Home / Prompt Guide

Prompt Guide for
Seedance 2.0

Seedance 2.0 doesn't need poetry. It needs direction. Think of your prompt as a shot list, not a creative writing exercise. This guide covers everything you need to get cinematic results from the model.

landscape The #1 Rule: Scene First, Action Second

Build the world first. Then fill in the action.

Testing across 1,000+ community generations confirms: describing the scene before the action improves visual consistency by 30-50%. Seedance builds the "static world" (composition, background, lighting) first, then layers in movement.

CORRECT ORDER

Scene: A dark cyberpunk alley at night, neon blue and pink lights reflecting on wet pavement, light rain falling. Then action: A man slowly turns toward the camera, glowing cybernetic eyes activating.

WRONG ORDER

A man turns toward the camera with glowing eyes in some kind of dark alley with neon lights...

Background drifts because the model invents the scene while processing the action.

attach_file The @ Reference System

The @ system is the core of Seedance 2.0's Omni mode. When you upload images, videos, or audio as references, they get labeled @Image1, @Video1, @Audio1, etc. You then tell the model exactly what role each file plays.

@Image1 as the character reference. @Video1 for the fluid dance movement. Camera executes a 360-degree orbit around the dancer. High-contrast lighting with volumetric rays.

What You Can Reference

FROM IMAGES
  • Character appearance & outfit
  • Scene/environment design
  • Product details & branding
  • Visual style & color palette
  • First or last frame
FROM VIDEOS
  • Camera movement & technique
  • Choreography & action sequence
  • Editing rhythm & pacing
  • Visual effects & transitions
  • Background music & sound

Reference Patterns

CHARACTER + MOTION TRANSFER
@Image1 performs the dance from @Video1. Maintain face consistency, fluid motion, medium shot, 4K cinematic.
CAMERA WORK REPLICATION
Reference @Video1's camera movements and transitions. Replace the subject with the product from @Image1. Smooth tracking shot, close-up details, 4K.
MULTI-SCENE WITH REFERENCES
The man from @Image1 walks through @Image2's corridor. Close-up: he takes a deep breath. He enters the room — the environment references @Image3. Natural dialogue throughout.
STYLE + EFFECTS TRANSFER
Black-and-white ink-wash style. The character from @Image1 references @Video1's effects and movements — performing a tai chi sequence in a misty mountain setting.

Pro tip: Be explicit about what to extract from each reference. "Reference @Video1" is vague. "Reference @Video1's camera movement and editing rhythm" is precise. The more specific you are about each file's role, the better the result.

function The Formula

Every good prompt follows the same spine. Five elements, in this order. The model reads top-down and gives the most weight to whatever comes first.

1. Subject
Who / What
→
2. Action
Does What
→
3. Camera
How It's Shot
→
4. Style
Visual Look
→
5. Constraints
Guardrails
Subject first

Pins the model to a center of gravity. Multiple subjects = split attention and drift.

Action second

The kinetic anchor. Tells the model what must keep moving even if everything else shifts.

Camera third

Sets framing logic so the model doesn't re-decide the lens every few frames.

Style & Constraints last

Adds flavor without hijacking motion. Constraints act as guardrails for faces, hands, lighting.

A bearded fisherman in a yellow raincoat hauls a heavy net from dark water, straining against the weight. Low-angle tracking shot from the side, handheld sway. Overcast morning light, desaturated blues, documentary grain. Face and clothing stay consistent, no distortion.
Subject Action Camera Style Constraints

Optimal length: 30 – 100 words. Short and structured beats long and poetic every time.

stacked_bar_chart 4 Levels of Prompting

Not all prompts need to be novels. Match your detail level to your ambition.

LEVEL 1 Minimal (30 words)

A [subject] in [scene]. [One action]. [Camera], [style]. No distortion.

Works for simple shots. Unpredictable results ~4/10.

LEVEL 2 Structured (50-100 words)

Follows the 5-part formula. One style anchor, one camera move. Clear constraints.

Good results ~7/10. The sweet spot for most use cases.

LEVEL 3 Multi-Shot (100-200 words)

Shot-by-shot descriptions with transitions. Seedance handles natural cuts within one video.

Cinematic results ~7/10. Great for storytelling.

LEVEL 4 Director's Script (200-500+ words) WOW RESULTS

Full shot-by-shot screenplay with timestamps. Each shot describes physics, materials, particles, lighting, camera, and audio separately.

Best results for showcase content. This is what produces viral videos.

