Distinguish three different audio roles
drama-skills handles visible dialogue, off-screen speech and voiceover separately. “She says” conflicts with an instruction that she silently thinks with her lips closed.
| Item | Record |
|---|---|
| Visible dialogue | Visual requirement: The right person speaks and performs the complete line; Example: The mechanic faces the customer and says the keys follow the test ride |
| Off-screen speech | Visual requirement: Voice belongs to someone outside the frame or with mouth unseen; spatial direction remains coherent; Example: The camera stays on the mechanic while the customer asks from opposite them |
| Voiceover or inner thought | Visual requirement: The visible person need not mouth the line; Example: The mechanic looks at a key under recollection voiceover |
For a line continuing across shots, specify where it starts, where it continues off-screen and where it ends. Do not finish it in one prompt and restart it in the next.
Record each line’s actual language and verbatim wording rather than inferring them from instruction language. If the audio uses the wrong language or changes the line, first use the language and dialogue check, then inspect lip sync here.
Budget the complete line, preparation and response
Use the actual speaking duration when you have the same line recorded in the same performance. Without audio, label the duration a text estimate rather than claiming it was measured. Also allow for preparation, actions that cannot overlap speech and the final response.
This is a budgeting example: a seven-second shot with half a second before and after leaves six seconds for speech. If your full recording lasts eight seconds, writing “finish within seven seconds” does not make it fit. Extend the shot or split at a meaningful boundary. If shortening the line, revise the script first and update the prompt and audio together.
| May adjust | Preserve |
|---|---|
| Unnecessary gestures, walking and complex camera motion | Complete words and the correct speaker |
| Simple actions that can naturally overlap | Actions that must occur in sequence |
| Duration or a meaningful shot split | A readable response at the end |
Speech rate varies by language, voice and emotion. A fixed characters-per-second estimate is not a capacity guarantee for every model.
A two-person prompt with distinct speaking and listening roles
This is an original shot design that still requires output review.
Keep the accepted shop camera position: mechanic screen-left, customer screen-right. The customer stops reaching and looks at the mechanic without speaking. Facing the customer, the mechanic says the full line: “Keys after the test ride.” This is the mechanic’s visible dialogue. The customer listens, then withdraws their hand; they do not repeat or mouth the line. Keep the withdrawal after the mechanic finishes, with the keys still on the bench. Add no voiceover or new words. Set duration from the full recording and required response.
If a visible listener repeatedly mouths the words, a permitted storyboard revision can isolate the speaker and show the hand withdrawal separately. If keeping both people visible, specify the silent listener and check the result.
For an inner-voice version, do not merely relabel the audio while keeping the speaking action. Show hands and keys under voiceover, or require closed lips when a face must remain visible. The skill’s small-sample observations show that textual restrictions can still fail; writing “mouth closed” does not prove the output obeyed it.
Match the repair to the failure
| Item | Record |
|---|---|
| The wrong person mouths the line | Check first: Speaker ownership and visible cast; Repair: Clarify the source; revise framing if needed |
| A face speaks during voiceover | Check first: Voiceover mislabeled as dialogue; Repair: Revise the shot or specify closed lips, then inspect output |
| The last words disappear | Check first: Full speech duration versus available window; Repair: Extend, split or revise the script without silently dropping words |
| A line repeats in two shots | Check first: Start, continuation and ending across prompts; Repair: Align prompts and the edited audio |
Listen to the full line, then watch at normal speed for mouth/audio correspondence and extra or missing words at cuts. Prompt text, subtitles and a still image cannot establish synchronization.
FAQ
Does “perfect lip sync” guarantee success?
No. Specify the right voice source, full line and feasible duration, then inspect the output. Repair speaker ownership, voiceover staging or capacity as appropriate.
Can I change difficult dialogue into voiceover?
That changes presentation and may change the story. Confirm the scene can work that way, then revise script and storyboard rather than silently converting it inside a motion prompt.
Do it with the skill
Ask short-drama-video-prompts to preserve the script and storyboard while specifying speaker, full line and visible/off-screen/voiceover role. Check the speaking window and propose extension or splitting when needed, without changing dialogue.
Read the method: Dialogue, voiceover and visible mouths · Timing dialogue and actions
About Drama SkillsThe short-drama-video-prompts skill on GitHub