Video model · ByteDance ·
Seedance 2.5 is a video model by ByteDance. In Endless it costs from 51 credits per video and is available on the Hobby plan and up.
Seedance 2.5 is ByteDance's video model for 30-second takes in one pass, guided by up to 50 references, with sound generated with the picture.
Seedance 2.5 generates up to 30 seconds of video in a single continuous pass, so a long take does not have to be stitched together from shorter clips. You can guide it with up to 50 image, video and audio references to keep characters, products and style consistent. Picture and sound come out of the same generation, so impacts and lip movement line up with the audio.
Attach a clip and the model works on that clip. The default editing task changes what happens in the shot and keeps its frame and length, the extension task continues it from where it ends, and the reference task builds a new clip that only takes guidance from the attachment. Aspect ratio and duration settings apply in the reference task only.
Seedance 2.0, its predecessor in the Endless catalog, makes clips of 4 to 15 seconds at up to 4K. Choose 2.5 when you need a longer take, many references or changes to footage you already have, and switch between the two in the same workspace, in chat or on the canvas.
Seedance 2.5 generates picture and sound together in one pass, for up to 30 seconds. ByteDance built it for storytelling across that whole length: in one generation the model can organize several connected shots, so a scene has a setup, a turn and an ending instead of one moment stretched across the clip. That changes how you write. A prompt sized for a 5-second clip runs out of instructions long before 30 seconds, and the model fills the rest on its own.
Write the prompt as a short shooting script. Say what the whole piece is in one sentence, describe what happens in order, then state what must stay the same throughout.
This layout works for longer clips:
@Image1: the courier (subject). @Image2: the rooftop (scene).
Summary: one sentence with subject, place, event, style and camera.
0–8 s: what we see, how the camera moves, what we hear.
8–16 s: ...
16–24 s: ...
Consistency: what stays fixed (wardrobe, light, camera feel).
Cover the full duration with timestamps and leave no gaps. Runway's guide reports that a gap between two timed blocks is left without direction rather than filled in, suggests about 6 to 8 seconds per beat, and treats timestamps as pacing guidance rather than frame-exact cuts. For a short clip you can skip the timeline and write two or three sentences in order: subject, action, setting, camera, light, sound.
ByteDance's own examples use plain camera language: one-take handheld gimbal tracking shot, slow push-in, pull back, arc around, top-down view, whip-pan, micro-FPV, wide shot. Name the move and what it follows ("the camera tracks her through the curtain into the corridor"), not only the shot size. Runway found that a beat longer than about four seconds can carry two movements, such as a side-follow that turns into an orbit, so you do not need a cut just to change the move. If you want one unbroken shot, write "single continuous take, no cuts".
Generate audio is on by default, and sound comes from the same pass as the picture, so footsteps, impacts and lip movement line up. Describe the sound the way you describe a shot: dialogue, ambience, effects and music. Write dialogue as plain labeled lines, such as Courier: 'Last one tonight.', and keep each line short enough to fit its beat.
Keep negative phrasing for audio and text, where it works: "No subtitles. No background music, room sound only." ByteDance says 2.5 reduces unwanted subtitles and background music, but stating it is still more reliable. To drive timing from a voice-over or a track, attach it as reference audio, call it @Audio1, and say which action lands on which line.
Attach images, a clip or audio, then name each one in the prompt by its tag: @Image1, @Image2, @Video1, @Audio1. Give every reference one job before you write the action ("@Image1 is the bottle, exact label and cap; @Video1 is the camera movement only"). A reference without a stated job leaves the model guessing what it controls, and the risk grows with every attachment. When a character appears in several beats, repeat two or three of their fixed traits at each beat.
With a single image, the First frame setting decides whether the clip starts on that exact image or uses it as a reference for subject and style. With a clip attached, Clip mode decides what happens:
ByteDance also describes edits limited to a time range ("4–8 s only") and green-screen editing, where a subject shot on green is placed in a new environment and reacts to it, with wind in the clothes and matching light. A textureless clay render can set the camera path and blocking while an image sets the materials and lighting.
| Feature | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
| Length per generation | Up to 15 s | Up to 30 s |
| References per generation | 9 images, 3 videos, 3 audio clips | 30 images, 10 videos, 10 audio clips |
| Extension | Video extension | Multiple rounds of extension |
| Editing | Targeted edits to clips, characters, actions and storylines | Timestamp-level edits, plus green-screen and camera perspective editing |
| Price | From 16 credits per video | From 51 credits per video |
| Availability | Available on every plan | Available on Hobby and above |
In your workspace, open Video and pick Seedance 2.5 in the model picker.
