Seedance 2.5
Prompting Guide

The guiding principle

Write a production brief, not a mood board

Seedance 2.5 works best when you give it a clear production brief rather than a collection of vague style words. Describe the subject, action, environment, visual treatment, camera, and audio, then define the role of every reference image, video, or audio file.

The core prompt formula

Use this structure for a straightforward generation:

Subject performs primary action or event in scene and environment.
The visuals feature visual style.
Use shot size, camera angle, camera movement, or cuts.
Audio includes dialogue, ambience, sound effects, or music.

Generation parameters such as resolution, duration, and aspect ratio are set on the generation page or through the API. They are usually not part of the prompt itself.

Build the prompt foundations

Use explicit audio syntax

When dialogue, music, sound effects, or subtitles need to be distinguished, use Seedance's visual syntax:

For dialogue, state the language and voice style before the line:

Map every reference explicitly

Never assume Seedance will understand what each uploaded asset is for. State what every image, video, or audio file contributes.

@Image 1 defines the woman's appearance, hairstyle, and clothing.
@Image 2 defines the laboratory layout and lighting.
@Video 1 defines the camera movement and pacing.
@Audio 1 defines the narrator's voice.

Add exclusions whenever a reference could bleed in something unwanted:

@Image 2 defines the workbench and window light. Do not use the people shown in the image.

Images are best used for distinct subjects, products, props, or locations. Videos generally define motion and pacing unless you explicitly say they define identity too.

Multi-reference prompts

When you have several characters, props, locations, or reference files, map them separately and group the materials by type.

[Characters]
The conservator corresponds to @Image 1. Use only the appearance, hairstyle, and clothing.
The registrar corresponds to @Image 2. Use only the appearance, hairstyle, and clothing.
Do not interchange their appearances, clothing, actions, positions, or dialogue.

[Props]
The sample case corresponds to @Image 3. It belongs only to the conservator.

[Scenes]
The conservation lab references @Image 4. Use only the room layout, architecture, and lighting.

[Motion and Audio]
@Video 1 defines the motion of the conservator opening the sample case.
@Audio 1 defines the guide's voice and dialogue.

For a recurring subject, create a central profile that fixes appearance, clothing, prop ownership, locations, and motion references. Then select only the references needed for each scene and describe the event and its end state.

Long videos and timing control

For videos longer than roughly 15 seconds, or videos with several distinct events, use stages instead of one long paragraph. Each stage should contain one primary event and a clear visible end state.

Use time ranges to allocate pacing, exact time points for key events, and relative timing for delays between actions. Time ranges are not frame-accurate edit points.

Editing an existing video

For video editing, define the source video as the sole editing master. State the edit target, scope, target material, and what must be preserved.

[Edit Goal]
Edit @Video 1. Within the entire video or a specific time range, replace the background outside the subject's silhouette with a nighttime city rooftop.

[Source Video Role]
@Video 1 is the sole editing master. It defines the subject, identity, clothing, action, composition, camera movement, audio, cuts, and event order.

[Edit Scope]
Modify only the background outside the subject's silhouette.

[Content to Preserve]
Keep the subject's identity, facial features, clothing, position, motion, camera movement, timing, dialogue, and sound effects unchanged.

For subject replacement, state that the new object inherits the original object's timing, motion, occlusion, path, speed changes, and exit. For background replacement, explicitly protect the subject's identity, facial features, clothing, expression, position, size, and motion.

Audio categories can be edited independently when you name the category and what must stay untouched:

Edit @Video 1. Remove only the original background music. Keep the character dialogue, lip sync, ambience, and action sound effects. Preserve the visuals, camera treatment, and editing rhythm from @Video 1.

Extending a video

For a forward extension, align the boundary frame before describing the new action:

For a backward extension, describe what happens before the source starts and make the source's first frame the final state of the new segment. Check the boundary frame, motion trend, and audio continuity in both directions.

Advanced workflows

One-click video from images

Always specify material roles, image order, motion amount, editing style, visual treatment, and audio. If image order matters, state it explicitly. If it does not, allow the model to arrange images by theme.

Seamless transitions

Define the before video, after video, trigger action, camera movement, visual transformation, arrival state, and audio transition. A transition aims for visual and audio continuity, not pixel-identical preservation of both clips.

Direct emotional performance visibly

Abstract words such as “tense,” “warm,” or “oppressive” leave too much to interpretation. Pair emotion with eye movement, brow tension, mouth movement, breathing, gaze direction, or hand movement.

Instead of “the character becomes nervous,” write:

After hearing the footsteps, the character stops breathing for a moment, shifts their eyes toward the doorway, tightens their brow, grips the edge of the table, and begins speaking in a whisper.

Final checklist

• The subject and primary action are clear.
• The setting, time, lighting, and spatial relationships are defined.
• The camera movement supports the action.
• Every reference has a clearly stated role and unwanted parts are excluded.
• Characters, clothing, props, and identities stay consistent.
• Long videos are divided into stages with visible end states.
• Editing prompts identify the sole source video and exact edit scope.
• Audio is distinguished as dialogue, music, ambience, or sound effects.
• Timestamps are used as pacing guidance, not frame-accurate guarantees.
• The prompt does not promise perfect subtitles, exact product text, or pixel-identical transitions.
• Generation settings such as resolution, duration, and aspect ratio are configured separately unless the workflow locks them automatically.