A Seedance 2.5 reference video can give creators more control over character movement, camera behavior, action timing, and scene composition than a text prompt alone. Instead of describing every turn, gesture, tracking shot, or transition in a long paragraph, creators can provide a short video that visually demonstrates how the final scene should move.
This approach is useful for product advertisements, fashion films, dance clips, action sequences, character performances, social media videos, and cinematic camera tests. The uploaded clip can function as a motion blueprint while separate reference images define the final character, product, clothing, environment, or visual style.
A reference video does not always need to resemble the desired final output. A phone recording made in an ordinary room can guide a science-fiction scene. A simple 3D animation can define a cinematic camera path. A performer wearing casual clothes can provide movement for a completely different digital character.
The key is to tell the model exactly what it should take from the reference—and what it should ignore.
Creators can explore reference-based generation with the workflow below to prepare clearer inputs and more reliable prompts.
Try the Seedance 2.5 AI video generator
Important: Seedance 2.5 availability, upload limits, reference labels, and interface controls may vary between platforms. This guide focuses on practical reference-video principles that apply across Seedance-style multimodal workflows.
What Is a Seedance 2.5 Reference Video?
A Seedance 2.5 reference video is an uploaded clip used to guide one or more temporal elements of a generated video.
These elements can include:
Character movement
Camera direction
Action timing
Scene blocking
Motion rhythm
Object interaction
Transition timing
Performance intensity
Lighting changes over time
The model does not necessarily need to copy every visible detail from the source clip. The creator can assign the video a limited role.
For example, a reference video may show a man walking through a kitchen while the camera tracks beside him. The final output could instead show a robot walking through a futuristic laboratory. The body path and camera motion remain useful even though the character, clothing, background, lighting, and visual style are replaced.
This role-based approach makes reference videos more flexible than ordinary video-to-video transformation.
Reference Video vs. Source Video
In this guide, a reference video means a clip used primarily for movement, timing, camera, or staging guidance.
A source video means footage that is directly edited, transformed, extended, or restyled.
Platform terminology may differ, but the practical distinction is useful:
Input | Main purpose | Typical result |
|---|---|---|
Reference video | Guide motion, timing, camera, or blocking | A new scene that follows selected movement |
Source video | Transform existing footage | An edited or restyled version of the original clip |
Reference image | Define identity, product, environment, or style | A generated video that preserves visual details |
Text prompt | Describe the creative objective | A scene based mainly on written instructions |
Audio reference | Guide rhythm, speech, ambience, or sound | Motion and audio aligned to a sound cue |
A strong Seedance 2.5 workflow gives each input a different responsibility instead of asking every reference to control everything.
What Can a Seedance 2.5 Reference Video Control?
The best reference clips usually have one clear purpose. A complex source containing multiple people, rapid cuts, background activity, and several camera changes may create conflicting instructions.
1. Character Movement
A motion reference can communicate physical actions that are difficult to describe precisely in text.
Examples include:
Walking speed
Head turns
Arm gestures
Sitting and standing
Reaching for an object
Turning around
Running or jumping
Dance choreography
Hand movements
Interaction between two characters
Consider the instruction:
The woman turns elegantly and looks toward the camera.
The word “elegantly” does not define the speed of the turn, the position of her arms, the direction of her feet, or the exact moment she looks at the camera.
A reference clip can demonstrate all those details visually.
2. Camera Movement
A Seedance 2.5 reference video can also act as a camera-motion reference.
It may guide:
Slow dolly-in
Dolly-out reveal
Side tracking
Circular orbit
Crane movement
Handheld following
First-person movement
Low-angle push-in
Overhead descent
Whip-pan transition
Drone-style flythrough
Rack-focus timing
Camera references are especially useful when a text-generated result looks visually correct but feels static.
A simple instruction can separate the camera motion from the unwanted content:
Follow only the camera path and shot timing from @Video1. Do not copy the people, clothing, objects, colors, or location from the reference.
3. Action Timing
Longer AI-generated clips may contain the right actions in the wrong order. The model may also perform everything too quickly or skip part of the sequence.
A reference video gives the scene a visible timeline.
For example:
Seconds 0–3: Character enters the frame.
Seconds 3–6: Character approaches the table.
Seconds 6–9: Character picks up the product.
Seconds 9–12: Camera moves into a close-up.
Seconds 12–15: Character presents the product to the camera.
The reference helps communicate not only what happens but when it happens.
4. Multi-Character Blocking
Scenes involving two or more characters are difficult because the model must track multiple identities, body positions, directions, and interactions.
