How many shots are there?

One continuous take, 15 seconds, 360 frames.

Why not one prompt for the whole film?

The film continues into a corridor after the elevator. This product stops at 15 seconds, inside the single elevator take.

Does the prompt ask for lip-sync?

It asks the performance to follow the reference's speech beats. The soundtrack is mixed afterwards, so the mouth shapes line up with the lines.

Direct answer

The whole video is one 15-second take: 0–3s both in frame, 3–7s the speaker turns and says the line, 7–11s the answer and the laugh, 11–15s the held look. One Wan call, 360 frames, 16:9, no cuts.

Make my video

The one-take prompt

One scene, one call, one continuous shot — and a prompt that tells the model to follow the reference beat for beat.

Published November 10, 2026

The beat sheet

Time	Frames	What happens
0.0–3.0s	0–72	Both in the elevator, doors closed, headphones on, folders held.
3.0–7.0s	72–168	The speaker turns toward the listener and says the line; the listener reacts.
7.0–11.0s	168–264	The answer, the smile, the laugh; the listener touches one headphone cup.
11.0–15.0s	264–360	The held look. No cut; the camera stays put.

Why one call instead of five

The source is one continuous take. Splitting it into five pieces was a competitor's 9:16 re-edit, and every seam is a chance for an identity to drift. One call keeps both faces in the same shot for the whole 15 seconds.

Wan caps reference plus output at 30 seconds: 15 seconds of reference plus 15 seconds of output is 30 billed seconds. That is $1.97 at 720p and $0.99 at 480p — the same order of magnitude as the five-shot build, with fewer seams.

What the prompt says

Reference video 1 is the whole 15-second take. Rebuild the framing, lens, light, grade and timing.
Keep the elevator, the warm wood, the brass rail, the mirror and the stainless doors.
Each uploaded image is one person. Name them by position, not by clothing.
Replace the original performer; drop the original face and hair.
One continuous shot: no cuts, no text, no logos, no watermark.
The performance follows the reference frame for frame — same glances, pauses and speech beats.

What it deliberately does not say

No new dialogue, no song lyrics, no brand and no second location. It does not ask the model to generate audio; the 15-second soundtrack is muxed in post. It also avoids asking for an exaggerated performance: the reference's own speech rhythm is the timing master.