Seedance 2.0 is what happens when AI video stops being a slot machine.
Old models were: type a prompt, cross your fingers, accept whatever weird hands you get.
Seedance 2.0 is: direct and produce - with references to images, videos, and audio so the model can copy what matters (motion, camera, style, voice) and invent what doesn’t.
And if you’re here for performance marketing: this is the same mindset that makes UGC actually scale. Human UGC is great - until you need 50 variations by Friday. Traditional UGC is about ~$200/video hiring creators. EzUGC-style AI UGC is about ~$5/video, with consistency you can actually control, in 32+ languages, in minutes.
This guide covers Seedance 2.0 end-to-end: the multi-modal @ reference system, prompt techniques, and real workflows (with lots of examples).
Table of Contents
- The Multi-Modal @ Reference System
- Ultra-Realistic Text-to-Video
- Consistency Across Every Detail
- Camera & Motion Replication
- Creative Template & Effects Replication
- AI Creativity & Story Completion
- Video Extension & Continuation
- Audio, Voice & Lip Sync
- One-Take Continuity
- Video Editing
- Beat-Synced Editing
- Emotion Performance
- Prompt Writing Fundamentals
- Camera & Cinematography Language
- Character Consistency Tips
- Video Quality Control
- Multi-Shot Storytelling
- Real-World Use Cases
- Pro Tips
- What to Avoid
- Ready-to-Use Prompt Templates
The Multi-Modal @ Reference System
The @ system is the core of Seedance 2.0.
It’s how you tell the model what each uploaded file is for - and it supports text, images, videos, and audio.
Here’s the key shift: you’re not just describing a scene. You’re assigning roles.
When you upload assets, Seedance labels them automatically: @Image1, @Image2, @Video1, @Audio1, etc. Then you reference them in the prompt.
Example prompt:
@Image1 as the first frame, reference @Video1 for camera movement, use @Audio1 for background music.
How It Works
- Upload your assets - images, videos, or audio files
- They get labeled - @Image1, @Image2, @Video1, @Audio1, etc.
- Reference them in your prompt - tell the model exactly how to use each file
- Generate - Seedance combines everything according to your direction
Special Usage Patterns
These patterns are where Seedance stops feeling like “AI video” and starts feeling like “a controllable tool.”
First/last frame + video motion reference
@Image1 as the first frame, reference @Video1's fighting choreography.
Extend an existing video
Extend @Video1 by 5 seconds.
Note: Set your generation duration to match the new portion only (e.g., if extending by 5s, set generation length to 5s).
Merge multiple videos
I want to add a scene between @Video1 and @Video2. The content is xxx.
Reference audio from a video (no separate audio file needed)
Reference @Video1's background music and sound effects.
Generate continuous actions
The character transitions from a jump directly into a roll, maintaining fluid and coherent motion. @Image1 @Image2 @Image3...
Important: When using multiple assets, double-check that each @ reference is correctly labeled. Don’t mix up images, videos, and characters. Also: clearly specify whether something is a reference or being edited.
Ultra-Realistic Text-to-Video
Even without references, Seedance 2.0 can do realistic text-to-video.
The difference is it handles multi-step actions described in normal language without falling apart halfway through.
Example Prompts
Everyday realism - laundry scene
A girl elegantly hanging clothes to dry. After hanging one piece, she reaches into the bucket for another, gives it a vigorous shake, and hangs it up.
Painting comes alive - commercial concept
The person in the painting looks nervous, eyes darting left and right, then peeks out from the frame. They quickly reach out of the painting to grab a cola, take a sip, and show a satisfied expression. Footsteps approach - the person in the painting hurriedly puts the cola back. A cowboy walks up and takes the cola, then leaves. The camera pushes in as the screen gradually fades to black with only a top light illuminating a can of cola. Artistic subtitles appear at the bottom: "Yikou Cola - a taste you can't miss!"
Period drama - 19th century London
Camera slowly pulls out (revealing the full street) and follows the female lead. Wind blows her skirt as she walks along a 19th-century London street. A steam-powered car speeds past from the right side, its wind lifting her skirt - she gasps in shock and quickly presses her skirt down with both hands. Background audio: footsteps, crowd chatter, vehicle sounds.