Sweet spot: 200-350 words, 3-4 shots, 40-60 words per shot. Going beyond 500 words causes the model to nail the first 2-3 shots but improvise on later ones — merging events, reordering, losing details. Keep your STYLE header rich (30-40 words), but each shot description tight: 1 action + 1 camera move + 2-3 physical details. Better 3 precise shots than 5 approximate ones.

FORMAT: 15s / 5 shots / [genre] / no dialogue
STYLE: [lenses, lighting, textures — in detail]

0:00-0:03 Shot 1: THE [NAME].
  [What's visible, physics, materials, particles, light]

0:03-0:06 Shot 2: THE [NAME].
  [Camera, motion, object interactions, textures]

0:06-0:10 Shot 3: THE [NAME].
  [Escalation, scale change, emotional shift]

...each shot detailed separately with physical interactions

videocam Camera Language

This is where Seedance shines. It understands real cinematography terms natively. Use them instead of vague descriptions like "dynamic camera".

One camera movement per shot. Compound moves cause chaos.

Term What It Does Best For
dolly in / outCamera physically moves toward or awayEmotional reveals, product shots
pan left / rightHorizontal rotation on fixed axisEstablishing shots, landscapes
tilt up / downVertical rotationScale reveals, head-to-toe intros
tracking shotCamera follows alongside subjectWalking, running, driving
orbit / 360Camera circles around subjectProduct showcases, hero intros
crane up / downVertical lift of entire cameraEpic establishing shots
push inSlow advance toward subjectBuilding tension
pull outSlow retreat from subjectEndings, reveals
staticNo movementDialogue, stillness

Feel Modifiers

handheld — documentary shake gimbal — smooth & stabilized steadicam — fluid follow aerial — overhead, grand POV — first-person
Combining movements?

Write them as beats, not simultaneous: "Start: slow dolly in for 3 seconds. Then: gentle pan right for the final 2 seconds." Seedance handles sequential much better than simultaneous.

palette Style & Mood

Pick one strong visual anchor. Don't stack adjectives. The model handles a single clear direction much better than five competing aesthetics.

DON'T

beautiful cinematic epic professional high-quality moody dark

DO

35mm film grain, overcast natural light, muted blues

Tested Style Anchors

DRAMA

anamorphic lens, shallow DOF, teal-orange grade

DOCUMENTARY

handheld, natural light, slightly desaturated

COMMERCIAL

soft high-key lighting, clean white studio, 60fps

NOIR

high contrast, hard shadows, rain on glass

DREAMLIKE

soft focus, muted pastels, slow motion

RETRO

16mm film stock, faded warm tones, visible grain

speed Motion & Physics

Seedance simulates physics. Vague motion descriptions produce vague results. Tell it how things move, not just that they move.

VAGUE

a flag moves

SPECIFIC

a flag snaps and ripples in gusting wind, fabric pulling taut then releasing

VAGUE

she walks

SPECIFIC

she walks deliberately, heels clicking on wet pavement, coat swaying

VAGUE

the car turns

SPECIFIC

the car drifts through the corner, rear tires smoking, weight shifting outside

Key rule: always add intensity. "wings flap" is generic. "wings flap powerfully, displacing dust" is cinematic.

face Character Consistency

Without explicit constraints, faces morph and outfits change between frames. This is the single most common mistake beginners make. Always add constraint keywords.

The Constraint Block

Consistent facial features throughout, no warping or distortion. Body proportions remain natural. Same outfit, same hairstyle. High detail, no flickering.

What to Lock Down

FACE

Clear facial features, stable face, no morphing, no deformation

BODY

Normal body proportions, natural structure, no stiffness, no distortion

CLOTHING

Same outfit throughout, consistent colors, unchanged accessories

ENVIRONMENT

Consistent lighting, no random style drift, stable background

Rule: Add a version of this constraint block to every prompt. It takes 10 words and saves you from unusable output. When using @Image references for characters, also add "same character as @Image1" to reinforce identity lock.

movie_edit Multi-Shot Stories

Seedance can generate natural cuts within a single video. Describe each shot separately.

Shot 1: Wide — a dark workshop, tools on walls, a single lamp swinging.
Shot 2: Close-up — weathered hands carefully assembling a clockwork mechanism.
Shot 3: Medium — the craftsman holds up the finished piece, lamplight catching the gears.

Warm tungsten tones, shallow depth of field, no flicker between shots. Same character throughout.

Tip 1: Keep each shot to one clear action.

