An off-screen speaker still needs a stable identity. Being outside the frame does not transfer the line to the nearest visible face. State who produces the voice, where that person belongs in the scene and whether the visible character is listening, speaking or simply being shown while a voiceover plays.
Separate three questions
Ask who is speaking, whether that source is visible now and whether the voice belongs to the scene or to narration. A person calling from the next room, a narrator describing a memory and a close-up person speaking directly are three distinct audiovisual relationships.
The H3 full-reference guide retains speaker identity when a referenced subject speaks off screen. The base guide also documents voiceover handling and closed lips for the corresponding on-screen character. These are provider rewrite conventions, not evidence that any website automatically converts a vague “voiceover” request correctly.
Write a visibility ledger for the shot
In a fictional scene, a child looks at a kite while a parent gives a short instruction from outside the frame. The child should respond by adjusting the string, not by speaking the parent's words. The parent may enter in the following shot without acquiring a different voice identity.
| Source | Visible now? | Intended behavior |
|---|---|---|
| Parent | No | Says the approved instruction |
| Child | Yes | Listens and adjusts the string |
| Narrator | Only if separately requested | Does not replace the parent's dialogue |
For an inner monologue over a visible adult, specify that relationship instead of describing it as ordinary physical speech. Do not add a new narrator character merely to fill a numbering gap. Likewise, do not show the off-screen parent inside the current frame just to provide a mouth for every audible word.
Check the listener, not only the speaker
Review the visible person's mouth through the entire off-screen line. Random lip motion can make a correctly voiced soundtrack appear assigned to the wrong character. Check gaze and reaction too: a listener can remain silent yet respond before the instruction, creating a different continuity error.
Listen for voice identity before and after the source becomes visible. If the same parent enters, compare the role assignment rather than demanding identical acoustic perspective; a change in location can reasonably change the room sound. Record that intended distinction in the shot notes.
Use ordinary role and visibility descriptions in an Imagild video brief, with actual site attachment labels for references. Keep H3 internal speaker formatting separate and retain the complete applicable rewrite structure when using it. Acceptance means the approved voice has the correct owner, only intended mouths articulate, and nobody is added to the frame without a story reason. This article describes a review plan; it does not claim an actual H3 generation passed it.