs nothing. Writing the report needs credits.