Dzine or APOB? Keep the Listener Silent

A visitor asks where the spare key is. The host answers, then the visitor takes a beat to think. In a generated two-person shot, a word attached to the wrong mouth changes who knows the answer. Even a small movement from the silent listener can look like a reply. Set the line ownership before choosing how to make the shot.
Prepare a speaking shot for your dialogue exchange
This Dzine and APOB comparison starts with a six-line key exchange and a map of the pauses. The examples show how to plan and inspect the scene. They do not report a lip-sync benchmark.
Give every spoken line one visible owner
Use two fictional adults and one neutral task. A is the visitor. B is the host. Keep the speaker IDs when you replace the dialogue with another language.
A question from the visitor
The visitor asks the odd-numbered questions. Line 1, A: “Did I leave the spare key here?” Line 3, A: “The blue dish by the door?” Line 5, A: “May I take it when I leave?” All three lines come from A. For the first pass, use unadorned readings so a reviewer can distinguish words, faces, and gaps.
An answer from the host
The host supplies the even-numbered answers. Line 2, B: “Yes, I put it in the dish.” Line 4, B: “Yes, beside the front door.” Line 6, B: “Of course. It’s still there.” The host knows where the key was put. A visitor whose mouth moves on line 2 could seem to answer their own question. Listen to the audio alone and check that the order still makes sense.
Read the lines with the same words and sequence in either workflow. The only planned difference is how the two faces are covered. Keep the host's reference shot and the visitor’s reference shot labeled A or B, with the six recorded line files carrying the same IDs. If an audio editor renames a file, update the line map before the next person previews it.
A pause owned by neither voice
Add Pause P1 after line 2 and Pause P2 after line 5. Neither pause owns a line of audio. P1 gives A a moment to recall which dish the host means. P2 lets B glance toward the door before answering. In a shared shot, write the intended action in the pause row instead of assigning a silent clip to whichever face was speaking last.
Inspect Dzine’s documented line-to-face assignment
A shared shot makes the face assignment part of the edit. The plan should still work when you change a voice file or discover that a reference hides one mouth.
A clear source with distinguishable faces
Dzine’s Multiple Lip Sync page describes starting with a visual source and assigning speech to characters. For this exchange, start with a source where the adults are readily distinguishable. Mark A by the entry and B by the counter in the shot note. A jacket, position, or name label outside the image makes it easy to compare the reference against the assignment.
Assign different lines to different characters
Map line 1 to A and line 2 to B, then repeat with the other four lines. If your source contains a camera move, check the first frame of each speaking turn where both faces are visible. Recheck any turn in which somebody walks behind the other person or looks away. The owner remains A or B even if positions within the frame change.
For a new take of B4, change that line alone and listen to the beginning of A5. The first word of a reply can be audible while the other face still occupies the frame. That can be a valid choice if the listener’s behavior reads as listening. Leave a note identifying the offscreen voice so another editor can tell the difference between a planned overlap and an accidental reassignment.
Preview ownership before exporting
Preview the changeover from each line to its following pause or answer. Put the scene on mute and see which face looks like it is speaking. Then restore sound and watch for a mouth movement that feels like a different person is supplying the words. This gives you two observations to save before export: visible owner and audible owner. It is an inspection procedure based on a documented workflow.
Pause on the first frame where the audience can see the answer taking shape. Confirm B’s identity from the reference and look for a smaller change in A’s face. In an actual review, note whether a camera angle or a cut prevents you from seeing the listener clearly enough to judge. Keep that row open until a usable angle is available.
Build separate speaking shots when coverage is the safer design
Separate coverage offers an editorial way to protect a line from being misread. You can hold the listener’s reaction in its own shot and reuse the approved audio order in assembly.
One approved speaker input per line
The APOB Lip Sync AI page describes a visual source plus audio workflow. Prepare one speaking input for each planned line with that person’s voice or an authorized recording. The instructions for this article use separate shots. They do not depend on both faces being assigned independent audio inside one image. Keep the speaker’s wardrobe, framing, and eyeline stable between lines so the six clips feel like one conversation.
For a small team, the advantage of the APOB route is that each speaking shot can have its own approved source and audio pair. A revision to B4 can be inspected in isolation before reinserting it. You still have to match the eye lines and the sound between cuts, and some scenes need both faces to share one continuous moment. Use the assignment workflow if a single shared frame matters more than independent shot revisions.
