Seedance 2.0 Basics: What It Is, What It Supports, and How to Think About It
Seedance 2.0 is a multimodal AI video generation model from ByteDance designed to create short, high-quality videos from text, images, video, and audio references. Instead of treating a prompt as a simple text command, Seedance 2.0 lets you combine multiple input types and direct the model toward a specific creative result.
That means you can use an image to lock a character or product look, a video to guide movement or camera choreography, audio to drive music timing or lip-sync, and text to explain the scene, action, style, mood, and output intent.
The most important shift is this: Seedance 2.0 works best when you stop prompting vaguely and start directing clearly.
A weak prompt says:
Make a cool cinematic video.
A strong Seedance 2.0 prompt says:
Use @Image1 for the product design. The product floats and rotates slowly against a clean white studio background. Soft gradient lighting highlights the metallic texture. Smooth orbit camera, macro commercial style. Text appears at bottom center: “Available Now” in elegant white serif type. Calm ambient music, premium mood.
The difference is not just length. The stronger version tells the model what each reference is for, what motion should happen, how the camera should behave, what the scene should feel like, and how the final video should be structured.
Seedance 2.0 rewards specificity.
What Seedance 2.0 is best at
Seedance 2.0 is built for short-form AI video generation, especially when you need stronger control over subject consistency, motion, camera direction, audio, and reference-based generation.
It is especially useful for:
- Cinematic portraits
- Product showcases
- E-commerce ads
- Character animation
- Music-driven edits
- Dance or action choreography transfer
- Stylized social videos
- Text-on-video ads
- Storyboard-to-video generation
- Video extension
- Natural language video editing
- Multi-shot short sequences
- Lip-sync dialogue
- Camera movement replication
- VFX and transition reference matching
The model can generate clips from text alone, but its strongest workflows come from combining references. If you want a product to stay recognizable, upload product images. If you want a movement copied, upload a motion reference video. If you want cuts synced to music, upload the audio. If you want a character to speak, provide an image for appearance and audio for the dialogue.
Think of Seedance 2.0 as a short-form video director that can read your creative brief, inspect your references, and generate the shot.
Core inputs and outputs
Seedance 2.0 supports four main input types: text, images, videos, and audio.
Text prompts describe the scene, subject, action, camera, timing, mood, style, and sound design. Images can be used for characters, products, logos, storyboards, environments, or visual style. Videos can be used for motion reference, camera movement, effects, editing, or continuity. Audio can be used for music, rhythm, narration, dialogue, or lip-sync.
The system supports up to 12 combined files per generation. A practical breakdown is:
- Up to 9 image files
- Up to 3 video files
- Up to 3 audio files
- 12 total files combined
Image formats include common file types such as JPEG, PNG, WEBP, BMP, TIFF, and GIF. Video inputs are typically MP4 or MOV. Audio inputs can include MP3 and WAV.
Outputs are short videos, usually between 4 and 15 seconds. This duration range is important because Seedance 2.0 works best when the creative idea fits the available time. A 4-second generation should usually focus on one clear action beat. A 10–15 second generation can support more complex timing, transitions, or multiple shots.
The most important concept: reference control
Seedance 2.0 uses an @ reference system to bind uploaded files to prompt instructions.
For example:
Use @Image1 for the character’s appearance.
Reference @Video1 for the camera movement.
Use @Audio1 for the background music.
Sync visual cuts to the rhythm of @Audio1.
This is one of the most important parts of prompting Seedance 2.0. Uploading a file is not enough. You need to tell the model what the file is and how it should be used.
A vague reference sounds like this:
Use the uploaded image.
A clear reference sounds like this:
Use @Image1 for the product’s exact shape, material, and logo placement. Keep the product visually consistent throughout the entire video.
Or:
Use @Video1 only for the camera movement and pacing. Do not copy the subject or background from @Video1.
This level of instruction prevents confusion. Without it, the model may misunderstand whether a file is meant to guide subject identity, scene composition, motion, style, camera movement, text, or timing.
Important restriction: realistic human faces
One critical limitation: Seedance 2.0 may block images or videos with clearly identifiable realistic human faces. This is enforced at the platform level.
Safer alternatives include:
- Illustrated character references
- Stylized human characters
- Animated or cartoon designs
- Abstract human representations
- Silhouettes
- Distant figures
- Non-photorealistic portraits
If a prompt or reference fails because of face restrictions, redesign the subject as stylized rather than photorealistic. For example, use “animated character,” “illustrated fashion model,” “stylized cinematic portrait,” or “distant silhouette” instead of a realistic identifiable person.