Tip 2: Specify lighting changes when the scene shifts (e.g., outdoor → indoor).

Tip 3: Always end with "maintain character consistency across shots".

cinematic_blur One-Take Continuity

Long, unbroken shots with consistent motion, scene transitions, and character persistence — no cuts. Seedance 2.0 excels at "oner" filmmaking when you describe the flow clearly.

TRACKING THROUGH SPACES
One continuous take. Camera pushes through the doorway into the first room, pans left to reveal the corridor, follows the character walking to the window, pushes through the window into the exterior. No cuts. Smooth continuous motion throughout.
IMAGINATIVE TRANSITION
@Image1 as the first frame. Camera zooms into the airplane window. Clouds drift slowly, one cloud morphs into ice cream. Camera pulls back into the cabin — a girl reaches through the window to grab it, takes a bite, beaming smile. One continuous take.
ELEMENT MORPHING
One continuous shot. Chess pieces on a table, camera pans left revealing sand on the floor, tilts up to a beach. A girl in white walks into the distance. Camera cuts to aerial — ocean waves wash the shore. Waves dissolve into flowing curtains. Camera pulls back to reveal the girl's face. Seamless throughout.

Key phrases: "One continuous take", "no cuts throughout", "seamless", "unbroken shot". These tell the model to maintain temporal coherence instead of cutting between shots.

sentiment_very_satisfied Emotion & Performance

Seedance 2.0 can convey nuanced emotions through facial expressions, body language, and voice. The key is describing the emotional arc — how feelings change over time.

EMOTIONAL BREAKDOWN
The woman walks to a mirror. She looks at her reflection, pauses in thought, then suddenly breaks down. Her expression shifts from contemplation to overwhelming sadness. Tears form. Cinematic close-up, warm interior lighting.
COMEDIC REACTION
@Image1 as the first frame. Camera rotates and pushes in. The character suddenly looks up with wide eyes, mouth drops open in exaggerated surprise. Beat. Then breaks into uncontrollable laughter, shoulders shaking. Comedy timing, bright lighting.
CONTRASTING EMOTIONS
Split composition. Left side: a woman cooking elegantly, calm and composed, warm soft lighting. Camera pans right to: a man sweating, face flushed, frantically cooking in heavy smoke. The contrast between serenity and chaos. 4K cinematic quality.

Emotion Keywords That Work

contemplative gaze tears forming eyes darting nervously satisfied smile exaggerated surprise deep breath, releasing tension intense focused determination expression shifts from calm to intense

trending_up Viral Formats

These formats consistently get the most views and shares. Each one exploits a specific visual hook that makes people stop scrolling.

Transformation

Object A morphs into Object B. The "how did they do that?!" effect.

airplane lands → transforms into a robot

Macro → Cosmic

Zoom into a tiny object and emerge in a vast world. Scale shift = shock.

coffee swirl → galaxy → back to cup

Bullet Time

Freeze a moment. Camera orbits the frozen scene. Time resumes.

freeze at peak action, orbit 360°, resume

Seamless Loop

End state = start state. Infinite replay = TikTok gold.

morph A → B → A, seamless loop-ready

Satisfying Physics

Tactile pleasure from realistic material interactions.

product explodes apart, floats, snaps back together

Escalating Multi-Shot

5-7 shots building from quiet to explosive. Cinematic tension.

macro detail → medium → wide reveal → epic crane

block What to Avoid

close
Multiple complex actions

Model tries everything at once — nothing looks right. One action per shot.

close
Vague words: "beautiful", "amazing", "epic"

These mean nothing visually. Use concrete terms: specific lighting, specific colors.

close
Contradictions: "extremely fast + perfectly stable"

Pick one. The model can't reconcile opposing instructions.

close
200+ word essays

The model loses focus. 30–100 words is the sweet spot.

close
Stacking styles

Five adjectives compete with each other. One strong anchor beats them all.

science Advanced Techniques

Cultural references as style anchors

Phrases like Apple keynote style, Blade Runner 2049 aesthetic, or National Geographic documentary give the model a powerful stylistic anchor. Community testing confirms this is one of the most effective techniques for consistent results.

Audio architecture

Seedance generates native audio synced to video. Describe sounds with specific adjectives: muffled for underwater, echoing for halls, crunchy for gravel, metallic clink for impacts. You can even specify audio as a separate block:

[Audio]: Low-frequency ticking build-up, staccato string rise,
cuts out abruptly at moment of impact.
SFX: leather on turf, wet foley, crunch of debris.

