Skip to main content

Break-Up on the Roof

One step back at the roof edge: a break-up told by hands going into pockets.

Story & dialogueImage to videoNeeds one reference image311 credits

Fills the prompt and every setting below into the Seedance 2.5 form. Nothing is submitted until you press generate.

The prompt

Keep both people and their clothing exactly as in the reference image. They stand at the roof edge a metre apart, not touching; the taller one takes a single step back and pushes both hands into pockets, and the other turns their head away toward the city. The camera drifts slowly right along the parapet, keeping both in frame. Blue hour, warm window lights coming on below, wind moving hair and jacket hems. Distant traffic and a loose flagline flapping.

Settings this prompt was written for

Settings this prompt was written for
FieldValue
ModeImage to video
Duration8 seconds
Resolution720p
Aspect ratioadaptive
AudioAudio on
Credits311

They are the real price of that exact configuration — computed with the same pricing engine the API charges with, at the resolution, duration and audio setting shown on the card. Change any of those in the generator and the price updates with it.

Why it works

The whole scene is distance, so the prompt measures it: a metre apart, not touching, one step back, hands in pockets. Those are the cues a model can hold for eight seconds without inventing facial animation it cannot do well. Locking the two-character still first also means the drift along the parapet never risks re-casting or merging the pair, which is the usual failure in multi-person image-to-video.

What to put in the reference image

Upload a still containing both characters at once — a two-shot with a clear gap between them and both faces visible works best. Avoid images where one person is cropped, heavily backlit, or much larger than the other, because image-to-video keeps that imbalance for the whole clip.

breakup scene promptrooftop blue hour videoimage to video two charactersdialogue scene ai video prompt

Prompts stay in English on every language of this site: that is the language the model was trained on, and translating a prompt changes what it renders. The page around it is translated.