TWO-SPEAKER PLANNING
Hypit two-speaker scenes: keep dialogue, voices and shots in sync
Plan a two-person scene with a speaker map, turn-by-turn script, consistent setting and review checklist, inspired by a new community conversation example.
Edited by HypitAI · Reviewed
Use the post as a scene idea
On September 20, @cantonese_slang described trying a two-person conversation with an assistant in a set inspired by 1980s Hong Kong television. The post supplies a scene direction, not a complete configuration or recipe. The asset sheet and example dialogue below are our original planning material, not the creator’s prompt or a tested reproduction.
Give each speaker a stable identity
Create a speaker sheet before generating a complete conversation. Keep the same ID in the script, asset list and review notes.
| Field | HOST | GUEST |
|---|---|---|
| Role | Asks one clear question | Explains one practical answer |
| Visual reference | host-front.png | guest-front.png |
| Voice reference | host-voice.wav or approved voice ID | guest-voice.wav or approved voice ID |
| Position | Left, facing toward the guest | Right, facing toward the host |
| Continuity | Fixed outfit, scale and lighting | Fixed outfit, scale and lighting |
Write turns before choosing voices
Official Script Role Cues identify who speaks; they do not create a character or select a voice. Ask the agent to map each label to the chosen performance and voice setup separately. If the service cannot produce both speakers reliably, request separate shots or another supported approach instead of assuming labels solve it.
A four-turn desk-lighting example
These are original example lines. They form a shot plan, not a promise about duration or model output. Approve the voices before assigning final timing.
| Turn | Speaker and line | Picture | Review |
|---|---|---|---|
| 01 | HOST: Why is half my face in shadow? | Two-shot, then host close-up | Only the host speaks |
| 02 | GUEST: Try moving the light toward the front. | Guest close-up, then lamp insert | The visible movement matches the line |
| 03 | HOST: Does this position look more even? | Host in the adjusted light | Same face, outfit and background |
| 04 | GUEST: Compare both sides before recording. | Two-shot and a short checklist | Hold the final image through the last word |
An editable two-speaker brief
Use your own people, voices and scene assets. This brief requests a small preview before the full conversation.
/hypit Plan a two-person desk-lighting conversation using script.csv and speakers.csv.
HOST stays left; GUEST stays right. Use the approved visual and voice references in assets/.
Keep the background, outfits, eye lines and lamp placement consistent.
First return the turn-to-speaker, voice and shot mapping plus missing assets.
Preview one HOST-to-GUEST handoff before producing the full sequence.
No overlapping speech; the listener must not mouth the other speaker’s line.
Keep captions readable without covering faces. Recheck timing whenever speech changes.
Return the editable project, final MP4 and a turn-by-turn review record.Check the handoff, not just each face
Listen without watching to catch swapped voices and repeated lines, then mute playback to inspect positions, mouth movement and reaction shots. Finally compare every turn against the approved script. A clear individual portrait does not ensure continuity across a cut.
If you change the language
Have a speaker of the target language approve the wording and pronunciations. Keep a separate spoken-text and subtitle-text column if they differ. Recheck voice support, speech duration and subtitle wrapping in the chosen setup; a creator’s multilingual context does not establish support for every voice or dialect.