Write a prompt with the subject, camera and motion, or start from an image you upload or generate.
Generate, then refine in chat or on a canvas. Every clip is saved to your workspace assets.
Grouped by use case. Each one opens in Endless with Seedance 2.5, ready to edit and generate.
@Image1 is the perfume bottle, exact faceted glass and gold cap, identical in every shot. 0–8 s: macro push-in across the bottle on wet black slate while a single drop runs down the glass, rain ticking softly. 8–18 s: the camera arcs around it as warm backlight rises through the liquid. 18–24 s: slow pull-back to a centered hero frame, no subtitles, a low cello note only.
A matte green insulated bottle sits on a mossy riverside rock at dawn. Single continuous take: the camera starts low on the water, rises over the bottle as a hand lifts it and pours steaming tea into the cap, then tilts up to reveal a misty valley. Sound of the river, the pour and a distant bird, no music.
@Image1 is the sneaker, exact white leather and orange sole. 0–6 s: top-down shot as the shoe drops onto cracked asphalt and dust bursts outward in slow motion. 6–14 s: low tracking shot beside a runner wearing it through an empty city street at sunrise, footsteps crisp and rhythmic. 14–20 s: hard cut to the shoe alone, rotating slowly on a concrete plinth as the footsteps fade out.
A ceramic coffee mug on a kitchen counter in the morning, steam curling, soft window light from the left. The camera slowly dollies in while a hand slides a folded newspaper next to it and the toaster pops in the soft background. Room sound only: the toaster, a spoon against ceramic, birds outside.
@Image1 is the lipstick, exact gold tube and berry shade. Fast-paced editorial ad on a seamless pink backdrop: 0–5 s, the tube spins and uncaps in extreme macro; 5–10 s, a model applies it in close-up with a slow rack focus to her eyes; 10–15 s, whip-pan to the tube standing in a field of pink petals. Upbeat electronic beat with crisp click effects on each cut.
A folding electric bike unfolds by itself on a quiet Tokyo side street at dusk, each hinge clicking into place as the camera orbits once around it. A rider in a navy raincoat steps on and sets off, and the camera follows from behind under flickering neon. Mechanical clicks, a soft electric whir, light rain.
Single continuous take, no cuts, handheld gimbal. The camera follows a chef in a white jacket from the back door of a busy restaurant through the steaming kitchen, past cooks plating dishes, through the swinging doors and into the candlelit dining room, ending wide as she sets a plate in front of a smiling couple. Kitchen clatter and sizzling pans, then soft jazz and murmured conversation.
Summary: a lighthouse keeper climbs the tower at night during a storm to restart the lamp. 0–8 s: wide shot of waves exploding against the rocks, the keeper leaning into the wind toward the door. 8–18 s: tracking shot up the spiral stairs, a lantern swinging in his hand, rain hammering the windows. 18–30 s: he pulls the lever, the lamp roars to life and the camera rises through the glass room to a wide aerial of the beam sweeping the sea.
One unbroken drone flight through a mountain village at sunrise: over terraced rice fields, between tiled rooftops where smoke rises from chimneys, down a narrow street where children run past, and out over a cliff edge to reveal the valley below. Cool morning haze, warm low sun. Natural ambience of roosters, wind and distant voices, no music.
Single continuous take of a street musician in a subway station. The camera starts on his hands tuning a battered acoustic guitar, pulls back as he begins to play, then slowly circles him as commuters slow down, a girl drops a coin and an older man starts to dance. The song swells with the station echo, ending on a wide shot as a train pulls in.
Summary: a fox crosses a winter forest at night, nature documentary style. 0–10 s: low tracking shot beside the fox trotting through deep snow, breath fogging under moonlight. 10–20 s: it freezes with ears up, then leaps headfirst into the snow and comes up empty, held in a locked-off medium shot. 20–30 s: slow pull-back to a wide shot of the fox walking on between dark pines, crunching snow, wind and a distant owl.