Action sequence - chase scene
Camera follows a man in black sprinting in a desperate escape, a crowd chasing behind. Camera shifts to a side tracking shot. He panics, crashes into a roadside fruit stand, scrambles up, and keeps running. Sounds of a chaotic crowd.
Consistency Across Every Detail
This is the part that matters if you’re doing anything commercial.
AI video used to be allergic to consistency: faces morph, logos drift, text turns into hieroglyphics, the scene changes style mid-clip.
Seedance 2.0 aims to lock it down: face identity, clothing, logos, text details, environment, and visual style.
What Stays Consistent
- Face identity throughout the video
- Product details - logos, text, fine elements
- Environment across shots
- Visual style - no random drift mid-clip
- Clothing and accessories
Example Prompts
Multi-scene family reunion - character consistency across complex narrative
The man (@Image1) tiredly walks down a hallway after work, his steps slowing. He stops at the front door. Close-up on his face - he takes a deep breath, adjusts his emotions, releases the negativity, and relaxes. Close-up of him finding his keys, inserting into the lock. After entering, his young daughter and a pet dog joyfully run over to greet him with hugs. The interior is warm and cozy. Natural dialogue throughout.
Style transfer - maintaining character through transformation
Replace the girl in @Video1 with a Chinese opera huadan (花旦). The scene is an ornate stage. Reference @Video1's camera work and transitions. Use the camera to match the character's movements. Ultimate stage aesthetics with enhanced visual impact.
One-take with consistent character across scene changes
Reference all transitions and camera work from @Video1. One continuous take. The scene opens on a chess game, camera pans left revealing yellow sand on the floor, camera tilts up to a beach with footprints. A girl in white walks into the distance on the beach. Camera cuts to an aerial overhead view - ocean waves washing the shore (no people visible). Seamless dissolve transition - the washing waves transform into flowing curtains. Camera pulls back to reveal the girl's face in close-up. One continuous take throughout.
Product commercial - brand consistency
Perform a commercial-grade showcase of the handbag in @Image2. The side view references @Image1. The surface material references @Image3. Show all details of the handbag. Background audio should be grand and majestic.
Multi-language commercial with brand elements
0–2s: Rapid four-frame flash cuts - red, pink, purple, and leopard-print bows displayed in sequence. Close-ups of satin sheen and the "chéri" brand text.
3–6s: Close-up of silver magnetic clasp clicking shut, then gently pulled apart - showcasing silky texture and convenience.
7–12s: Quick cuts of wearing scenarios - wine-red bow on a coat collar for commuter style; pink bow tying a ponytail for sweet street look; purple bow on a bag strap for understated luxury; leopard-print bow on a suit collar for bold style.
13–15s: All four bows displayed side by side with brand name.
First-person POV with multi-scene references
Set @Image1 as the first frame. First-person perspective. Reference @Video1 for camera movement. Upper scene references @Image2, left scene references @Image3, right scene references @Image4.
Consistency is the difference between “cool demo” and “something you can spend money on.”
And yes - this is exactly why performance teams like AI UGC. When you’re testing hooks and angles, you don’t want the product label changing every render.
Camera & Motion Replication
This is the “stop prompt-engineering like it’s 2022” feature.
Instead of describing a dolly move in poetry, you upload a reference video. Seedance extracts the camera language and motion.
Seedance 2.0 can replicate:
- Choreography - fight scenes, dance moves, action sequences
- Camera techniques - dolly shots, tracking, crane, handheld, Hitchcock zoom
- Editing rhythm - cut timing, transitions, pacing
- Specific moves - whip pans, orbit shots, mechanical arm tracking
Example Prompts
Hitchcock zoom + mechanical arm tracking
Reference @Image1 for the man's appearance. He's in @Image2's elevator. Fully replicate @Video1's camera movements and the protagonist's facial expressions. Hitchcock zoom when startled, then several orbit shots showcasing the elevator interior. Elevator doors open - tracking shot follows him out. The exterior scene references @Image3. The man looks around. Reference @Video1's mechanical-arm multi-angle tracking following the character's line of sight.
