top of page

The Master Guide / 09 Performance · Reactions and the eyes

The reaction clock

Every planned reaction has a type and a time. In a one-shot prompt, write both in ACTION / TIMELINE (block 8 of the prompt, the timed action); in a multi-shot prompt, in that character's PERFORMANCE line (block 9). Both forms: Video with Seedance → Templates A, B and D.

Every planned reaction has a type and a time. In a one-shot prompt, write both in ACTION / TIMELINE (block 8 of the prompt, the timed action); in a multi-shot prompt, in that character's PERFORMANCE line (block 9). Both forms: Video with Seedance → Templates A, B and D.

The four types [proven that reactions are timed; the four latencies best available, not yet validated]. The latency is the delay between the cause and the first visible change on the face.

Type

When the face changes

Use it for

Overlapping

Before the line ends

Dialogue; the default when someone listens

Half a beat late

About 0.25–0.5 s after the cause

Naturalistic scenes and UGC (ads styled as a creator's own phone video)

Instant

With the cause

A shock; a comic freeze

Held, then late

0.5–1 s of nothing, then the reaction

The comic take

Written in the timeline, a held-then-late reaction reads like this: "at 6.0 s the van door slams; she keeps kneading; at 6.8 s her hands stop flat on the dough, and stay there."

Comedy shapes: a double take (the glance, a beat, then the look back as it lands) runs about 3 s; a slow burn (the reaction spreading slowly over the face) runs at least 4 s. Each gets a 1–2 s hold after it for the editor. Generate the handle and choose the in-point in the edit, because a quarter-second difference may not come back exactly as written.

One reaction per beat. Option: use it when you write a planned reaction, before any take has shown a repeat [single strong source]. State each reaction once, with its end state ("…and it stays there").

After a take repeats a reaction (decided): if it reacts, glances away and reacts again, write the reaction once with its end state, then the next movement.

Two readings of the key reaction (decided). The sample batch is the first batch you generate: one sample shot per scene, made as two 480p drafts (see Planning, budgets and takes → The generation rhythm). For one key reaction per film, those two drafts may be two different readings. Note in your take log which one wins.

Check the clock on takes you made anyway. A latency of a quarter to half a second needs every frame:

  1. Step through the window (the part of the take that plays in the cut) frame by frame in a timeline set to the take's own frame rate, 24 fps, or export every frame of the window as images (steps below).

  2. Note the frame of the cause and the frame of the first visible change on the face. Divide the difference by the frame rate: at 24 fps, 6 frames is 0.25 s and 12 frames is 0.5 s.

  3. Set the gap beside the type you wrote.

Frames every 0.5 s are enough to spot a double take or a reaction that comes twice, never to time a latency. If the face is too small to read, record it as inconclusive: it has not passed.

Do it yourself

Step through every frame of the window (Premiere, DaVinci Resolve or ffmpeg)

  1. In Premiere, New Sequence From Clip sets the timeline to the take's frame rate. In DaVinci Resolve, set the project's timeline frame rate to 24 before you add the clip.

  2. In Premiere and DaVinci Resolve the arrow keys move one frame.

  3. Or export every frame of the window as images. For a window from 5.5 s to 7.5 s, this line gives 48 images, the first at 5.5 s and each 1/24 s after the one before:

ffmpeg -ss 5.5 -to 7.5 -i take.mp4 r_%04d.png

Export frames every 0.5 s (ffmpeg)

  1. Export with the filter -vf fps=2. The full line: Camera and lens → Measuring the dose on the sample batch → Do it yourself.

Sources

bottom of page