Negative prompts

Add at the end to remove unwanted elements:

no subtitles no text overlays no watermark no music avoid static shots

Content moderation tips

If your prompt gets flagged, try rephrasing without direct mentions of weapons or dangerous actions. Focus on physics (sparks, particles, speed) rather than violence. "Parkour jumps from building" gets flagged; "Runner vaults over obstacles in an urban environment" usually passes. Sometimes a second attempt with the same prompt goes through.

Name your shots

Give each shot a capitalized name: Shot 1: THE CONTACT, Shot 3: THE FLIGHT. This isn't just style — the model separates shots more cleanly when each one has a distinct label. Reduces blending between scenes.

Material specificity

Don't write "a sword" — write "a blade of folded Damascus steel with visible wave patterns." Don't write "a dress" — write "a crimson silk dress that pools on the reflective black floor." Seedance renders materials more accurately when it knows the type:

brushed metal wet leather volcanic obsidian translucent jade hammered copper frosted glass

Emotional arc in 15 seconds

The best prompts have a story arc even in 15 seconds: stillness → build-up → climax → resolution. Example: dark room (0-3s) → light appears (3-6s) → magic intensifies (6-12s) → grand reveal (12-15s). Monotone action = boring output. Rising energy = cinematic output.

Zero-gravity and unusual physics

Seedance handles unusual physics well: zero-gravity (objects float, liquid forms spheres, hair drifts), reverse time, extreme slow motion. Be specific: "coffee mug floats in mid-air, steam forming a perfect sphere around it in zero gravity" works much better than "things float around."

FORMAT/STYLE header

Start your director-level prompts with metadata. This sets a "contract" with the model before the actual scene description:

FORMAT: 15s / 5 shots / [genre] / no dialogue
STYLE: [specific camera/lenses], [lighting], [color palette], [film reference]

content_copy Quick Templates

Minimum Viable Prompt

A [subject] [doing one action].
[Camera movement], [one style anchor].
Consistent appearance, no distortion.

Full Director Prompt

Subject: [who or what, with key visual details]
Action: [one specific continuous motion]
Camera: [shot size] + [one movement] + [speed]
Style: [one visual anchor] + [lighting]
Constraints: [consistency rules]

Multi-Shot Template

Shot 1: [shot size] — [subject + action + setting]
Shot 2: [shot size] — [subject + action]
Shot 3: [shot size] — [subject + action]

[Style anchor], [lighting]. Consistent character across all shots.

Product Showcase Ad

Reference @Video1's camera work and transitions. Replace the product with @Image1. Smooth tracking shot, close-up of product details, cinematic lighting, 4K ultra-HD, professional commercial quality.

Character Dance Transfer

@Image1 performs the dance from @Video1. Maintain character face consistency, fluid motion matching the reference choreography, medium shot, slight orbit, 4K cinematic quality, natural lighting.

Atmospheric Landscape

Seaside sunset, waves gently lapping the shore, camera slowly panning horizontally. Warm orange tones, calming and serene. Silky smooth footage, 4K ultra-high definition, no flickering, no ghosting.

One-Take Transition Sequence

@Image1 as the first frame. One continuous take. Camera pushes through the doorway into the first room (@Image2), pans left to reveal the corridor (@Image3), follows the character walking to the window (@Image4), camera pushes through the window into the exterior (@Image5). No cuts. Smooth continuous motion throughout.

Emotion-Driven Scene

@Image1 as the first frame. The character slowly looks up — expression shifts from calm to intense. Camera rotates and pushes in during the emotional peak. Cinematic lighting, 4K, no deformation. Consistent facial features throughout.

Beat-Synced Photo Montage

@Image1 through @Image6 synced to @Video1's rhythm and beats. Each image transitions on the beat. Characters are dynamic with strong visual tension. Dreamlike visual style. Add dramatic lighting shifts between beats.

Cinematic Nature

FORMAT: 16:9 cinematic, ultra realistic, warm color grade

Massive volcanic cliff rising above turquoise ocean at golden hour. Camera slowly pushes in. Waves crash against the rocks far below. Seabirds circle in the warm updrafts. Volumetric light rays through the sea spray. National Geographic quality. 4K, rich detail.

Style Transfer

Recreate @Video1 in black-and-white ink-wash painting style. Maintain the original camera movements and character actions. Flowing brush strokes, traditional Chinese aesthetic, misty atmosphere, rice paper texture. 4K, cinematic.

Ready to create?

Take these rules and generate your first cinematic video.

Start Generating arrow_forward