Seedance 2.5 Prompt Guide: Start With a Clear Video Brief
This Seedance 2.5 prompt guide explains how to turn a rough video idea into instructions a model can follow more consistently. A well-structured prompt improves instruction following, reference-material consistency, and control over the result. It does not remove randomness or guarantee a fixed level of visual realism, complex camera execution, or frame-perfect editing.
Before writing, decide whether you need a text-only generation, reference-led creation, an edit of an existing clip, or a sequence assembled from several assets. The right prompt is not the longest prompt. It is the one that clearly tells the model what each supplied material means and what must happen on screen.
For reusable starting points, browse the Seedance 2.5 prompt library, then try it in the built-in Seedance 2.5 generator.
Use the Core Prompt Formula
The most reliable default structure is:
Subject + action or event + scene and environment + visual style + camera movement or cut + audio
Start with the subject and its main action. Then add only the details that affect the result: place, lighting, material, mood, shot behavior, dialogue, ambience, sound effects, or music.
<Subject> performs <primary action or event> in <scene and environment>.
The visuals feature <lighting, color, material, texture, or mood>.
Use <shot size, camera angle, camera movement, or cuts>.
Audio includes <dialogue, ambience, sound effects, or music>.
For example:
A ceramic artist finishes a pale blue cup in a studio at dawn, lifts it from the wheel, and places it on a wooden shelf.
Soft morning light enters through the window. The wet clay has a delicate sheen and the workbench remains tidy.
Begin with a medium shot of the wheel-throwing process, slowly push in toward the cup's surface texture, then cut to a frontal shelf view.
Retain the wheel's low hum, clay friction, and subtle indoor ambience.
Generation settings such as duration and aspect ratio belong in the product controls when they are available; do not treat them as substitutes for a clear scene brief.
Assign Every Reference a Role
References work best when the prompt states both what to inherit and what to ignore. Seedance 2.5 can combine up to 50 reference materials, including up to 30 images, 10 videos with a combined duration of up to 30 seconds, and 10 audio clips with a combined duration of up to 30 seconds. More inputs are not automatically better: use the smallest set that directly supports the scene.
@Image 1 defines <subject>'s appearance, clothing, structure, or material.
@Image 2 defines <scene>'s layout and lighting. Do not use people from this image.
@Video 1 defines motion, pacing, or camera movement. Do not use its identity or setting.
@Audio 1 defines the voice, dialogue, ambience, or music.
Then write the generation instruction normally. If several images show one object from different angles, say that they are views of the same object and state that only one instance must appear. Do not assume labels embedded inside an image will map people or props correctly.
Keep character and prop mappings explicit
For advanced multi-reference work, name each entity and bind it one by one. Group characters, props, scenes, motion, and audio, then write a short profile for any subject that appears across scenes. This prevents appearance swaps, incorrect prop ownership, and accidental background carryover.
[Characters]
<Conservator> corresponds to @Image 1. Use appearance, hairstyle, and clothing only.
<Registrar> corresponds to @Image 2. Do not interchange the two characters' clothing, actions, or dialogue.
[Props]
<Sample Case> corresponds to @Image 3 and belongs only to <Conservator>.
[Scenes]
<Conservation Lab> references @Image 4. Use only the space, materials, and lighting.
Select references by scene instead of asking every supplied asset to appear at once. This is especially important when you are preparing a longer sequence or a product workflow.
Build Longer Videos as Stages With End States
For a multi-event or 30-second video, split the story into consecutive stages. Give each stage one main change and an observable end state. End states preserve continuity: they tell the next stage who holds the prop, where a character stands, and what remains visible.
[Generation Goal]
Generate an instructional video about packing a flower-shop order.
[Stage 1]
Initial state: The florist stands behind a workbench with stems, scissors, and wrapping paper.
Primary event: The florist arranges and trims the bouquet.
End state: The florist holds the bouquet in the left hand; the scissors are back on the right side of the bench.
[Stage 2]
Continue from the previous stage. The assistant unfolds wrapping paper and the florist places the bouquet inside.
End state: The wrapped bouquet lies in the center of the workbench with its green ribbon facing the camera.
[Maintain Consistency]
Keep both identities, clothing, workbench orientation, scissors position, and bouquet ownership consistent.
Use timestamps only when a critical handoff, entrance, transition, or beat needs extra pacing guidance. Consecutive ranges such as 0-5 seconds, 5-10 seconds, and 10-15 seconds are a time budget, not a frame-accurate edit decision. Avoid asking for too many actions in a short interval.
Prompt Video Editing, Keyframes, and Extension Carefully
Editing prompts need a strict boundary. Treat the source clip as the sole editing master; then state the edit target, scope, reference material, and content to preserve.