A helpful reminder
Seedance 2.0 is not just a text-to-video model. It is a reference-driven video generation system. The more clearly you assign each input, the more control you have over the output.
Use text to direct. Use images to preserve appearance. Use videos to guide movement, camera, effects, and continuity. Use audio to drive rhythm, mood, dialogue, and lip-sync.
Key takeaway
Before writing advanced prompts, understand the basic system: Seedance 2.0 works best when you provide clear direction and clearly assign every reference. The foundation is simple: know what you want the model to make, know which inputs matter, and tell the model exactly how to use them.
How to Prompt Seedance 2.0: Structure, References, Motion, Camera, Audio, and Text
Good Seedance 2.0 prompting is less about writing a long prompt and more about writing a clear creative brief.
A useful prompt tells the model what the subject is, what should happen, where it happens, how the camera moves, what the lighting feels like, what style to use, and how any uploaded references should shape the result.
The goal is not to describe everything. The goal is to describe the right things.
The universal prompt formula
A reliable Seedance 2.0 prompt usually includes six ingredients:
[Subject with specific detail]
+ [One concrete action beat]
+ [Environment or setting]
+ [One camera movement with framing]
+ [Lighting source and mood]
+ [Style or aesthetic]
A more advanced version can add timing, transitions, text overlays, audio, and reference assignments:
[Subject/Character]
+ [Scene/Environment]
+ [Action/Motion]
+ [Camera Movement]
+ [Timing]
+ [Transitions/Effects]
+ [Audio/Sound Design]
+ [Style/Mood]
+ [Reference Instructions]
You do not need every element every time. But you should always include the subject and motion. Those are the minimum ingredients for a useful video prompt.
A simple but strong prompt could be:
A luxury watch rotates slowly on a black marble surface.
Light rays catch the polished steel bezel, creating sparkle reflections.
Extreme close-up macro shot with a smooth 360-degree orbit.
Soft studio lighting with a dark gradient background.
High-end commercial style, crisp details, premium mood.
This works because it gives the model a clear subject, action, surface, lighting, camera path, style, and mood.
How to write better subjects
The subject is the anchor of the video. It tells the model what the viewer should focus on.
Weak subjects are generic:
A woman.
A car.
A product.
A city.
Strong subjects are specific:
A young woman with flowing auburn hair wearing a red silk dress.A vintage blue motorcycle with chrome details and worn leather seats.A luxury skincare bottle with frosted glass and a gold pump.A neon-lit city street at night with wet pavement and steam rising from vents.
Specific subjects give Seedance 2.0 more visual information to work with. The more concrete the subject, the easier it is for the model to maintain consistency.
For products, mention material, shape, logo position, color, surface texture, and design details. For characters, mention outfit, silhouette, hairstyle, posture, and visual style. For environments, mention location, weather, lighting, atmosphere, and time of day.
How to write stronger motion
Seedance 2.0 is a video model, so motion matters. A beautiful subject with no action can produce a static or underwhelming clip.
Good motion is concrete and visible.
Instead of:
The character moves.
Write:
The character slowly raises their gaze toward the camera while wind moves strands of hair.
Instead of:
The product is shown.
Write:
The product floats upward, rotates 180 degrees, and settles into center frame as light glints across the logo.
Instead of:
The city looks cinematic.
Write:
Rain falls onto the pavement as the camera glides forward through neon reflections and passing headlights.
Motion should usually focus on one main action beat. Too many actions in a short clip can make the generation feel rushed or incoherent.
Match complexity to duration
One of the most common mistakes in Seedance 2.0 prompting is asking for too much in too little time.
A 4-second clip should usually contain one subject, one action, one camera movement, and one mood.
Example:
A ceramic perfume bottle sits on a reflective black surface.
The bottle slowly rotates as soft light sweeps across the glass.
Macro close-up, smooth orbit camera, luxury commercial style.
A 10–15 second clip can support a more structured sequence:
0-3s: Wide shot of a neon-lit city street at night, rain reflecting signs on the pavement.
3-6s: Medium shot of a detective stepping out of a car in a trench coat.
6-10s: Close-up of their eyes scanning the street, rain dripping from the hat brim.
10-15s: Over-shoulder shot moving toward a flickering bar sign.
Noir style, high contrast, moody blue-orange palette.
The rule is simple: shorter clips need simpler ideas. Longer clips can handle more stages.
