Core structure
What makes a Seedance 2.0 prompt actually land?
Most Seedance 2.0 clips fail because the prompt describes a static picture rather than a dynamic shot.
Seedance 2.0 generates motion, audio, and multiple camera angles in one pass. A static, image-style prompt wastes the model's core strengths. If you don't define the action, the camera, and the sound, the model guesses—leading to inconsistent results.
Structure each prompt as Subject → Action → Setting → Camera → Lighting → Audio. Use the templates below, address your files as [Image N] / [Video N] / [Audio N], and describe the sound explicitly so the native audio track has a clear render target.
A prompt lands when it describes change over time. Give the model a job, not a riddle.
The six fields
- Subject: Who or what, with 2–3 concrete details like hair, wardrobe, or material.
- Action: A real verb—turns, sips, sprints, dissolves—not just a static state.
- Setting: Where it happens, including the time of day.
- Camera: One move per clip: static, slow pan, dolly in, orbit, or handheld.
- Lighting: Light source and color, e.g., 'cool blue window light, warm sodium lamp'.
- Audio: Describe the sound out loud: dialogue, ambient, sound effects, music.
Copy the templates below and change the nouns to fit your scene.
Copy-paste starters
Video templates to start from
These templates are built around the model's real levers: multimodal references, lens switches, and native audio.
Cinematic
Kisses - 10s
A romantic sequence using lens switches for emotional impact.
Close-up of two lovers in a dimly lit jazz club. Lens switch to a wide shot as they lean in. Soft warm spotlight, shadows on the brick wall. Ambient jazz music, clinking glasses, soft whispers.
- Use a lens switch to transition from the close-up to the wide shot.
- Ensure the audio cue matches the club atmosphere.
Action
Beach lift with jet overhead
Combining character motion with environmental scale.
A man lifts a woman on a sun-drenched beach. A jet flies overhead in the distance. Slow orbit camera move. Bright natural daylight, high contrast. Sound of waves, jet engine roar, laughter.
- The orbit camera move adds depth to the beach setting.
- Describe the jet sound clearly for the native audio engine.
Urban
Dance at parking lot
Capturing rhythmic motion in a stark urban environment.
A dancer in street clothes performs a fluid routine in a concrete parking lot at dusk. Handheld camera tracking the movement. Cool blue ambient light with harsh orange street lamps. Rhythmic bass-heavy track, footsteps on concrete.
- Use a handheld camera move to give the dance a raw, energetic feel.
- The lighting contrast between blue and orange defines the mood.
Atmospheric
Dancing at the rooftop
A high-angle sequence focusing on movement and city lights.
A couple dancing on a rooftop terrace overlooking a city skyline. Slow dolly out. Golden hour lighting, warm glow on faces. Distant city traffic, wind, soft melodic music.
- A slow dolly out reveals the scale of the city.
- Ensure the music cue is specific to guide the audio generation.
Cinematic technique
How do you write a cinematic Seedance 2.0 prompt?
Cinematic quality comes from restraint, not from stacking adjectives.
Pick one camera move per clip. An orbit, a dolly, and a pan in five seconds will look broken. Commit the lighting to a single source and color rather than asking for 'beautiful lighting.'
Seedance 2.0 animates and scores audio in one pass. Motion and sound must be written into the prompt. If you don't name the sound, the model has nothing to render.
Use the soundtrack to reinforce what the viewer can see. Footsteps should match the surface, ambience should match the location, and dialogue should be written as an exact line in quotation marks.
Rules for restraint
- One camera move per clip.
- One light source and color palette.
- Describe the sound out loud.
- Focus on action verbs.
- A verb that changes the scene over time
- Enough duration for every requested beat
Reference system
How do multimodal reference prompts work in Seedance 2.0?
The reference system is why creators choose Seedance 2.0—it turns a prompt into a director's brief.
| Reference Type | Capacity | Role |
|---|---|---|
| [Image N] | Up to 9 | Character consistency, background, or wardrobe |
| [Video N] | Up to 3 | Motion vocabulary and camera rhythm |
| [Audio N] | Up to 3 | Voice, music bed, or ambient track |
Address files as [Image N], [Video N], and [Audio N] in your prompt.
Reference recipes
Prompts for image, video, and audio references
Use bracket tokens as assignments, not decorative tags.
Two images
Character plus environment
Lock an identity and place it in a separate location reference.
[Image 1] is the main character; preserve the face, hairstyle, and olive jacket exactly. [Image 2] is the environment. Place the character in its doorway, walking slowly toward camera while dust moves through the afternoon light. Warm interior practicals, cool daylight outside. Boots on wood, distant birds, quiet room tone.
- Use a clean front or three-quarter character image.
- State which details belong to each reference.
Image + video
Borrow motion and pacing
Transfer the movement logic without losing the subject identity.
Use [Image 1] as the performer and copy only the walk, turn, and camera pacing from [Video 1]. Place the performer on a neon street after rain, keeping the face and clothing from [Image 1]. Match the timing of [Video 1] while changing the environment and color palette. Footsteps, wet traffic, distant music.
- Say what to copy from the video and what not to copy.
- A short reference with one readable action transfers best.
Image + audio
Voice-led café scene
Use a face reference and an audio file as the speaking track.
[Image 1] is seated at a small café table and speaks the line from [Audio 1]. Match mouth movement, pauses, and head gestures to [Audio 1] while preserving the face. Soft daylight from the left, shallow depth of field, static medium shot. Use [Audio 1] as the primary voice and add only quiet espresso-machine ambience.
