Two characters speaking different lines in one wide shot require separate performance ownership and a shared dialogue timeline. If a performance tool handles only one character at a time, prepare synchronized individual passes and composite them in an external editor. A prompt naming two speakers does not overcome a documented single-character input limit.
Build the conversation before recording performances
Write each spoken line with its speaker, start, end and listener response. Include the pauses when neither person speaks. Keep one reference audio timeline for rehearsal, but retain each person's clean voice separately for the final mix. Overlap should be deliberate; accidental duplicate voices cannot be repaired by adding speaker labels to captions.
Runway's Act-Two multi-character tutorial describes separate performances, individual character generations and a later composite. That is an external workflow example, not evidence that Imagild provides its performance controls, compositing tools or multiple voice tracks.
Give each pass the complete shared duration
The first person's pass must include listening while the second person talks, not just their own isolated sentence. The other pass needs the complementary silence and reaction. Keep the common start time and frame rate. If a tool generates a shorter usable pass, do not stretch it blindly to match the other character; review its timing against the approved conversation first.
Use a common base composition with separable character regions. A fixed wide shot is easier to evaluate than a camera move with one person crossing in front of the other. Preserve the unmodified base and identify which pixels each generated layer is allowed to replace.
A hypothetical ten-second exchange
Imagine Mira asks a question from seconds one to three. Jon listens, pauses, and replies from seconds four to seven. Both remain visible until second ten. Mira's full-length performance includes her question and a silent listening response; Jon's includes the early listening and his answer. Neither pass should mime the other person's line.
On the shared timeline, align the approved voice starts with the relevant mouth movement. Inspect the silent character during every spoken phrase. Then mute the audio and ask whether the visual speaker order is still readable. This is a proposed acceptance exercise, not a demonstrated generation result.
Review the composite as a conversation
Check character overlap, edges, contact shadows and the common background separately from the performance. A correct mouth can still belong to a layer that slides across the floor. Play each voice alone, then the final mix, checking for rehearsal audio or duplicated speech left inside a pass.
Accept only when each line belongs to its intended speaker, listener mouths remain plausibly silent, reactions fit the actual pauses and the full-duration composite is stable. If crossings or simultaneous speech cannot be made credible, use approved separate views or a simpler staging plan. Archive the independent performances, dialogue timeline and final assembled version so that later line revisions remain reviewable.
Sources
Prepared by Imagild Editorial with AI assistance and reviewed against the linked sources on October 13, 2026. Examples are hypothetical; external procedures do not establish Imagild features or guarantee results.