Seedance 2.0 Text-to-Video Prompt Guide: Subject, Motion, Camera, and Audio

An effective Seedance 2.0 text-to-video prompt does more than request something "cinematic, epic, and 8K." It lets the model answer four concrete questions: who or what moves, how it moves, how the camera observes it, and when each sound occurs.

ByteDance Seed's official page says Seedance 2.0 uses a joint audio-video architecture and accepts text, image, audio, and video inputs. Its technical report describes generations from 4 to 15 seconds. The duration, aspect ratios, resolution, and reference limits available to you still depend on the platform and model variant you select.

Start with one event that fits the duration

A common mistake is putting an entire story into one prompt:

A woman leaves home, crosses the city, meets her friends,
attends a concert, and watches the sunrise.

That is a sequence of scenes, not one shot. Reduce it to a single event:

On a rooftop before dawn, a woman walks into the wind,
stops near the edge, and looks up as the first sunlight reaches the city.

A short generation works best with a starting state, one continuous action chain, and a recognizable final frame. For multiple locations, plan a storyboard and generate separate shots or use a carefully limited timeline.

Seedance text-to-video prompt formula

[duration and aspect ratio] + [subject and starting state]
+ [continuous action] + [one primary camera move]
+ [environmental change] + [lighting and style]
+ [sound or dialogue] + [continuity constraints]
+ [final frame]

Example:

8-second cinematic shot, 16:9.
A lone woman in a dark wool coat stands on a rocky sea cliff at blue hour.
Wind moves only her coat and hair. She takes two careful steps forward and stops.
The camera begins in a tight profile close-up, then slowly moves over her shoulder
and tilts down to reveal waves striking the rocks below.
Cold blue ambient light, realistic ocean spray, stable horizon.
Audio: strong coastal wind and distant surf, no music, no dialogue.
End on a high overhead wide shot with the full figure visible.
Maintain the same face, coat, cliff, and weather throughout.

Write subject motion and camera motion separately

The Seedance 2.0 Pro Gallery example Woman on a Cliff Overlooking the Ocean begins close on the woman's face, eases over her shoulder, tilts toward the cliff, and finishes overhead above the waves. Its strength is a camera path with a start, a route, and an endpoint rather than a vague request for "dramatic camera movement."

The prompt also demonstrates a language problem. Wind tearing at her coat may mean wind pulling violently on the fabric, but tearing can also imply that the coat is physically ripped. Replace ambiguous prose with observable action:

strong wind pushes and flaps the coat fabric; the coat remains intact

A generation prompt is not a literary contest. If a word could create two visibly different interpretations, rewrite it as a concrete action.

Bind every sound to a time and source

Do not end with only "cinematic sound." Specify where the sound comes from and when it happens:

0-3s: soft footsteps on gravel, wind from camera left
3-6s: footsteps stop, ocean surf grows louder
6-8s: one strong wave impact, no dialogue, no background music

The sound source should exist in the frame. Do not add dialogue when the character never opens her mouth. If the environment is far away, avoid making it sound as if it touches the microphone. Product, food, and mechanical clips can use specific sounds such as tapping, friction, a cap opening, or carbonation.

When should you use a timeline?

A 4–8 second single shot usually needs only a clear beginning, process, and endpoint. Use a timeline when the prompt includes an outfit change, beat-synchronized actions, a transition, or multiple sound events:

[00:00-00:03] Establish the subject and environment
[00:03-00:06] Complete the main action with one camera move
[00:06-00:08] Stop the action and hold a clear final composition

For a multi-shot ad, use the 15-second breakdown in the Seedance 2.0 product video prompt guide. To anchor a person or product with a reference image, pair it with the Seedance 2.0 image-to-video prompt guide.

Fix failures in this order

1. The action does not finish: reduce the number of actions and remove irrelevant background events.
2. The person or object drifts: repeat only the critical identity, clothing, color, and quantity constraints.
3. The camera moves unpredictably: keep one primary move and define its start and endpoint.
4. Audio is out of sync: state the sound's time, source, and whether music is allowed.
5. The clip has atmosphere but no event: add a starting state and final frame instead of more style adjectives.

Change one category of variable per test and keep successful versions. This makes it possible to tell whether the problem comes from the event, camera, audio, or continuity.

Seedance 2.0 text-to-video FAQ

Can I prompt Seedance in Chinese?
Yes, but camera terms, directions, and movements still need to be unambiguous. If one phrase behaves inconsistently, rewrite that sentence as simple visual English rather than stacking Chinese and English synonyms throughout the prompt.

Why does a detailed timeline still skip events?
A timeline organizes instructions; it does not make too many events easy. If every segment changes the location, clothing, character, and camera position, the model still has to perform multiple hard transitions. Reduce the changes before adding more timestamps.

Should I choose parameters or write the prompt first?
Decide the use case, aspect ratio, and available duration first. Then write an event that fits. Resolution changes the output specification but cannot repair conflicting movement instructions.

Generate a Seedance 2.0 text-to-video prompt

Write one event that fits the target duration, then organize its action, camera, sound, and final frame: Open Genvizu AI Video Prompt Generator