Camera vocabulary Seedance 2.0 understands
Camera instructions are one of the best ways to make Seedance 2.0 outputs feel intentional. Instead of saying “cinematic camera,” use specific camera language.
Useful terms include:
- Slow dolly-in: camera moves toward the subject
- Dolly-out: camera pulls away
- Orbit shot: camera circles the subject
- Tracking shot: camera follows the subject laterally
- Push-in: camera moves forward into the scene
- Pull-back reveal: camera pulls wider to reveal context
- Crane up: camera rises vertically
- Low-angle tilt up: camera looks upward from below
- Handheld: slight shake, documentary feel
- Steadicam: smooth continuous motion
- Whip pan: fast horizontal snap
- Hitchcock zoom: dolly out plus zoom in
- First-person POV: camera sees through a character’s eyes
- Rack focus: focus shifts from foreground to background
Camera movement should not conflict with itself. Avoid asking for a fast zoom-in and a slow dolly-out at the same time unless you are intentionally describing a Hitchcock zoom. Keep the camera direction coherent.
A strong camera line looks like:
Slow dolly-in from medium shot to close-up, shallow depth of field, subject centered.
Or:
Smooth orbit camera around the product, starting from a low macro angle and ending in a centered hero frame.
Working with image references
Image references are useful when you need the model to preserve identity, design, composition, or style.
You can use images for:
- Character references
- Product references
- Logo references
- Multi-subject references
- Storyboard frames
- Style boards
- First-frame control
- Multi-angle subject understanding
For character or product consistency, provide clean, well-lit images. Multiple angles are helpful because they give the model a better sense of the subject’s full form.
Example:
Use @Image1 for the product’s front view.
Use @Image2 for the side profile.
Use @Image3 for the logo and material detail.
Generate a premium product video where the product rotates slowly on a reflective black surface.
Keep the product shape, logo placement, and material consistent across the full clip.
For multiple characters:
@Image1 is Character A.
@Image2 is Character B.
Character A hands a glowing object to Character B in a quiet forest clearing.
Medium two-shot, slow push-in, moonlit atmosphere, fantasy film style.
Maintain both characters’ appearances clearly and separately.
For storyboard references:
Use @Image1 through @Image4 as storyboard frames in sequence.
Generate a continuous video transitioning through each frame.
Maintain character consistency, lighting continuity, and cinematic pacing.
The key is to label each image and assign its purpose.
Working with video references
Video references are powerful because they can guide motion, camera movement, effects, and continuity.
You can use video references for:
- Action choreography
- Dance movement
- Athletic motion
- Gestural performance
- Product interaction
- Camera path replication
- Drone movement
- Transition effects
- VFX style
- Video extension
- Editing an existing clip
If you want the model to copy movement but not copy the subject, say that clearly.
Example:
Use @Video1 as the motion reference only.
Apply the same dance choreography and rhythm to the character from @Image1.
Do not copy the original dancer’s appearance, clothing, or background.
Urban street background with neon lights.
Medium-wide tracking shot, high-energy music video style.
If you want the camera movement copied:
Use @Video1 as the camera movement reference.
Use @Image1 as the new subject.
Keep the camera movement, pacing, and framing from @Video1, but replace the subject with the product from @Image1.
Cinematic lighting, natural motion, premium commercial style.
If you want effects copied:
Reference @Video1 for the particle effect and transition style.
Apply the same glowing dust transformation to the character from @Image1.
Keep the background as a dark studio environment with dramatic rim lighting.
The more specific you are about what to copy and what not to copy, the better the result.
Working with audio references
Audio can shape timing, rhythm, dialogue, lip-sync, and mood.
Use audio references for:
- Background music
- Beat-synced edits
- Voiceover
- Dialogue
- Lip-sync
- Sound design
- Mood and pacing
For music-driven videos:
Use @Audio1 for the music track.
Sync visual cuts to the beat of @Audio1.
Each downbeat triggers a new camera angle.
High-energy montage of urban dancers, fast cuts on downbeats, slow motion during the bridge.
For lip-sync dialogue:
Use @Image1 as the character’s appearance.
Use @Audio1 for the dialogue track.
The character speaks directly to camera with lips synced to @Audio1.
Medium close-up, soft indoor lighting, natural conversational tone.
Subtle head movements and realistic facial expression.
Audio instructions are often overlooked, but Seedance 2.0 can generate richer outputs when you describe sound effects, music mood, narration, or dialogue behavior.
