Prompts and multilingual dialogue

A useful prompt explains what happens before describing how to film it. Separate subjects, action, environment, style, camera and sound so the instructions remain easy to inspect and revise.

Core structure

Subjects and references → action/event → location → style → camera → sound

Subjects: identify everyone present and which references define them.

Action: describe visible behavior and cause-and-effect order. “She looks at the envelope, then raises her eyes” is more actionable than “she is very shocked.”

Location: establish place, time, objects, light and spatial relationships.

Camera: specify framing, position and one principal movement.

Sound: assign dialogue, language and voice, and state whether ambience, effects, music or subtitles are wanted.

Single-shot example

@Image1 defines Lin's identity and clothing; @Image2 defines the bookshop at night.
Lin stands on the left of the counter. She puts down the envelope with her right
hand, pauses, then looks toward screen right.
Medium close-up, fixed camera, a warm desk lamp illuminating her face.
Using the voice in @Audio1, she says quietly: “You finally came.”
Keep subtle paper and clothing sounds. No background music or subtitles.

Insert real studio references using @; example labels are placeholders.

Plan action with time segments

For longer Seedance 2.5 clips, describe segments such as 0–5 seconds establishing the relationship, 5–10 seconds for dialogue and 10–15 seconds for a listener reaction. These timestamps guide creation; inspect the actual output timing.

Allow natural speaking time before adding pauses, turns or movement. Avoid packing many actions into one second, and do not request an uninterrupted take together with several hard cuts.

Sound notation

The guide suggests the following conventions to distinguish audio tasks:

Type	Example
Music	(soft strings)
Effect	<a quiet click from the door lock>
Dialogue	{Lin: You finally came.}
Subtitle	【Three years earlier】

These are prompting conventions, not mandatory programming syntax. For clean finishing footage, explicitly exclude subtitles, text and background music, then inspect the result.

Languages and accents

Specify language, regional accent, speaker, delivery and the exact line. For English dialogue, provide the actual English sentence instead of only asking for translation.

Lin speaks in English with a natural American accent.
Her voice is quiet and controlled, with a short pause before the final word.
Dialogue: “I kept the letter. I just couldn't send it.”
Use @Audio1 for her voice identity. No narration or subtitles.

Audition the voice in the target language. Distinguish speakers and turn order, and check that the reference recording's original words have not entered the new dialogue. Supported languages are listed under Models and parameters.

Iterate deliberately

Identify whether the problem concerns identity, space, movement, camera or sound. Preserve successful conditions and change one major variable. Restore historical prompts to compare versions. After an AI-assisted rewrite, check that it has not introduced unwanted story events or shots.