- Keep the face large enough for readable lip motion.
- Do not ask background sound to overpower the voice reference.
Multi-shot control
How to use “lens switch” inside one clip
The cue marks a cut to a new angle while the subject and scene continue.
Three-shot sequence
Warehouse suspense
A longer clip with enough time to establish three connected beats.
Wide shot: a detective in a rain-dark trench coat approaches an abandoned warehouse door at night. Lens switch to a close view of one gloved hand pressing the rusted handle. Lens switch to a low angle inside as a single bulb swings above the doorway and the detective enters frame. Tense low strings rise across all three shots, hinge creak, echoing steps.
- Use 12–15 seconds for two lens-switch beats.
- Carry wardrobe, lighting, and sound across every angle.
One quick cut
First sip reaction
A compact two-shot sequence that fits a shorter render.
Close-up of espresso filling a white cup on a marble counter, steam curling into warm morning light. Lens switch to the customer’s face as she takes the first sip and smiles. Slow push-in on the reaction. Pouring liquid, ceramic tap, a soft satisfied breath, quiet café murmur.
- One cut is enough for a 5-second clip.
- Describe the continuity object—the cup—on both sides of the cut.
Native controls
Prompts for audio, first/last frames, and a fixed camera
Turn on the matching generation control, then make the prompt describe only the motion that control should govern.
Native dialogue
Barista line with room tone
State the spoken words and the surrounding audio layers explicitly.
A barista leans across the counter, smiles, and says, “The usual today?” Cozy café at sunrise, static medium shot, soft window light from camera right. Clear natural voice, espresso-machine hiss, cup set on wood, low conversation behind her.
- Put exact dialogue in quotation marks.
- Keep ambience quieter than the spoken line.
First + last frame
Door-to-library transition
Use supplied start and end images to define a controlled transformation.
Begin exactly on [Image 1], the closed antique door. End exactly on [Image 2], the candlelit library fully revealed. Over the clip, the handle turns, the door opens, and the camera makes one slow push through the threshold. Preserve the door geometry throughout. Wood creak, low room tone, small candle flickers.
- Make the beginning and ending images compositionally compatible.
- Describe the bridge between them, not two unrelated scenes.
Fixed camera
Locked product turntable
Keep the frame immobile and direct only the subject movement.
Fixed camera. [Image 1] is the exact perfume bottle. The bottle rotates once on a white pedestal while the camera remains completely locked. Clean softbox from the left, narrow rim light from behind, controlled reflections on the glass. Quiet studio tone and one subtle mechanical rotation sound.
- Enable fixed camera in the model controls.
- Avoid any pan, orbit, zoom, or handheld language.
Generate and refine
A fast workflow for testing Seedance prompts
Change one variable at a time so you can tell what improved the result.
Start with one subject, one action, one camera move, one lighting design, and one sound layer. Generate a short draft before adding a lens switch or a larger reference stack.
If identity drifts, improve the character image and make [Image 1]’s role more explicit. If motion breaks, simplify the action or lengthen the clip. If the camera wanders, remove competing camera verbs or enable fixed camera where appropriate.
Once the visual beat works, refine ambience, dialogue, and music. Audio is part of the original generation, so it deserves the same concrete direction as the camera.
Debug by symptom
- Static result: add a visible action and a camera verb.
- Chaotic motion: remove secondary actions and extra camera moves.
- Identity drift: strengthen and explicitly assign the image reference.
- Weak soundtrack: name the exact voice, effect, ambience, or music cue.
- Broken multi-shot sequence: reduce lens switches or increase duration.
- Product geometry drift: use a clean image reference and fixed camera.
Common questions
Seedance 2.0 prompting FAQ
What is the best Seedance 2.0 prompt structure?
Use Subject → Action → Setting → Camera → Lighting → Audio. Give the subject concrete visible details, use a real action verb, select one primary camera move, commit to a light source and color, and name the sound you expect.
How many reference files can Seedance 2.0 use?
The full Seedance 2.0 model interface can support up to 9 images, 3 videos, and 3 audio files, although the exact controls depend on the provider. Address exposed references as [Image N], [Video N], and [Audio N], and give each file one explicit role.
What does “lens switch” do?
It tells Seedance to cut to a new camera angle while continuing the same scene. Use one switch in a short clip; reserve two or three beats for 12–15 second sequences.
How do I get audio in a Seedance 2.0 clip?
Enable native audio and describe what should be heard. Put spoken lines in quotation marks, identify ambience and effects, and reference an uploaded sound as [Audio N] when exact timing or voice matters.
What resolution and duration can I generate in VideoAny?
Seedance 2.0 currently supports 480p or 720p and clips from 5 to 15 seconds. Short clips suit one beat; longer clips leave room for multi-shot direction and more complete audio.
Should I use text-to-video or image-to-video?
Choose text-to-video when the scene can be invented from direction alone. Choose image-to-video when the character, product, wardrobe, or opening composition needs to remain anchored to a reference.
Put it into practice
Paste one prompt, replace the nouns, and direct the first shot
Start simple, verify the action and camera, then add the reference files, lens switches, and audio controls available in your current interface only when each extra control has a clear job.
- One action and camera move first
- Assign every available [Image], [Video], and [Audio]
- Name the sound you want rendered