Text in video
Seedance 2.0 can render text directly inside video. This is useful for ads, social content, product videos, educational clips, and dialogue-driven videos.
There are three main text types:
Title or slogan text
Use this for hero messaging, campaign lines, and ad copy.
Example:
Text appears at center: “DREAM BIGGER” in bold white sans-serif.
It fades in at 2 seconds and holds for 3 seconds.
Subtitles
Use subtitles for narration or dialogue.
Example:
Subtitles appear at bottom of screen, synced with narration in @Audio1.
White text with black outline for readability.
Speech bubbles
Use speech bubbles for stylized character dialogue.
Example:
The character says “Hello!” with a speech bubble near their mouth.
Cartoon style, rounded white bubble, playful mood.
For best results, specify what the text says, where it appears, when it appears, how long it stays, and what it should look like.
A helpful reminder
A Seedance 2.0 prompt is strongest when every phrase has a clear purpose. Use the subject to define the focus, motion to define the action, camera language to define the viewing experience, references to preserve or transfer specific details, and audio or text instructions to shape the final presentation.
Seedance 2.0 prompting works best when you write like a director. Be specific about the subject, concrete about the motion, intentional with the camera, explicit with references, and realistic about what can fit inside the selected duration.
Advanced Seedance 2.0 Workflows: Editing, Extensions, Multi-Shot Videos, Templates, and Common Mistakes
Once you understand the basics of Seedance 2.0 prompting, you can use it for more advanced production workflows: editing existing footage, extending clips, connecting separate videos, creating one-take camera moves, generating multi-shot sequences, transferring choreography, building music videos, and using reusable templates.
The key is to stay focused. Advanced workflows work best when the prompt is structured, references are clearly assigned, and each generation has one primary goal.
Natural language video editing
Seedance 2.0 can edit existing videos. You can add, remove, or modify elements by describing the change in natural language and supplying the original video as a reference.
Examples include:
Replace the red car in @Video1 with a vintage blue motorcycle.
Keep the original camera angle, lighting, background, and motion path.
Remove the plastic bottle from the table in @Video1.
Keep the rest of the scene unchanged, including camera movement, lighting, and character motion.
Add glowing fireflies around the character in @Video1.
The fireflies should move naturally through the scene and match the existing lighting.
Editing works best when each generation focuses on one major change. Asking for many simultaneous edits can reduce control and consistency.
Video extension and track completion
Seedance 2.0 can extend videos forward or backward while preserving continuity.
A good extension prompt includes:
- Which video to extend
- How many seconds to add
- What should happen next
- What should stay consistent
- Camera continuity
- Lighting continuity
- Character or product continuity
- Environment continuity
Example:
Extend @Video1 by 5 seconds.
Continue the same camera movement and lighting.
The character keeps walking forward, reaches a doorway, and pauses.
Maintain the same color grade, outfit, environment, and motion physics.
Seedance 2.0 can also connect multiple short videos into a continuous sequence. This is useful when you have separate clips that need smooth transitions and consistent visual style.
Example:
Use @Video1, @Video2, and @Video3 as sequence references.
Connect them into one continuous 15-second video.
Maintain consistent lighting, color grade, camera energy, and character appearance.
Generate smooth transitions between the clips.
One-take continuous shots
One-take prompts are useful when you want a continuous camera move with no cuts.
Example:
Reference all transitions and camera movements from @Video1 as a one-take continuous shot.
Start in a busy kitchen, camera pushes through a doorway into a dining room, then continues past a window to reveal a garden.
No cuts. Smooth steadicam feel. Natural lighting transitions between rooms.
For one-take shots, avoid overloading the scene. The camera path should be clear, and the sequence should unfold naturally.
Multi-shot sequences
For 10–15 second generations, you can structure a prompt by timecode.
This gives Seedance 2.0 a clear timeline:
0-3s: Wide establishing shot of a futuristic train station at sunrise.
3-6s: Medium shot of a traveler stepping onto the platform, coat moving in the wind.
6-10s: Close-up of a glowing ticket in their hand.
10-15s: Pull-back reveal as the train arrives through mist.
Cinematic sci-fi style, soft orange-blue lighting, atmospheric sound design.
Timecoded prompts are useful when you need multiple beats, but every beat should still be simple enough to fit the available duration.
Prompt templates by use case
Cinematic portrait
A young woman with flowing auburn hair slowly raises her gaze toward the camera.
Soft golden hour light casts warm shadows across her face.
