Seedance 2.5 Prompting Guide

Seedance 2.5 prompting is not a prettier caption. Treat the box like a director brief: who is on screen, what happens, which reference owns which job, when each beat lands, how the camera moves, what must stay continuous, what you hear, and what to keep out. On Versely you run the three catalog rows (Seedance 2.5, image-to-video, reference-to-video) at 34 credits each, then iterate the brief instead of restacking adjectives.

This guide is the technique layer. For click-by-click studio setup see how to make viral videos with Seedance 2.5. For filling a full 30-second continuous take see how to brief a 30-second Seedance 2.5 take. The Seedance 2.0 prompting guide remains the sibling for shorter trend clips on the 2.0 row.

The director brief stack

Only subject and action are mandatory. Add the other slots in the order your last take failed.

1. Subject - who or what stays readable in frame
2. Action - one verb chain the camera could hold
3. Reference roles - each @Image / @Video / @Audio gets one narrow job
4. Timeline - integer-second beats for multi-event takes
5. Camera - one move per beat, tied to an event
6. Continuity - identity and state that must not reset
7. Audio - dialogue, SFX, ambience, music in the same pass
8. Constraints - exclusions live in the prompt (no dedicated negative field)

Write it as labeled lines if that helps you edit. Seedance reads the clauses either way. What it does not forgive is five unrelated ads glued with commas.

Timestamped beats for multi-event takes

Seedance 2.5 responds to integer-second ranges. Seedance 2.0 largely ignored them. Use contiguous windows with no gaps. Treat stamps as pacing budgets, not frame-accurate cuts.

Weak:

A florist wraps a bouquet, ties ribbon, and hands it to a customer while smiling.

Stronger:

0-6s: A florist in a linen apron gathers stems on a wood counter in soft morning light. 6-14s: She wraps kraft paper, twists it closed, and ties a cream ribbon. 14-20s: She turns the finished bouquet toward camera and offers it to a hand entering from the right.

Give each beat room. Crowding three micro-actions into one second forces hard cuts or dropped events. Match duration to beat count: one clean beat often wants 4 to 8 seconds; three connected beats want closer to 15 to 20; pad to 30 only when events fill the time.

Worked prompt 1 (single beat, short)

FORMAT: 9:16, 6 seconds, photoreal product.
SUBJECT + ACTION: A frosted citrus soda bottle on wet slate. A hand lifts it and pours a steady stream into a tall glass of ice until foam rises.
CAMERA: Slow push-in at close range. Locked framing on bottle and glass.
LIGHT: Bright daylight from upper left, fresh commercial grade.
AUDIO: Crisp fizz of the pour, ice settling, light room tone. No music.
CONSTRAINTS: No text, no subtitles, no warped hands, no extra bottles.

Worked prompt 2 (multi-beat timeline)

FORMAT: 16:9, 18 seconds, grounded kitchen realism.
SUBJECT + ACTION: A woman in a green wool coat sits at a kitchen table with an unopened letter beside a mug.
TIMELINE:
0-6s: She notices the letter, pauses, keeps both hands on the mug.
6-12s: She slides the letter closer, opens it carefully, unfolds one page.
12-18s: She reads silently. Expression moves from concern to quiet relief.
CAMERA: Medium close-up, eye level. Slow push only after she unfolds the page.
CONTINUITY: Same short black bob, green coat, silver hoop earrings, same kitchen throughout.
AUDIO: Soft room tone, paper rustle, distant street. No music. No dialogue.
CONSTRAINTS: No recast, no new room at midpoint, no subtitles, no text overlay.

Narrow reference jobs (and what not to copy)

Uploading a still is not the instruction. Name the asset by order (@Image1, @Video1, @Audio1) and give it one job. State what it must not leak into the take.

REFERENCE ROLES:
@Image1 controls the barista face and hair only. Do not copy clothing, background, or lighting from @Image1.
@Image2 defines the navy apron and cream logo patch only.
@Video1 supplies camera trajectory and pacing only. Do not copy its subject, location, or wardrobe.
@Audio1 supplies voice timbre for one short line only. Do not copy its music bed.

If two images disagree on environment, you created a conflict the model must invent its way out of. Prefer one subject lock, one wardrobe lock, one motion lock, one audio lock. Crowded group photos work better when you crop to a single face first.

Worked prompt 3 (reference-to-video)

FORMAT: 9:16, 12 seconds, editorial fashion, soft daylight.
REFERENCE ROLES: @Image1 defines the model's face, hair, and proportions only. Do not copy pose or studio backdrop from @Image1. @Image2 defines the camel trench coat cut, lapels, belt, and color only. Do not copy @Image2's lighting.
SUBJECT + ACTION: The model walks slowly toward camera, stops, and makes a half-turn so the coat hem flares then settles.
CAMERA: Full-body start. Push to waist-up only after the hem settles. No orbit.
CONTINUITY: Same face, hair, coat construction, belt position throughout.
AUDIO: Soft footsteps on concrete, light fabric rustle, quiet studio. No dialogue. No music.
CONSTRAINTS: No redesign of the coat, no duplicated limbs, no text, no watermark.