Complex chase sequence with multiple camera techniques
Reference @Image1 for the man's appearance. He's in @Image2's corridor. Fully replicate @Video1's camera work and the protagonist's facial expressions. Camera follows the protagonist sprinting around a corner in @Image2, then in @Image3's long hallway - camera tracks from behind, low angle, orbiting around to the front. Camera pans right 90° to capture @Image4's forked intersection, hard stop, then pans 180° for a close-up of the protagonist's face: gasping for breath. Camera follows the protagonist's POV looking around - reference @Video1's rapid left-right orbital camera work to showcase the scene. Pull back to @Image5's setting, continue side-angle tracking of the protagonist running.
Product showcase with cinematic camera
The tablet from @Image1 as the main subject. Camera work references @Video1 - push in to a screen close-up, camera rotates as the tablet flips to reveal its full form. Data streams on screen constantly change. The surrounding environment gradually transforms into a sci-fi data space.
Dance with synchronized camera rhythm
The female star from @Image1 as the main subject. Reference @Video1's camera techniques for rhythmic push-pull-pan movements. The star's movements also reference the dance choreography from @Video1's dancer - performing energetically on stage.
Multi-character fight choreography
Reference @Image1 and @Image2 for the spear-wielding character. @Image3 and @Image4 for the dual-blade character. Replicate @Video1's choreography. They fight in @Image5's maple leaf forest.
Fight scene with environment and camera fusion
Reference @Video1 for character actions. Reference @Video2 for orbiting camera language. Generate a fight scene between Character 1 and Character 2. The fight takes place under a starry sky. White dust rises during combat. The fighting is spectacular, the atmosphere intensely tense.
Commercial car showcase
Reference @Video1's camera work, frame transitions, and rhythm. Replicate using the red supercar from @Image1.
Creative Template & Effects Replication
Find a video style you like. Feed it in. Generate your own content in that exact structure.
You don’t need film-school vocabulary. You just need to be explicit about what you’re copying: rhythm, camera work, effects, transitions, style.
Example Prompts
VR/Sci-fi scene transitions with template video
Replace @Video1's subject with @Image1. @Image1 as the first frame. The character puts on a virtual sci-fi headset. Reference @Video1's camera work. From third-person view to the character's subjective POV. Through the AI headset, arrive at @Image2's deep blue universe. Several spaceships appear, flying into the distance. Camera follows the ships, transitioning to @Image3's pixel world. Camera flies low over pixel mountains with trees growing in stylized formations. Then the view tilts up, accelerating through to @Image4's green-textured planet, camera gliding over the planet's surface.
Fashion montage with fish-eye effect
Reference the first image for the model's facial features. The model wears outfits from images 2–6, approaching the camera with playful, cool, cute, surprised, and suave poses respectively. Each outfit change triggers a cut. Reference the video's fish-eye lens effect and ghosting/flickering visual effects.
Advertisement recreation
Reference the video's ad concept. Using the provided down jacket image, reference the goose feather image and swan image. Pair with this ad copy: "This is goose down. This is a warm swan. This is a wearable arctic swan-down jacket. Dress warm for the new year, live warm every day." Generate a new down jacket advertisement.
Ink-wash style with effects transfer
Black-and-white ink-wash style. The character from @Image1 references @Video1's effects and movements - performing an ink-wash tai chi martial arts sequence.
Particle and texture effects transfer
Starting from black. Reference @Video1's particle effects and materials. Gold-gilded sand drifts in from the left side of the frame, covering rightward. Reference @Video1's particle dispersion effect. The text from @Image1 gradually appears at the center of the frame.
Character effects mashup
The character from @Image1 references @Video1's actions and expression changes - demonstrating the absurd act of eating instant noodles.
Puzzle-shatter transition recreation
Starting from @Image1's ceiling. Reference @Video1's puzzle-shattering effect for the transition. Replace the "BELIEVE" text with "Seedance." Reference @Image2's font style.
AI Creativity & Story Completion
Seedance 2.0 isn’t just obedient.
It can also fill gaps: add drama, pacing, and connective tissue when you give it a premise instead of a screenplay.
Example Prompts
Comic panel to animated story
Animate @Image1 as a comic strip, reading left-to-right, top-to-bottom. Keep the characters' dialogue consistent with what's written in the image. Add special sound effects for scene transitions and key story moments. The overall style should be humorous and playful. Reference @Video1 for the animation style.