A gentle breeze moves strands of hair.
Slow dolly-in from medium shot to close-up.
Cinematic film grain, shallow depth of field, warm amber tones.
Product showcase
Use @Image1 for the product.
The product floats and rotates slowly against a clean white background.
Soft studio lighting highlights its texture and material.
Smooth orbit camera around the product.
Minimal premium aesthetic.
Text appears: “Available Now” in elegant serif font at bottom of frame.
Dance choreography transfer
Use @Image1 for the character’s appearance.
Reference @Video1 for the dance choreography, rhythm, and energy.
The character performs the same dance moves in an urban street environment with neon lights.
Medium-wide tracking shot, high-energy music video style.
Sync movement to the rhythm of @Audio1.
Lip-sync character video
Use @Image1 as the character’s appearance.
Use @Audio1 as the dialogue track.
The character speaks directly to camera with lips synced to @Audio1.
Medium close-up, soft indoor lighting, shallow depth of field.
Natural conversational tone, subtle head movement, expressive eyes.
Subtitles appear at bottom of screen synced with the dialogue.
Storyboard-to-video
Use @Image1 through @Image4 as storyboard frames in sequence.
Generate a continuous 12-second video transitioning through each frame.
Maintain the same character appearance, lighting direction, and cinematic style.
Smooth transitions between shots, emotional pacing, atmospheric sound design.
Video extension
Extend @Video1 by 5 seconds.
Continue the same camera movement, lighting, outfit, color grade, and environment.
The character walks forward, reaches the doorway, pauses, and turns slightly toward the light.
Maintain visual continuity and natural motion.
Element swap
Replace the red car in @Video1 with a vintage blue motorcycle.
Keep the original camera angle, lighting, background, and motion path.
The motorcycle should match the same speed and position as the original car.
Beat-synced music video
Use @Audio1 for the music track.
Sync visual cuts to the beat of @Audio1.
Each downbeat triggers a new angle or scene.
High-energy montage of urban dancers.
Fast cuts on downbeats, slow motion during bridges.
Neon lighting, handheld camera energy, music video style.
Common mistakes to avoid
Mistake 1: vague references
Do not upload media and assume the model knows what to do with it.
Weak:
Use this image.
Strong:
Use @Image1 for the character’s appearance, outfit, hairstyle, and color palette.
Keep the character consistent throughout the video.
Mistake 2: missing @ assignments
Every uploaded file should be referenced clearly. If you upload three files, assign all three.
@Image1 is the product.
@Image2 is the logo.
@Audio1 is the music track.
Mistake 3: conflicting camera instructions
Avoid contradictory directions.
Weak:
Fast zoom in, slow dolly out, static camera, handheld orbit shot.
Strong:
Slow dolly-in from medium shot to close-up, smooth stabilized movement.
Mistake 4: overloading short clips
Do not force a full story into four seconds. A short clip should focus on one strong visual beat.
Mistake 5: ignoring sound design
If the video includes movement, atmosphere, music, dialogue, or impact, add audio direction.
Example:
Soft ambient music, faint city traffic, gentle rain sounds, subtle whoosh as the camera pushes in.
Mistake 6: duration mismatch
If your prompt describes 10 seconds of action, do not generate a 4-second clip. Either simplify the idea or choose a longer duration.
A practical workflow for better results
A strong Seedance 2.0 workflow usually follows this order:
- Decide the exact output type: portrait, ad, music video, lip-sync, product showcase, edit, extension, or storyboard.
- Choose the right duration for the idea.
- Upload only the references that matter.
- Assign each reference with
@syntax. - Write one clear subject.
- Write one concrete action beat.
- Add camera movement and framing.
- Add lighting, style, and mood.
- Add audio or text instructions if relevant.
- Remove contradictions before generating.
- If the result misses the mark, revise one variable at a time.
The best prompts are not necessarily the longest. They are the clearest.
A helpful reminder
Advanced Seedance 2.0 workflows work best when you give the model a focused job. Use natural language to edit, extend, connect, or structure footage, but avoid asking for too many changes at once.
When prompts become more complex, clarity matters even more. Assign every reference, match the idea to the duration, keep camera and motion instructions coherent, and use templates as starting points rather than rigid formulas.
Seedance 2.0 can support production-style video workflows, but the best results come from focused direction. Use clear references, one primary goal per generation, realistic timing, and structured prompts to guide editing, extension, one-take shots, multi-shot sequences, and template-based generation.