A listener insert without a speech track
During P1, show A’s reaction without a speech track for that insert. During P2, show B glancing toward the door. The reaction source may be a still or a separately prepared clip. Inspect it for mouth shapes that look like formed words. Allow the ambient sound already chosen for the scene to continue in the assembled sequence, if that suits the cut.
Make a reference for B’s listening turn too: “Adult host at the kitchen counter, eyes shift toward the front door, chin lifts slightly, no spoken line, face shown in a stable three-quarter view.” Compare the expression with the previous speaking shot, keeping the door on the same side. If a movement implies a new line, reduce it before approving the insert.
Assembly that preserves the original speaker order
Lay out A1, B2, P1, A3, B4, A5, P2, and B6 in order in your editor. If the end of a line overlaps a reaction insert, mark the line’s offscreen owner in the notes and confirm the listener does not appear to mouth it. Keep those decisions separate from line deliveries. The APOB voice direction guide can help when cadence or emphasis is the problem.
Distinguish silence from a failed speaking turn
Give the pause its own behavior and allowed sound. A glance, nod, or breath can advance a conversation before anyone says another word.
A listener who is thinking
P1 is a thinking pause. A looks down for a beat, then looks back to B. In the shot note, write “A listens, lips resting, eyes move briefly toward the dish, room tone only.” For an original reaction reference try: “Adult visitor near the entry, quiet attention, gaze briefly toward a small blue dish, mouth at rest, no other face in frame.”
A brief acknowledgement without a word
P2 is a brief acknowledgment. B can give a small nod before answering. A storyboarding note from Walt Disney Animation Studios discusses expression, timing, and staging as part of planning a scene. Here, a nod has a place in the cut, and the first word of B6 gets its own start. If a nod repeatedly reads as the beginning of a spoken answer, choose a quieter gesture or give it a separate shot.
An audio gap that must not become another line
An audio gap stays a gap in the ownership map. If a chair creaks or the key touches the dish, note the sound separately from speech. Use a simple grid while reviewing.
Beat | Line owner | Intended visible behavior | Allowed sound |
|---|---|---|---|
B2 | B | B speaks, A attends | B2 and room tone |
P1 | None | A thinks | Room tone |
A3 | A | A asks, B attends | A3 and room tone |
P2 | None | B nods | Room tone |
B6 | B | B answers | B6 and room tone |
Give the reviewer a second column for what they actually saw. Leave it blank until they watch. A movement can be a planned inhale, an unexpected opening of the mouth, or a legible word shape. Preserve that distinction in the record. You can ask the reviewer to name who spoke without announcing the intended answer. Write uncertainty down before you choose which shot to repair.
Repair ownership without rewriting the conversation
The repair starts from the recorded mismatch. Keep the approved words in place while you change either the assignment or the coverage.
A line assigned to the wrong face
Hypothetical error: the visible owner column says A on B2. Assign that line back to B if the shared shot supports a clean face selection. Mark B2 as revised and inspect the exact frame before and after the first syllable. Recheck P1 even though its notes did not change. A corrected speaking turn can expose a gap that was previously hidden by the mistaken movement.
A reaction mistaken for an answer
Hypothetical error: P1 looks like an answer. Keep its audio free of speech and review the listener’s lips in the reaction source. A shot of A’s eyes or hand hovering over the dish may communicate thinking with less ambiguity. Recheck the last word of B2 and the first word of A3 so the change preserves the sense of both speakers waiting for the next turn.
A silence that reads correctly on a large display can still be hard to interpret on a small one. Check the final crop after captions have been added. If the captions obscure the dish or the listener’s glance, move them within the edit before deciding to replace a shot. Record whether the remedy involved the mouth, the sound, or what the viewer could see.
A shared shot that needs separate coverage
When both faces repeatedly look like they speak the same line, try separate coverage and label the offscreen audio. Start with the two turns around the error, then play the whole exchange in order. Save the observed behavior against the corrected map and flag any ambiguity that remains. For a moving setup, APOB’s image-to-video guide can help plan a visual reference before you assemble the speaking shots.
Use the same three points when you inspect an export: the first word, the last visible syllable, and the transition into silence. If you swap shots, play them with the room tone you will use in the finished cut. If you only change assignment, check that a face which was correct on the adjacent line still owns it. Finish by playing all six lines without stopping.
Related reading: Morphic vs APOB: Rehearse a Chase Without Cloned Extras

