Storyboard to cinematic intro
Reference @Image1's storyboard - follow its shot breakdowns, framing, camera movements, visuals, and copy. Create a 15-second healing-style opening for a piece about "Childhood Through the Four Seasons."
Mood-driven video from image inspiration
Using @Image1 through @Image5 as inspiration, create an emotionally-driven video. Background music references @Video1's audio.
Video Extension & Continuation
This is “keep filming” instead of “generate another unrelated clip.”
You can extend forward or backward with continuity. Set your generation duration to match the new content you want.
Example Prompts
Wild commercial extension (15s)
Extend the video by 15 seconds. Reference @Image1 and @Image2's donkey-on-motorcycle character. Add a wild advertisement sequence:
Scene 1: Side fixed shot - donkey bursts out of a fence on a motorcycle, nearby chickens are startled.
Scene 2: Donkey does donuts on sand. Close-up of motorcycle tire, then aerial overhead shot of the donkey performing spinning stunts, kicking up dust.
Scene 3: Snowy mountain backdrop - donkey flies over a hillside on the motorcycle. Ad text "Inspire Creativity, Enrich Life" appears behind the subject through a masking effect as the motorcycle flies past, ending with a dust trail.
Fitness ad extension (6s)
Extend the video by 6 seconds. An intense electric guitar riff kicks in. "JUST DO IT" text appears center-screen then gradually fades. Camera tilts up to the ceiling - a muscular man pulls himself up on gymnastic rings, wearing @Image1's tight-fitting workout gear with @Image2's "Fitness" logo on the back. He powers through the pull-up, then "DO SOME SPORT" text appears as the closing ad frame.
Coffee commercial extension (15s)
Extend @Video1 by 15 seconds.
1–5s: Light and shadow slowly slide across a wooden table and cup through venetian blinds, with branches gently swaying.
6–10s: A single coffee bean gently drifts down from above. Camera pushes into the bean until the screen goes black.
11–15s: Text fades in - first line: "Lucky Coffee", second line: "Breakfast", third line: "AM 7:00–10:00".
Backward extension - adding a prequel (10s)
Extend backward by 10 seconds. In warm afternoon light, the camera starts from awnings fluttering in the breeze at a street corner, slowly tilting down to small daisies poking out at the base of a wall. The protagonist's red sneakers appear - he's crouching at a flower stand, smiling as he gathers a big bunch of sunflowers into his arms, petals brushing his white T-shirt. As he turns to step onto his skateboard, the flower stand owner laughs and calls out "Watch out for flying petals!" He waves back, then starts skating - a few golden petals have already escaped the bouquet, landing on the skateboard deck.
Audio, Voice & Lip Sync
This is the part that makes AI video feel expensive.
Seedance 2.0’s sound engine can:
- Generate realistic lip-sync matched to mouth movements
- Work in multiple languages - Mandarin, English, Spanish, Korean, and more
- Create environmental sound - wind, rain, traffic based on what's on screen
- Produce sound effects matched to on-screen actions
- Generate background music following visual rhythm
- Reference voice timbre from uploaded videos
If you’re a marketer, you should immediately think “localization.”
Same creative, different language, native-feeling delivery. That’s also a core EzUGC workflow: one script, many markets, 32+ languages, consistent avatar performance.
Example Prompts
Animal dialogue with voice reference
Fixed camera. Central fish-eye lens view looking down through a circular opening. Reference @Video1's fish-eye lens. Make the horse from @Video2 look into the fish-eye camera. Reference @Video1's talking movements. Background music references @Video3's sound effects.
Real estate documentary with voiceover reference
Based on the provided office building photos, generate a 15-second cinematic documentary in realistic style. 2.35:1 widescreen, 24fps, with delicate visuals. The narrator's voice timbre references @Video1. Capture "the ecosystem of the office building" - showing different companies operating inside, with narration explaining how the building has become a vibrant commercial ecosystem.
Character dialogue - comedy talk show
A comedy roast between a cat and a dog in a "Cat & Dog Roast Show":
Cat host (licking paw, eye-roll): "Folks, who even understands? This guy next to me does nothing all day but wag his tail, destroy the couch, and use those 'I'm so good, please pet me' eyes to scam treats from humans. Honestly, you destroy everything when nobody's looking, and you still go by 'Lucky'? I'd call you 'Wrecky'! Hahaha."