A blocking reference can show:
Where each character begins
Who moves first
Which direction each person faces
When characters cross paths
Who touches an object
How distance changes
Where the camera is positioned
Where each character ends the scene
The reference does not need to look professional. Actors, mannequins, stick figures, or simple 3D models can all communicate staging.
5. Motion Rhythm
Motion rhythm describes the energy and pacing of the scene.
A luxury product advertisement may need slow, controlled movement. A sports commercial may require sharp acceleration. A fashion film may use long pauses and deliberate gestures.
Words such as “dynamic,” “premium,” and “cinematic” are subjective. A reference video can communicate the intended rhythm more clearly.
How to Prepare a Good Reference Video
A professional camera is not required. The most important factor is whether the movement is easy to see and understand.
Keep the Main Action Visible
The relevant motion should remain inside the frame.
Avoid:
The subject leaving the frame
Heavy motion blur
Extremely dark footage
Objects blocking the body
Rapid cuts during the main action
Unnecessary background movement
Camera shaking that is not part of the intended result
A clean silhouette makes character motion easier to interpret.
Use a Simple Background
A plain wall, studio floor, green screen, open room, or basic 3D environment can work well.
The background does not need to resemble the final location. Its purpose is to separate the action from distracting visual information.
If the final video will place a character inside a fantasy palace, the motion reference can still be recorded in a normal room.
Match the Intended Aspect Ratio
Whenever possible, prepare the reference in the same orientation as the final output.
Use:
9:16 for short-form vertical video
16:9 for cinematic or landscape video
1:1 for square social content
A major aspect-ratio change can affect:
Subject placement
Walking distance
Camera path
Negative space
Visible body movement
Final composition
A horizontal movement may not translate perfectly into a narrow vertical frame.
Trim Empty Frames
Remove long pauses before and after the important movement.
If the first three seconds of the source contain no action, the generated result may also delay the main event.
The reference should begin close to the first meaningful movement and end shortly after the final pose or camera position.
Use One Clear Motion Pattern
A focused clip is often more effective than a long video containing unrelated actions.
Good examples include:
One walking path
One camera orbit
One product reveal
One dance movement
One character interaction
One transition
One hand gesture
When multiple types of motion are needed, separate them into different references and assign each reference a clear role.
How to Use a Seedance 2.5 Reference Video
The exact controls may differ across platforms, but the creative workflow generally follows the same steps.
Step 1: Define the Final Scene
Before uploading files, describe the intended result in one sentence.
For example:
A cinematic fashion video showing a woman in a silver dress walking through a futuristic glass corridor while the camera tracks beside her.
This sentence establishes:
Subject
Clothing
Environment
Main action
Camera relationship
Overall style
Without a clear final objective, it is difficult to decide what each reference should control.
Step 2: Assign the Video One Primary Role
Decide what the reference video contributes.
Possible roles include:
Body movement
Camera movement
Timing
Character blocking
Object interaction
Transition structure
Performance energy
Motion rhythm
Do not simply upload a clip and write:
Make it like this video.
That instruction forces the model to guess which details matter.
A clearer instruction would be:
Follow the body movement and action timing from @Video1. Do not copy the actor, clothing, room, lighting, or camera style.
Step 3: Add Visual References Separately
Use images for details that must remain visually consistent.
These may include:
Character face
Full-body appearance
Clothing
Product design
Environment
Color palette
Vehicle
Prop
Logo position
Lighting style
A useful division of responsibility is:
Video controls how it moves.
Images control how it looks.
Text explains what happens.
Audio controls rhythm or sound.
Preservation instructions define what cannot change.
Step 4: Label Each Reference Clearly
When the platform supports reference labels, assign each file a specific job.
For example:
@Video1— body movement@Video2— camera path@Image1— character identity@Image2— clothing@Image3— location@Image4— product@Audio1— music rhythm
Then explain these roles directly in the prompt.
Use @Image1 for the character’s face, @Image2 for the outfit, and @Video1 only for body movement and action timing.
This is more reliable than expecting the model to infer each file’s purpose.
Step 5: Add Preservation Instructions
Preservation instructions explain which details must remain stable.
Examples include:
Preserve the face from @Image1.
Keep the product shape unchanged.
Maintain the same clothing throughout the shot.
Preserve the logo position.
Keep the character’s hairstyle consistent.
Do not add extra people.
Maintain the same lighting direction.
End with the product centered in the frame.
Negative instructions are equally important:
Do not copy the actor from @Video1.
Do not copy the original background.
Ignore the clothing colors in the reference video.
Do not reproduce visible text or logos.
Use only the camera path from @Video2.
Use only the action timing from @Video1.
Step 6: Test One Variable at a Time
After generating the first version, evaluate the result in categories.
Ask:
Did the movement match?