[Edit Goal]
Edit @Video 1. Only from 4-7 seconds, change the cool blue light on the right wall to warm orange light.
[Source Video Role]
@Video 1 is the sole editing master. It defines the character, room layout, action, composition, camera movement, audio, and event order.
[Edit Scope]
Change only the wall light and the illuminated area.
[Content to Preserve]
Keep identity, clothing, expression, position, motion, room structure, camera movement, dialogue, and ambience from @Video 1.
Some task types lock parameters automatically. Video editing preserves the source aspect ratio and approximately its duration, so neither can be set separately. First-frame and first-and-last-frame generation use the first image's aspect ratio; use matching ratios for the first and last image to avoid stretching. Video extension preserves the source aspect ratio, while the extension duration can be set. These rules are constraints, not prompt details to repeat.
Case video: transfer a source performance
This example assigns the source video to performance, movement, and audio, while the image reference supplies the character. It illustrates a narrow, named media-role workflow rather than a promise of frame-perfect preservation.
Use One-Click Video and Transitions as Directed Workflows
For one-click video, state material roles, image order, how much each image should move, editing rhythm, visual treatment, and audio. Do not write only “turn these images into a video.” If upload order matters, say so explicitly; if it does not, permit a thematic order.
For a seamless transition, define the before clip, after clip, trigger action, camera movement, visual transformation, arrival state, and audio handoff. A bridge can aim for visual and audio continuity, but it is not a pixel-identical splice of two existing clips.
@Video 1 is the rainy street before-transition clip. @Video 2 is the gallery after-transition clip.
At the end of @Video 1, a red umbrella fills the frame. The camera continues forward as the umbrella's circular edge becomes the gallery skylight.
End at @Video 2's opening composition and fade rain into reverberant footsteps.
Describe Emotion and Camera in Visible Terms
Words such as “tense” or “warm” set direction but leave room for interpretation. Add two to four observable cues: a change in gaze, brow tension, mouth movement, breathing, hand movement, or posture.
After the applause begins behind the stage, the actor's fingers stop on the program and the gaze turns toward the curtain.
After confirming the curtain call is over, the shoulders relax, a restrained smile appears, and the eyes slowly well with tears.
Common camera terms—wide shot, close-up, push in, pan, orbit, follow shot, low angle, and overhead view—can be written directly. For a niche technique, pair the term with the target and visible result. For example, do not only say “rack focus”; say that focus shifts from foreground leaves to the person in the background, leaving the leaves soft and the face sharp.
Pre-Generation Checklist
- Does the prompt name the subject and main event?
- Does every reference state what to use and, where needed, what not to use?
- Is every character, product, prop, and scene mapped separately?
- Does each long-video stage have one primary change and a visible end state?
- Are identity, clothing, prop ownership, spatial direction, and audio relationships consistent?
- For editing, have you named the sole master video, edit scope, and preserved content?
- Are emotional and camera directions observable rather than abstract?
Use the showcase to compare your intended density and camera language with finished examples. If the brief is still vague, use chat to turn it into a structured scene before generating. When your workflow is clear, review pricing based on how often you expect to iterate.
FAQ
What is the best Seedance 2.5 prompt structure?
Start with the subject and its primary action, then add the scene, visual treatment, camera behavior, and audio only when they materially affect the output. For references, state each file's role and exclusions before the scene instruction. A clear hierarchy is more useful than a long, unstructured paragraph.
How many Seedance 2.5 reference materials should I use?
Use the fewest references that establish the identities, scene, motion, and audio you need. The guide allows up to 50 materials, but recommends smaller, distinct sets for stability. Add more only when each asset has a specific, named job in the current scene.
How do I keep multiple characters consistent?
Name each character, bind each to a specific image, and state which appearance attributes to use. Separately assign props and scenes, and explicitly prohibit swapping clothing, actions, positions, or dialogue. For multiple scenes, repeat the continuity rules and visible end state at each handoff.
How should I write a 30-second Seedance 2.5 prompt?
Break the story into consecutive stages. Each stage should contain one main event and end with a visible state that the next stage can inherit. Use timestamp ranges only to allocate pacing around a critical moment; do not use them as promises of frame-accurate cuts.
Can a prompt preserve an edited video exactly?
No. A precise editing prompt can define a master video, a narrow edit scope, and the content to preserve, which improves control. It does not guarantee frame-by-frame overlap. Editing also automatically preserves the input aspect ratio and approximately preserves its duration.
Should I include camera specifications such as focal length?
You can, but the intended visible result is usually clearer. State the subject, focus behavior, movement direction, speed, and composition change. For uncommon techniques, explain what the foreground and background should visibly do instead of relying on a term alone.