Dog host (head tilted, tail wagging): "You're one to talk! You sleep 18 hours a day, and the moment you're awake you rub against humans' legs for canned food. You shed so much the humans' black clothes are covered in your fur. They finish vacuuming and you immediately roll on the couch again. And you still pretend to be royalty?"
Chinese opera dialogue with period drama tones
The prelude to "The Case of Chen Shimei" begins. The black-robed Judge Bao on the left points at the red-robed Chen Shimei on the right, teeth gritted, singing in traditional opera style: "The blade meets its sheath - with irrefutable evidence, do you dare deny it?" Chen Shimei's eyes dart nervously, seeking an escape, his face deeply embarrassed. Then, from off-screen, a female opera voice calls out: "Hold!" Both Bao and Chen turn to look to the right of the frame.
Multi-character multilingual dialogue
The girl in the hat center frame warmly sings "I'm so proud of my family!", then turns to hug the girl next to her. The girl responds emotionally: "My sweetie, you're the heart of our family." The boy in yellow on the left says excitedly: "Folks, let's dance together to celebrate!" The girl on the far right responds: "I'll bring the music!" Latin music kicks in. The woman in the orange dress on the left nods and smiles. The woman with braids on the right pumps her fist. People in the crowd start stomping their feet, children clap along to the beat. The whole family forms a circle, dancing joyfully on a colorful street, skirts swirling, spreading warmth and joy.
Action squad dialogue
Fixed camera. The standing squad leader clenches his fist and commands in Spanish: "We strike in three minutes!" The knife-wielder sheathes his blade. The blond team member checks his firearm. The green-haired member grips a tactical flashlight. The dark-skinned member puts a hand on his partner's shoulder and asks in Spanish: "Flank them?" The captain nods: "Standard play - leave survivors for interrogation." Everyone goes silent, equipment clinking as they complete tactical hand signals and rise in unison, ready for battle.
Morning scene with voice timbre reference
0–3s: An alarm clock rings. A blurry image of Scene 1 fades in.
3–10s: Quick camera pan to the man's face in close-up. He tiredly calls the girl to wake up - his voice and timbre reference @Video1.
10–12s: The girl pouts and dives under the covers.
12–15s: Cut to full body shot of the man. He sighs and says: "I really can't do anything with you!"
Science explainer with educational narration
In a science-documentary style and tone, animate the content from @Image1. The story covers: Wukong needs the Banana Leaf Fan to cross Flaming Mountain, so he goes to Princess Iron Fan at Emerald Cloud Mountain to borrow it. But she refuses - her son Red Boy was subdued by Wukong and made a disciple of Guanyin, separating mother and child. She won't lend the fan and even tries to take revenge. Wukong's polite request fails, and a confrontation erupts.
One-Take Continuity
One-takes are hard for humans.
They’re even harder for generative video because continuity is where models usually break.
Seedance 2.0 is built to handle long, unbroken shots with consistent motion, scene transitions, and character persistence.
Example Prompts
Urban tracking shot
@Image1 through @Image5 - one continuous tracking shot following a runner up stairs, through corridors, onto the roof, ending with an overhead view of the city.
Imaginative scene transition - airplane to ice cream
Starting with @Image1 as the first frame. Camera zooms into the airplane window. Clouds drift slowly across the frame, one of them studded with colorful candy sprinkles. This cloud stays centered, then gradually morphs into @Image2's ice cream. Camera pulls back into the cabin - the girl from @Image3 reaches through the window to grab the ice cream, takes a bite, cream on her lips, beaming with a sweet smile. Voiceover references @Video1.
Spy thriller - continuous tracking
Spy-thriller style. @Image1 as the first frame. Camera tracks in front of a female spy in a red trench coat walking forward. Full tracking shot - passersby repeatedly obscure her. She reaches a corner (building references @Image2). Fixed camera - she exits frame around the corner and disappears. A masked girl (appearance references @Image3 - reference appearance only) lurks at the corner, glaring menacingly after her. Camera pans forward toward the spy - she walks into a mansion (@Image4) and vanishes. No cuts throughout - one continuous take.
