The whip pan and readable text

A whip pan only reads as intentional if you tell Claude two things: how long the swing takes, and which subject is which. Skip either and Seedance either drifts the timing or loses track of who's on screen when the camera lands.

Naming the whip pan

The ask here is a two-shot scene: a half-second wide to anchor who sits where, then a hard cut into a single continuous shot where the camera plants itself between the psychologist and Santiago and whip-pans from her to him mid-line. The prompt labels her Subject A and him Subject B, and gives the whip an exact budget — settled until 0.3s before the move, a 0.5s motion-blur swing, the new subject settled by 1.4s. Without those numbers, Seedance guesses at pacing and the swing either drags or snaps before the actor's face has resolved.

First whip-pan take — a nostalgic smile lands on him seconds after he was crying in scene 4. Timing and continuity both hold; the emotional read doesn't.

The first render nailed the mechanics but broke continuity with the previous scene — he'd just been crying, so a warm smile on the answer read as a non sequitur. Feeding Claude that context (he's still shaky, red-eyed, no smile) fixed the performance without touching the whip-pan spec at all — proof the technical prompt and the emotional prompt are separate knobs.

A second take introduced its own tell: the pan opened on a medium shot and landed on a close-up.

A medium-to-close-up whip pan — the frame size changes mid-swing, which reads as a cut hiding inside a pan. A whip pan must land on the same shot type it started on.

That's the rule a whip pan can't break: it's one continuous camera move, so it has to open and close on the same shot size — medium to medium, close-up to close-up. A whip pan that changes frame size mid-swing isn't a whip pan anymore, it's a disguised cut, and it reads as a glitch rather than a transition.

Catching slop in on-screen text

Readable text is where a generation quietly fails even when the acting and camera are perfect. The therapist's notebook came back as scribbles standing in for handwriting — a tell that's easy to miss on a quick watch but instantly reads as AI slop on a real look.

The fix is to give Claude the exact words to render, not a vague "she writes something": three specific lines of clinical shorthand, spelled out verbatim in the prompt, with the shot held on her hand long enough (a 2-second close-up) for the text to actually resolve on screen.

Final scene 5 — the notebook now reads GUILT — missed penalty / Self-blame, unresolved / Roots: childhood / football, clean and legible in the raking window light.

If a prop carries text, name that text word-for-word in the prompt and hold the shot long enough to read it — anything vaguer and the model fills the page with slop.

A whip pan is a physics problem before it's a style choice: fix the duration and the settle-in time first, then layer performance notes on top — don't try to fix pacing and emotion in the same prompt edit.