Did the camera follow the correct path?
Did the character remain consistent?
Did the actions happen in the correct order?
Did the timing feel natural?
Did the model copy unwanted details?
Did the product, background, or clothing drift?
Change only one major instruction before the next test.
If the prompt, character image, environment, camera, timing, and reference clip all change together, it becomes impossible to identify what improved or damaged the result.
Seedance 2.5 Reference Video Prompt Formula
A practical Seedance 2.5 reference video prompt can be organized into seven parts:
Final subject + final environment + required action + reference role + camera behavior + preservation rules + visual treatment
Use this reusable structure:
Create a [duration and aspect ratio] video of [subject] in [environment]. Follow the [movement, timing, blocking, or camera path] from @Video1. Use @Image1 for [character or product identity] and @Image2 for [clothing, location, or style]. Preserve [important details]. Do not copy [unwanted details] from the reference video. Use [lighting, lens, atmosphere, and visual style].
The prompt does not need to be extremely long. It needs to clearly divide responsibility between the inputs.
Copy-Ready Seedance 2.5 Reference Video Prompts
Example 1: Fashion Walk
Create a 12-second vertical fashion video of the woman from @Image1 wearing the silver dress from @Image2. Follow the walking pace, posture, arm movement, and final turn from @Video1. Place her inside a futuristic glass corridor at night. Use only the body movement from @Video1; do not copy the original person, clothing, or background. Add soft blue lighting, reflective floors, a premium editorial look, and a smooth side-tracking camera.
Example 2: Product Reveal
Create a cinematic product advertisement for the black smartwatch shown in @Image1. Follow the camera orbit and reveal timing from @Video1. Replace the object in the reference with the smartwatch while preserving its exact screen shape, buttons, strap, and logo position. Use a dark studio background with controlled rim lighting and subtle floating particles. End on a centered close-up of the product.
Example 3: Dance Performance
Create a 15-second performance video using the character from @Image1. Follow the choreography, body timing, foot placement, and arm movement from @Video1. Do not copy the dancer’s face, body shape, clothing, or studio background. Place the final character on a neon-lit rooftop at night. Use a stable full-body composition with a slow forward camera movement and no additional dancers.
Example 4: Cinematic Camera Path
Create a cinematic shot of the red sports car shown in @Image1 inside a modern concrete garage. Follow only the camera path, movement speed, and shot timing from @Video1. Do not copy any people, vehicles, text, colors, or location details from the reference. Begin with a low rear angle, orbit toward the front, and end on a close-up of the headlights. Preserve the vehicle’s exact body shape and design.
Example 5: Two-Character Blocking
Create a 15-second dramatic scene using the characters from @Image1 and @Image2. Follow the starting positions, walking paths, pauses, and interaction timing from @Video1. Character A enters from the left, stops near the table, and hands a sealed envelope to Character B. Preserve both faces and outfits. Do not copy the performers or location from @Video1. Use warm interior lighting and a subtle handheld camera.
Example 6: Food Commercial
Create a close-up food commercial featuring the dessert from @Image1. Follow the hand movement, spoon action, camera push-in, and final product hold from @Video1. Preserve the dessert’s shape, layers, colors, texture, and plate design. Do not copy the original food, hand appearance, or background. Use soft morning light, shallow depth of field, realistic texture, and gentle natural steam.
Example 7: 3D Previsualization
Transform the blocking and camera movement from @Video1 into a realistic science-fiction scene. Use @Image1 for the main character, @Image2 for the spacecraft interior, and @Image3 for the control console. Follow the camera path and character position from the reference, but replace all gray 3D geometry with the final environment. Use cinematic volumetric lighting and maintain spatial continuity throughout the shot.
Common Reference Video Mistakes
Asking the Model to Copy Everything
A source clip may contain a person, clothing, lighting, background objects, sound, camera motion, and several actions.
Writing:
Make a video like @Video1.
does not explain which parts should be preserved.
Use a more specific instruction:
Follow the body movement and timing from @Video1, but replace the character, environment, clothing, lighting, and visual style.
Combining Conflicting References
Two reference videos may show different walking speeds, camera paths, or action timing.
If both are assigned to the same role, the model must decide which one has priority.
Separate their responsibilities:
@Video1 controls body movement.
@Video2 controls camera movement.
@Video3 controls transition timing.
When possible, begin with one primary video reference and add secondary clips only after the first motion works.
Using an Overly Long Clip
Long references may contain empty frames, irrelevant actions, cuts, and camera changes.
Trim the clip to the exact movement needed for the final result. A focused source is easier to interpret than an entire scene.