Cozy cabin scene - POV one-take
Starting from @Image1's exterior, first-person subjective POV with a fast push-in to the interior of the cabin. A deer (@Image2) and a sheep (@Image3) sit by the fireplace drinking tea and chatting. Camera pushes in for a close-up - the teacup's design references @Image4.
Roller coaster POV
@Image1 through @Image5 - subjective POV, one continuous take of a thrilling roller coaster ride, speed increasing.
Video Editing
This is the “don’t throw away the whole render because one thing is wrong” section.
Seedance 2.0 supports targeted edits: swap characters, change hair, insert brand elements, rewrite the narrative - while preserving everything else.
Example Prompts
Complete narrative reversal
Subvert @Video1's entire storyline.
0–3s: A man in a suit sits at a bar, calm, swirling his glass. Camera slowly pushes in - elegant lighting, serious atmosphere. He whispers: "This deal... it's big."
3–6s: The woman behind him looks tense and asks "How big?" He looks up, lowers his voice: "Very big." Camera cuts to a hand close-up as he sets the glass down - full gravitas.
6–9s: Suddenly he pulls out from under the table - an absurdly oversized snack gift pack, slamming it on the table with a thud.
9–12s: The woman's hand, tensed at her waist, visibly relaxes. Her expression softens. The entire mood shifts to playful.
13–15s: The man offers her a snack. Camera pulls wide to the full bar. The image goes transparent and blurry - subtitle appears: "No matter how busy, remember to have a snack~"
Character swap preserving all motion
Replace the female lead singer in @Video1 with the male vocalist from @Image1. All actions should exactly replicate the original video. No cuts. Band performance music throughout.
Simple element addition
Change the woman's hairstyle in @Video1 to long red hair. The great white shark from @Image1 slowly surfaces behind her, half its head visible.
Commercial edit with brand insertion
@Video1: Camera pans right. The fried chicken shop owner busily hands fried chicken to queuing customers, saying in Mandarin: "After his order, yours - everyone please line up." He then grabs a paper bag printed with @Image1's design. Close-up of the branded bag. Close-up of the hand-off to the customer.
Beat-Synced Editing
If you’ve ever tried to cut a montage to music, you know the pain.
Seedance can sync visual rhythm to a reference music video so the cuts actually hit.
Example Prompts
Fashion beat-sync
The girl in the poster keeps changing outfits. Clothing styles reference @Image1 and @Image2. She holds the bag from @Image3. Video rhythm and beat references @Video1.
Multi-image rhythm montage
@Image1 through @Image7: the characters sync to @Video1's keyframe positions and overall rhythm. Characters should be more dynamic. The overall visual style is more dreamlike with strong visual tension. Adjust framing of reference images as needed for the music and scene requirements. Add lighting variations to complement the visuals.
Landscape rhythm montage
@Image1 through @Image6 scenic landscape images. Reference @Video1's visual rhythm. Transitions between scenes should match the style and musical beat for perfect sync.
Anime combat with precise timing
8-second intense strategic battle anime clip with a revenge theme.
0–3s: The female lead from storyboard 1 turns and sits. Cut - she places a chess piece, saying "You lose." Reference storyboard 2.
3–4s: Quick pan to the opposing man's face close-up (storyboard 3). He grits his teeth, furious at the result.
4–6s: Cut to overhead shot. She plays a piece - onlookers gasp in amazement (storyboard 4).
6–8s: Camera whips down. Screen goes black, then fades in - dim interior - she gazes out the window at the moonlight and quietly says "We'll see about that" (storyboard 5).
Emotion Performance
A lot of AI video looks technically fine and emotionally dead.
Seedance 2.0 improves facial expression, body language, and voice so characters can actually act.
Example Prompts
Emotional breakdown
The woman from @Image1 walks to a mirror. She looks at her reflection - pose references @Image2. She pauses in thought, then suddenly breaks down screaming. Her grabbing-the-mirror action and her breakdown expression and emotion fully reference @Video1.
Comedic transformation
@Image1 as the first frame. Camera rotates and pushes in. The character suddenly looks up - facial appearance references @Image2 - and begins roaring loudly. Intense with comedic flair - expression and demeanor reference @Image3. Then the character's body transforms into a bear - reference @Image4.
