Seedance Image-to-Video Prompts: A Motion-First Guide

A Seedance image-to-video prompt should explain what happens after the reference image, not redescribe every pixel already visible. The image has already chosen the subject, wardrobe, setting, color palette, and opening composition. Your prompt has a narrower job: direct the motion, protect important details, and define where the shot should end.

This motion-first method is useful for Seedance and Doubao Seedance workflows because it separates instructions the model can act on from decorative language that adds little control.

Use a five-part Seedance prompt formula

Build the prompt in this order: primary subject motion, camera behavior, environmental response, stability constraint, and final state. A compact example is: “The woman slowly turns toward the window. The camera makes a gentle lateral track to the right. Curtain fabric and loose hair move in the same soft breeze. Preserve her facial features and the room geometry. The movement settles on a clean three-quarter portrait.”

Each sentence has one role. You can remove or replace a section without rewriting the whole prompt, which makes testing faster.

Read the reference image as frame one

Before writing, identify what the image makes physically possible. Check the direction of the subject’s gaze, the amount of open space around the body, which objects are partially hidden, and whether the image contains foreground and background layers. A seated subject can lean, look, breathe, or reach; asking that subject to sprint out of frame requires the model to invent far more unseen anatomy and space.

Use motion that continues the existing pose. If a coat is already lifting in the wind, continue that direction. If a car points toward the left edge, either track with it or reveal more space with a pull-back. The strongest image-to-video prompts cooperate with the source frame.

Give the subject one primary action

Short shots rarely need a complete performance. Choose one readable action such as opening a box, lifting a cup, turning toward a sound, stepping into light, or looking up as rain begins. Small secondary motions can support it, but they should not compete.

Instead of “She walks, smiles, waves, turns, and the city transforms,” try “She takes two measured steps toward the neon doorway and pauses as its light reaches her face.” The second version has a clear cause, action, and ending.

Describe one controlled camera move

Camera instructions work best when they include direction and pace. “Slow push-in toward the product,” “steady side tracking as the cyclist moves,” and “gentle pull-back to reveal the full room” are more actionable than “cinematic camera.” If the composition is already strong, a locked camera with subtle subject motion may preserve identity better than an ambitious orbit.

Avoid combining a fast zoom, orbit, crane, and handheld shake in one short clip. When you need a complex sequence, generate separate shots and edit them together.

Make environmental motion agree

Hair, fabric, leaves, smoke, reflections, and dust can make a shot feel alive, but they need a shared physical direction. During a rightward breeze, loose elements should not move independently in opposite directions. During a camera push, foreground layers should pass faster than the distant background.

Write environmental motion as a response: “As the train arrives, warm reflected light travels across the platform and lifts the edge of her coat.” Cause-and-effect language often creates a more coherent result than a list of effects.

Protect the details that matter

Name only the details that must remain stable: facial identity, product shape, printed label, horizon, vehicle design, or architecture. A useful constraint is specific: “Keep the bottle centered and preserve its rectangular silhouette and label placement.” Generic phrases such as “perfect consistency” provide less guidance.

When a reference contains an existing hand, reflection, or small text error, repair the still first. Animation tends to amplify those weaknesses.

End on a deliberate frame

Define how the movement resolves. Ask the subject to pause, the camera to settle, or the product to finish centered. A usable ending makes the clip easier to cut, loop, caption, or use as a thumbnail. It also limits late-clip drift because the generation has a target state.

Diagnose one failure at a time

If identity drifts, reduce subject and camera motion before rewriting the style. If the camera moves incorrectly, remove secondary actions and state one direction. If the clip becomes chaotic near the end, shorten the action or request an earlier settle. If the scene looks frozen, add one physical response such as fabric, light, steam, or parallax.

Keep the source image, prompt, and output together. Save the phrases that work for portraits, products, landscapes, and character shots. A small tested Seedance prompt library will become more useful than a huge collection of unverified examples because it reflects your own visual material and review standard.