Prompt field guide

Seedance 2.5 prompt guide

A production-oriented way to plan subjects, action, camera, look, audio, and continuity before Seedance 2.5 generation becomes available on Cineriq.

Important context

These are planning templates, not claimed 2.5 outputs

Seedance 2.5 generation is not active on Cineriq. The examples below use model-neutral directing practices and the categories described in official Seedance 2.0 materials. Final syntax, supported controls, input limits, and output behavior must be verified after production documentation is available.

Last reviewed:

Prompt anatomy

Seven fields for a directable shot

You do not need every field in every prompt. Use the smallest set that makes the intended video observable, ordered, and testable.

  1. 1

    Subject

    Name the person, object, product, creature, or environment that anchors the shot.

  2. 2

    Action

    Describe observable movement in chronological order with one clear action beat at a time.

  3. 3

    Environment

    Set location, time, weather, surface, background activity, and spatial relationships.

  4. 4

    Camera

    Specify shot size, camera position, movement, lens character, focus, and transition intent.

  5. 5

    Look

    Define lighting, palette, contrast, material texture, realism, and production design.

  6. 6

    Audio

    If the released interface supports it, state dialogue, ambience, effects, music, and timing.

  7. 7

    Constraints

    State continuity requirements and exclusions without contradicting the requested action.

Reusable structure

Start with a shot brief

Fill only the relevant lines, remove labels if the released interface works better with natural prose, and keep one versioned copy of every prompt used in testing.

SUBJECT:
ACTION:
ENVIRONMENT:
CAMERA:
LIGHT + COLOR:
AUDIO (only if supported):
CONTINUITY:
EXCLUDE:

Prompt examples

Six production-oriented starting points

Adapt these prompts to assets you own or have permission to use. They have not been presented as videos generated by Seedance 2.5.

Example 01

Cinematic product reveal

A controlled hero shot with material detail and restrained motion.

Planning example

A matte-black wireless speaker stands on wet volcanic stone at blue hour. Fine mist moves behind it while a narrow rim light reveals the grille texture. Start on an extreme close-up of a water droplet, then make a slow, level dolly backward to reveal the full product centered in frame. Premium commercial lighting, deep charcoal palette, realistic reflections, no text, no logo distortion, no extra objects.

Why this structure helps: The shot has one subject, one camera move, a defined reveal, and explicit continuity constraints.

Example 02

Character action

A readable action beat with a clear beginning, middle, and end.

Planning example

At dawn, a courier in a yellow rain jacket runs across a narrow rooftop, plants one foot on a low concrete wall, vaults cleanly over it, lands, and looks back over her left shoulder. The camera tracks beside her at waist height with natural handheld energy but stable framing. Wind moves the jacket and loose cables. Overcast soft light, realistic city texture, preserve the same face, jacket, and bag throughout.

Why this structure helps: Chronological verbs reduce ambiguity, while the final sentence names the continuity anchors.

Example 03

Three-shot micro story

A compact narrative organized by shot rather than one dense paragraph.

Planning example

SHOT 1 — Wide static frame: an empty neighborhood cinema at night, rain reflecting the marquee lights. SHOT 2 — Medium tracking shot: the projectionist walks down the aisle carrying one glowing film canister. SHOT 3 — Close-up: she opens it in the projection booth and warm light fills her face. Keep the same woman, coat, canister, and color palette across every shot. Quiet rain ambience, soft projector hum, no dialogue.

Why this structure helps: Numbered shots make the intended edit structure and cross-shot continuity easy to audit.

Example 04

Image-to-video direction

Animate a still while protecting composition and subject identity.

Planning example

Use the reference image as the exact opening composition. Keep the subject’s face, clothing, body proportions, and background architecture unchanged. Add a gentle breeze through the hair and tree leaves; the subject blinks once and slowly turns toward the window. Make a subtle five-percent camera push-in with no reframing. Preserve natural skin texture and window reflections. Do not introduce new people, objects, signage, or camera cuts.

Why this structure helps: The prompt separates locked reference properties from the small set of allowed changes.

Example 05

Food texture study

Macro motion with physical cues instead of broad quality adjectives.

Planning example

Macro close-up of a chef breaking the caramelized top of a crème brûlée with the back of a silver spoon. The shell cracks into irregular shards, the custard compresses slightly, and a small curl of steam rises. Side light from a warm restaurant window reveals crisp sugar texture and glossy custard. Locked camera, shallow depth of field, realistic hand motion, no jump cuts, no extra utensils.

Why this structure helps: Concrete physical events give the model visual evidence to resolve instead of relying on “high quality.”

Example 06

Dialogue and ambience plan

Separate spoken words, performance, environment, and sound cues.

Planning example

Medium two-shot inside a stationary train carriage at sunrise. A tired traveler looks at her companion and quietly says, “We missed the city, but not the morning.” Deliver the line naturally with a small relieved smile. Outside light sweeps across both faces as another train passes. Audio plan: restrained carriage hum, one distant rail squeal, no music. Keep eyelines, lip timing, and background direction consistent.

Why this structure helps: Use audio directions only when the released model and interface document those controls.

Reference workflow

Lock what matters, direct what changes

Official Seedance 2.0 materials describe reference-based control using images, video, and audio. That is a confirmed 2.0 baseline, not a promise of identical Seedance 2.5 limits.

Identify each reference

Give every asset a role: subject identity, wardrobe, location, motion, camera, sound, or style.

Name locked properties

State which face, proportions, colors, materials, layout, or voice characteristics must persist.

Name allowed changes

Describe the exact action, expression, camera move, transition, or environmental motion to introduce.

Resolve conflicts

If two assets disagree, state which one controls identity, composition, motion, and style.

Preflight review

Check the brief before spending a generation

Can every required subject and prop be identified without inference?
Are actions written in the order they should happen?
Is the camera instruction physically compatible with the scene?
Are reference elements separated into locked and changeable properties?
Do style words describe observable lighting, color, texture, or movement?
Are continuity requirements specific across shots?
Are exclusions short, relevant, and free of contradictions?
Do you have permission to use every reference image, voice, face, logo, and track?

Rights and safety

References need permission, not just a prompt

Use faces, voices, music, footage, trademarks, and private information only when you have the necessary rights and consent. The official Seedance 2.0 launch materials specifically note identity verification or prior legal authorization for real human portrait references. A prompt does not remove those obligations.

Primary-source context

Confirmed 2.0 reference material

Open the generator preview

Draft a prompt in the non-generating AI Video preview.

Track release status

Check which Seedance 2.5 details are actually confirmed.

Compare 2.5 vs 2.0

Review confirmed differences and remaining unknowns.