Commercial with contrasting emotions
This is a range-hood advertisement. @Image1 as the first frame - a woman elegantly cooking, no smoke. Camera quickly pans right to @Image2 - a man sweating profusely, face flushed, cooking in heavy smoke. Camera pans left and pushes in on a range hood on @Image1's table - the hood references @Image4 - frantically sucking up all the smoke.
Prompt Writing Fundamentals
Seedance prompts work best with a simple formula:
Subject + Action + Setting + Lighting + Camera Language + Style + Quality + Constraints
Example:
A young woman walking slowly along the beach, gentle breeze moving her hair, smiling toward the camera, warm golden-hour lighting, 4K HD, cinematic feel, stable camera movement, smooth and fluid footage, sharp details.
Write like a director. Not like a poet.
Also: be clear whether something is a reference or an edit. With many assets, double-check each @ label.
Action Description Tips
Seedance is strong at motion. Which means your action verbs matter.
Do use:
- Slow, gentle, continuous, natural, fluid, smooth
- "Slowly turns around," "gently raises hand," "light footsteps," "slightly lowers head," "sways with the wind"
Avoid:
- Exaggerated, high-speed, complex multi-person interactions, extreme twisting
- Single vague words like "dancing" or "walking" - be specific about the motion
Camera & Cinematography Language
Seedance understands camera language. Use it.
Shot types
- Close-up / Medium shot / Wide shot / Extreme close-up
Camera movements
- Slow push-in / Gentle pull-out / Steady pan / Half-orbit
- Fixed shot / Handheld-stable / No shake / Silky smooth
Example:
Medium shot, slow push-in, steady follow-cam, silky smooth with no jitter.
Character Consistency Tips
If you want fewer morphing faces and spaghetti limbs, add constraints.
Use phrases like:
- Clear facial features, stable face, no distortion, no deformation
- Normal body proportions, natural structure, no stiffness
- Same character, consistent clothing, unchanged hairstyle
Example:
Character's face remains stable without deformation, normal body structure, natural and fluid motion.
Video Quality Control
Don’t assume the model will “make it 4K” unless you tell it.
Resolution & clarity
- 4K, ultra-high definition, rich details, sharp resolution
Visual feel
- Cinematic quality, natural colors, soft lighting
Stability
- No blur, no ghosting, no flickering, stable footage
Multi-Shot Storytelling
Seedance can understand sequential shots as one story.
Describe shots in order:
Shot 1 is a wide view of the city. Shot 2 is a close-up of the character from @Image1. Shot 3 is an over-the-shoulder tracking shot as they walk into the building.
You can also describe transitions naturally:
Opens with a close-up of the face, slowly pulls out to a wide shot, character walks slowly, camera follows steadily, ends on a freeze-frame smile.
Real-World Use Cases
Seedance is fun for art.
But it’s useful when you apply it to work.
Ads & E-commerce
- Product demos with synced narration
- Replicate winning ad creatives with your own products
- Template-based ad scaling
- Brand-consistent multi-scene commercials
- Multi-language ads with native lip-sync
Content Localization
- Multi-language versions with native lip-sync
- Reference original video for motion, generate new dialogue in the target language
Short-Form Content
- Combine multiple clips, audio, and effects in minutes
- TikToks, Reels, YouTube Shorts
- Beat-synced montages from still images
Storyboard to Video
- Upload storyboard panels as reference images
- Describe the motion between them
- The model fills in creative details automatically
Tutorials & Guides
- Add voiceover, show step-by-step processes with AI avatars
- Science explainers with narrated animations
Film & Narrative
- Complex chase sequences with multi-angle coverage
- Emotional dialogue scenes with lip-sync
- One-take continuous shots with scene transitions
- Period dramas with historically-appropriate settings
Video Remix Ideas
Tested workflows you can try:
- Fashion runway swap - Input a runway video + a clothing image - "Replace the clothing of the model in @Video1 with the clothing in @Image1"
- Element addition/removal - Input a video - "Add a glowing neon sign in the background" or "Remove the car from the scene"
- Character replacement - Input a video + character image - "Replace the lead in @Video1 with @Image1, replicate all original actions"
- Video continuation - Input a video - "Extend @Video1 by 10 seconds with a dramatic reveal"
- Narrative reversal - Input a video - "Subvert @Video1's storyline - turn the dramatic scene into a comedy"
- Style transfer - Input a video + style reference - "Recreate @Video1 in black-and-white ink-wash style"
- Commercial recreation - Input a trending ad + your product images - "Reference @Video1's ad concept and rhythm. Replace the product with @Image1"
- Multi-character swap - Input a fight video + 2 character images - "Replace the 2 characters in @Video1 with @Image1 and @Image2"
- Music video from stills - Input photos + a music video - "Sync @Image1 through @Image6 to @Video1's rhythm and beats"
Pro Tips
Use High-Quality References
If @Image1 is blurry, your video will be blurry. Use 2K or 4K images.