Hiding the Motion Behind Visual Effects
Flashing lights, foreground objects, heavy blur, fast cuts, and strong camera shake may hide the movement the model needs to follow.
Use a clean motion reference first. Add cinematic styling through text and visual references.
Forgetting Exclusion Instructions
Without exclusions, the model may transfer details that should remain only in the reference.
Useful instructions include:
Do not copy the original actor.
Do not copy the source background.
Do not reproduce visible text.
Ignore clothing colors in @Video1.
Use only the camera path.
Use only the movement timing.
Adding Too Many New Actions
A reference video may clearly demonstrate one action, but the prompt may add four more unrelated actions.
This creates competing instructions.
Start by matching the primary movement. Add secondary actions only after the motion transfer is stable.
How to Fix Weak Reference-Video Results
When a generation fails, identify the specific problem before rewriting the entire prompt.
Problem | Likely cause | Recommended fix |
|---|---|---|
Motion does not match | Reference role is unclear | State exactly which movement should be followed |
Character identity changes | Visual identity is underdefined | Add clear face and full-body reference images |
Original background appears | The source scene was not excluded | Explicitly reject the reference background |
Camera remains static | Camera instructions are buried | Put the camera direction in a separate sentence |
Actions happen too quickly | Timing is undefined | Add second-by-second phases |
Extra objects appear | The source video contains clutter | Crop or simplify the reference |
Product shape changes | Identity and motion instructions conflict | Add multiple product views and preservation rules |
Two characters merge | Positions and identities are unclear | Use separate identity images and clearer blocking |
The best solution is not always a longer prompt. In many cases, the result improves when each input receives a narrower and more explicit role.
Best Use Cases for Seedance 2.5 Reference Video
Product Advertising
A simple physical mockup or phone recording can define:
Product rotation
Camera movement
Hand interaction
Reveal timing
Final framing
A product image controls appearance while the reference video controls motion.
Character Animation
A performer can record body movement without appearing in the final result. A separate character image defines the generated identity.
This is useful for:
Digital characters
Fantasy heroes
Mascots
Historical scenes
Stylized avatars
Consistent campaign characters
Fashion and Dance
Reference clips can guide walking pace, poses, turns, choreography, and performance rhythm.
Separate clothing and character images help prevent the model from copying the performer’s appearance.
Film Previsualization
Directors can record a phone video or export a simple 3D blockout to communicate:
Camera path
Character placement
Action order
Shot duration
Final composition
The reference becomes a low-cost visual plan for the generated scene.
Social Media Content
The same motion template can be reused across different characters, products, settings, and visual styles.
This helps creators produce multiple variations while preserving a successful pacing structure.
Frequently Asked Questions
Can Seedance 2.5 use a video as a reference?
Seedance-style multimodal workflows can use video inputs to guide movement, camera behavior, timing, and scene structure. Exact Seedance 2.5 controls, supported formats, and upload limits may vary depending on the platform and rollout stage.
Does the reference video determine the final character?
Not necessarily. A reference video can control only the movement while a separate image defines the final character. The prompt should clearly state that the actor’s face, clothing, body shape, and background must not be copied.
Can a reference video control camera movement?
Yes. A reference clip can communicate camera direction, speed, framing, and timing. State that only the camera path should be followed when the original subject and location should not appear in the result.
Can multiple reference videos be used?
Where the selected platform supports multiple video inputs, each clip should be assigned a separate role. One video may guide body movement while another controls the camera. Avoid combining clips with contradictory timing or motion.
Does a reference video need to be high resolution?
Motion readability is usually more important than cinematic image quality. The subject, action, and camera movement should be clearly visible and easy to separate from the background.
Can a phone recording be used?
Yes. A simple phone recording can work well as a movement, camera, or blocking reference when the important action remains visible and the background is not overly distracting.
Why does the generated video copy the original background?
The prompt may not clearly separate motion from appearance. Add an instruction that uses only the body movement and action timing from @Video1 and does not copy the person, clothing, objects, colors, lighting, or background.
Final Thoughts
A Seedance 2.5 reference video is most effective when it has one clearly defined responsibility.
It can tell the model:
How a person moves
How a camera travels
When an action happens
Where characters stand
How a product is revealed
How quickly a scene develops
Separate images can then control the character, product, clothing, environment, and visual style.
The strongest workflow is not to upload as many references as possible and hope the model understands them. It is to assign every input a specific job:
Text defines the creative objective.
Images define visual identity.
Video defines movement and timing.
Audio defines rhythm and sound.
Preservation instructions define what cannot change.
Start with one short, readable reference clip and one clear prompt. Once the movement works, add character images, product references, styling, lighting, and audio in controlled stages.
Ready to test a reference-based workflow?