People & dialogue · Hard · 5 s

Pavo AI Video: Two-Shot Dialogue Hold

Two people in one frame is a continuity test, not a conversation test. A Pavo AI two-shot dialogue preset that prioritises a stable eyeline and a held ending over fake lip sync. Includes two identity fixes and edit notes.

Load this preset in the studio →
Bright interior portrait with a single subject and soft window light
Camera
Locked, slight three-quarter
Length
5 s
Difficulty
Hard
Best for
Short-drama beats, reaction shots, scripted dialogue inserts
Shot family
People & dialogue
Reviewed
18 November 2026

Before you render: the first-frame check

A generated clip inherits every ambiguity in its opening still. Approve the still as a usable composition first, then ask for only the motion this shot actually needs. If one of these checks fails, fix the still — not the prompt.

  • Both faces are fully in frame with no overlap of hair or shoulders.
  • One eyeline direction is defined before any motion is requested.
  • Clothing colours are distinct and flat, with no patterns or type.
  • The background has no mirrors, screens or framed text.

Three beats, in order

One action with a beginning, a middle and a held ending gives the model priorities and gives an editor recognisable cut points. Secondary movement stays subordinate to the main action.

Beat 1 · 0–1.5s

Both subjects hold a neutral listening pose.

Beat 2 · 1.5–4s

Only the left subject inclines the head slightly; the right subject stays still.

Beat 3 · 4–5s

Both settle into the original pose and hold.

Anchors that must not drift

List the identity and geometry anchors that must stay fixed. Review them across the whole result, especially during turns and camera moves. If one detail fails, revise that instruction alone.

  • Two face shapes
  • Two shirt colours
  • Eyeline direction
  • Shoulder positions
  • Background geometry

The six-line prompt

Copy all six lines into your own generator. Each line owns one decision, so a failed render tells you which line to change. Do not add extra adjectives that do not change a visible production decision.

SUBJECT: Two adults in a bright interior, one in a plain cobalt-blue shirt on the left and one in a plain cream shirt on the right.
FRAME: Locked medium two-shot, slight three-quarter angle, both faces fully visible, plain off-white wall, one window light from the left.
ACTION: Hold a neutral listening pose; only the left person inclines their head slightly; both return to the opening pose. No mouth movement, no speaking.
CAMERA: Locked off, no pan, no push, no cut, no reframing.
FIXED: Both facial structures, both hairlines, two shirt colours, eyeline direction, shoulder positions, no text, no mirrors.
END ON: Both subjects in the original pose, fully still, for 12 frames.

Want to generate your own version? Use the six-line prompt builder.

If it breaks: two signals, two single-line fixes

SignalChange this one line
The two faces start to resemble each other.Split into two single-subject clips and cut between them. Identity bleed between two generated people is the most common failure in this format.
Mouths move without audio.Remove every dialogue, speaking and lip-sync word. Add voice and captions in the editor instead.

See all six failure → fix entries →

Editing handoff

Judge this as footage, not as a demo. These are the notes that make the clip usable downstream.

  • Record dialogue separately and place it over the holds, not the movement.
  • Cut to a single-subject reaction shot for any line longer than three seconds.
  • Never generate captions inside the frame; keep them editable.

Read the full export handoff guide →

Keep going

Related shot presets

All 12 shots

A preset library, not a lucky prompt.

Twelve presets across six shot families, every one with a held end frame and an editing handoff. Load any of them into the studio.