Prompting Veo for dialogue scenes
Veo's defining trick is native audio: the dialogue, the room, the effects all render with the picture. Most prompts ignore this and describe only what the shot looks like. The fix is simple, and it changes everything about how the take lands.
Write the sound
Name what you want to hear with the same care you give the image. The line of dialogue in quotes, the rain against the window, cups on saucers, a bus hissing past outside. Sound you don't describe is sound the model invents, and invented sound is where takes go wrong.
Two old friends argue over the last pastry in a rain-soaked café.
She says "you had the last one in 1987" and slides the plate away.
Handheld two-shot. Rain on the window, cups on saucers,
a bus hissing past outside. Natural sound, no music.Direct, don't decorate
One clear camera instruction beats three mood adjectives. Veo reads lens language closely, so spend your words there rather than on words like cinematic, which every prompt already implies.
| SPEND WORDS ON | SKIP |
|---|---|
| Handheld two-shot, slow push-in | Cinematic, stunning, epic |
| The line of dialogue, in quotes | A conversation happens |
| Rain on glass, cups on saucers | Cozy atmosphere |
| Rack focus from hands to face | Beautiful lighting |
One beat per take
Eight seconds is one beat of one scene: the argument, not the whole reconciliation arc. Write the prompt around a single exchange with a clear emotional turn. A reliable checklist:
- +One location, one action, one emotional turn
- +Dialogue line in quotes, eight words or fewer
- +Two or three named sounds, then stop
- +One camera instruction, written like a shot list
Sound you don't describe is sound the model invents.