One camera move per beat, tied to events

Stacking "dolly, orbit, whip, crane" in one sentence without triggers makes the model guess timing. Tie the move to a cause.

Weak: Orbit, then whip right, then crane up.

Stronger: As the food leaves the pan, the camera orbits the chef's face, then whip-pans right to follow the arc.

Prefer crew vocabulary Seedance already reads: push-in, pull-back, orbit, crane, handheld, locked-off. Separate subject motion from camera motion in different clauses so neither flattens the other.

Native audio in the same pass

Audio is generated with the picture when the Audio toggle is on (the default on Versely). Write an Audio: clause. Split layers when the mix matters: dialogue, SFX, ambience, music.

AUDIO:
DIALOGUE: She looks at the cup and says quietly, "Okay. That one."
SFX: Ceramic clink as the cup meets the saucer.
AMBIENCE: Soft cafe murmur, espresso machine hiss under.
MUSIC: None.

There is no dedicated negative_prompt field on common Seedance 2.5 APIs, so exclusions live in the prompt text: No subtitles. No BGM. No text overlay. No warped hands. Put them at the end so they stay easy to edit.

Duration matched to beat count

Beats in the brief	Sensible length	Why
One clean action peak	4 to 8s	Fast iterate; less empty padding
Two or three connected steps	12 to 20s	Room for cause and aftermath
Full continuous story	up to 30s	Only when events fill the clock

Iterate at 480p when you are still locking motion. Deliver at 720p once the brief holds. There is no 1080p on this row. Catalog details stay on /models/seedance-2-5.

Failure modes

Failure	Likely cause	Fix in the brief
Events pile into the first half	No timeline, or empty late windows	Add contiguous second ranges; cut micro-actions
Face or product drifts mid-take	Vague "use my image"	Narrow @Image1 to face or pack only; name what not to copy
Background jumps with the reference	Reference leaked environment	Explicit exclusion: do not copy background or lighting
Camera feels random	Move list without triggers	One move per beat, tied to an on-screen event
Mix fights the picture	No Audio clause, or stacked VO	Layer dialogue / SFX / ambience; say No BGM when needed
Identity reset after a turn	Continuity not restated at risk points	Re-anchor two visible traits when the subject reappears
Empty 30s with dead air	Duration longer than the events	Shorten the clock or add real beats
Burned-in captions you did not want	Model inventing titles	Constraints: No subtitles, no text overlay

Skip the dialect: brief the Versely agent

You can learn this stack. On Versely you can also skip memorizing every Seedance dialect and brief the agent instead of writing prompts: content type, photos or references, the edits you will accept, target platforms, and a budget ceiling. Name the row when it matters ("use Seedance 2.5 Reference to Video"). The agent plans the job; you approve the plan instead of babysitting clause order.

A prompt still wins when you are hand-tuning one hero take. A brief wins when the job is bigger than one generate: variants, captions, and a post path with a spend cap.

After the take: edit, post, collections

When the Seedance pass is close enough:

1. Light trim and pacing in the studio editor if a beat overruns. Do not ask the model to be CapCut.
2. Burn mute-proof captions with /tools/ai-caption-generator or the on-device /free-tools/burn-captions path. Keep type off faces when the product is the point.
3. Upload or schedule to the social accounts Versely already connects for your workspace. If a network is not connected, export the mp4 and post from the native app.
4. Save keepers into a Versely collection so the next brief reuses the same stills, refs, and winning prompt lines.

Honest limit: Seedance 2.5 is not a full multi-scene movie editor. Board separate generates when you need separate setups, then stitch. The continuous single-pass is the product.

FAQ

What is the best Seedance 2.5 prompt formula?

Subject and action first, then reference roles, timestamped timeline, camera, continuity, Audio, and constraints. Drop any slot you do not need. Add slots in the order your last failure showed up.

How do I write Seedance 2.5 timestamps?

Use integer-second ranges with no gaps (0-7s, 7-14s, 14-20s). They pace events; they are not frame-accurate edit points. Finish hard cuts in an editor.

How should I use reference images?

Address them as @Image1, @Image2, and assign one job each. Always say what not to copy (background, lighting, pose) when the still contains signals you do not want.

Does Seedance 2.5 generate audio?

Yes, in the same pass when Audio is on. Write an Audio: clause. Put exclusions such as No BGM or No subtitles in the prompt text.

How is 2.5 prompting different from Seedance 2.0?

2.5 takes integer-second timelines seriously, supports longer single-pass takes, and leans harder on narrow multimodal reference roles plus native audio. 2.0 still suits short tone-first trend clips; use the 2.0 prompting guide for that dialect.

Where do I run this on Versely?

Open the AI video generator or Imagine with Seedance 2.5 selected. The three rows (text, image-to-video, reference-to-video) are listed at 34 credits each in the catalog. Composer totals still depend on length and quality before you send.

Takeaway

Prompt Seedance 2.5 like a short shoot brief, not like an image caption. Timestamp multi-event takes, give every reference one narrow job, write Audio inline, and park exclusions in the prompt. Run the rows on Versely, or hand the agent a job brief when you would rather not memorize the dialect. Then caption, post where the product supports it, and file the keepers in a collection so the next take starts warmer.