Be Explicit About References
"Reference @Video1's camera movement" beats vaguely mentioning a file. Tell the model what to extract: camera work, choreography, rhythm, effects, transitions, or voice timbre.
Combine Video + Image Tags
Upload a photo as @Image1, a dance video as @Video1, then prompt:
@Image1 performs the dance from @Video1.
This is motion transfer with a custom character. It’s the pro move.
Iterate Small Changes
Don’t rewrite your entire prompt. Change one word or swap one reference.
Specify Edit vs. Reference
Make clear whether you want to edit an existing video or use it as reference.
Label Your Assets Carefully
Mixing up image numbers is the #1 cause of unexpected results.
Use Time Segments for Long Videos
For 10–15 second videos, break your prompt into time segments:
0–3s: Close-up of the character's face.
4–8s: Camera pulls back to reveal the full scene.
9–12s: Action sequence with tracking shot.
13–15s: Closing frame with brand text.
Reference Audio from Videos
You don’t need a separate audio file:
Background music references @Video1's audio.
Quick Recap
- Write actions slowly and continuously
- Write camera movements steadily and simply
- Always add stability / no-deformation / no-stiffness constraints
- Append quality + style keywords at the end
- Avoid intense, complex, multi-person actions
- Clearly label each @ reference's role
- Specify whether you're referencing or editing
What to Avoid
- Complex multi-person interactions - fighting, running, jumping, intense choreography with many people
- Vague descriptions - "looks good," "very beautiful," "super cool" (these tell the AI nothing)
- Contradictory requirements - "ultra-fast speed + extremely stable" (pick one)
- Mixing up @ references - accidentally referencing the wrong image or video
- Prohibited content - explicit, violent, dangerous actions, or copyrighted characters
- Uploading realistic human faces - current policy does not support uploading materials containing realistic human faces (official examples shown may use internal permissions)
Ready-to-Use Prompt Templates
Template 1: Cinematic Portrait Video
A young woman walking slowly through a forest, gentle breeze lightly sweeping her hair, natural smile, warm sunlight, medium shot, slow push-in, smooth and stable footage, 4K HD, cinematic feel, clear face without deformation, normal body structure, rich details.
Template 2: Atmospheric Landscape
Seaside sunset, waves gently lapping the shore, camera slowly panning horizontally, warm orange tones, calming and fresh, silky smooth footage, 4K ultra-high definition, no flickering, no ghosting.
Template 3: Image-to-Video (First Frame Lock)
Based on the reference image, maintain consistent character appearance and clothing, slow hand-raise and turn, natural and fluid, no stiffness or deformation, stable camera movement, high-definition details, cinematic quality.
Template 4: 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.
Template 5: 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.
Template 6: Multi-Scene Narrative with Audio
The man from @Image1 walks through the corridor. Close-up: he takes a deep breath at the door, adjusts his expression. He unlocks the door and enters - his daughter and pet dog run to greet him with hugs. Warm interior. Natural dialogue throughout with ambient home sounds.
Template 7: 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 pu
Where EzUGC Fits (If You’re Making Ads)
Seedance is a powerful generator. But most ad teams don’t just need “a video.” They need volume, consistency, and variants.
That’s the gap EzUGC is built for.
- Traditional UGC: ~$200/video hiring creators
- EzUGC AI UGC: ~$5/video with better consistency
- Real-looking AI avatars that speak 32+ languages
- Built for DTC brands, agencies, and performance marketers
- Create video ads in minutes, not days
If you want to pump out 20 hook variations, localize them, and keep the product and pacing consistent, try EzUGC here.