One-take tour of a small ceramics studio, the camera moving steadily at chest height: from the open door past shelves of drying bowls, to a potter centering clay on a spinning wheel, then through a doorway into the kiln room where an apprentice opens the kiln and orange light floods the frame. Wheel hum, the kiln door creaking, no music.
Two-shot, locked-off camera at eye height inside a cramped food truck at night. The cook, a woman in her forties with a red bandana, slides a box across the counter. Cook: 'Last one tonight, and it's on the house.' Customer, a tired young man in a delivery jacket, grinning: 'You say that every night.' Sizzling grill, traffic outside, no subtitles.
@Image1 is the grandmother, @Image2 is the granddaughter. Medium close-up at a kitchen table under warm lamplight as the grandmother teaches her to fold dumplings and laughs when the girl's dumpling falls apart. Grandmother: 'Again, slowly this time.' Soft clatter of bowls, rain on the window, clean lip sync, no subtitles.
A detective in a rumpled grey suit stands in a rain-soaked alley, a flashlight beam cutting through the fog. Slow push-in as he crouches over a dropped umbrella and says under his breath: 'She was here. Ten minutes ago, maybe less.' Steady rain, a distant siren, a low synth drone.
0–6 s: a nervous groom paces in a church side room while his best man fixes his tie. 6–12 s: Best man, steady voice: 'Breathe. She already said yes once.' The groom laughs and his shoulders drop. 12–18 s: the doors open, organ music swells and the camera follows them out into the bright aisle.
A puppet-style owl in a tweed waistcoat sits at a tiny desk, peering over half-moon glasses at a young hedgehog. Owl, dry and slow: 'Homework, Mr. Prickle. Where is it?' Hedgehog, squeaking: 'The wind took it, sir. Truly.' Stop-motion texture, warm desk lamp, a clock ticking.
@Audio1 is the narrator's voice-over; generate no other voice. A young woman opens the shutters of a bakery at dawn, and the camera follows her as she pulls trays of bread from the oven and arranges them in the window. Time each action to the narration and end on her flipping the sign to open as the last line finishes.
Edit @Video1. Keep the characters, their actions and the lighting exactly as they are. Change only the season: snow covers the street, breath fogs in the cold air and the trees are bare.
Edit @Video1. Keep the composition, camera position and performance rhythm. Change only the actor's expression from neutral to quietly amused, and keep his voice and lip sync as they are.
Edit @Video1. Keep the dancer and her movements unchanged and replace the green-screen background with a rooftop at sunset over a dense city. Let the wind move her hair and dress from the left, and light her with warm low sun to match.
Extend @Video1. Continue from its last frame with the same characters, scene, look and sound. The cyclist reaches the top of the hill, stops, takes off her helmet and looks out over the bay as the camera swings around to her side.
Edit @Video1, 4–8 s only. Replace the red car with a vintage blue convertible and leave everything before and after untouched, including the camera move and the street sound.
Edit @Video1. Leave the people, what they do and the overall look as they are, and rework only the camera: a slow push-in on the table, a rise to a top-down view as the plates are set down, then a descent into a close two-shot of the couple.
@Image1 is the mascot, a round orange robot with one blue eye, identical in every shot; @Image2 is the city park. 0–10 s: the mascot rolls along a path, waving at joggers. 10–20 s: it helps a child get a kite down from a tree with a telescopic arm. 20–30 s: wide shot as they fly the kite together at sunset, light orchestral music and playful beeps.
@Image1 is the presenter, @Image2 is the studio set, @Video1 is the camera movement only. The presenter walks toward the camera across the set explaining a budgeting app, gesturing at a large screen behind her. Presenter: 'Three taps, and your month is planned.' Clean studio sound, no music, no subtitles.
@Image1 for the café interior, @Image2 for the barista, @Image3 for the cup design. The barista pours latte art in close-up, slides the cup across the counter, and the camera follows it to a customer by the window. Keep the printed logo on the cup sharp and unchanged, with espresso machine hiss and quiet chatter.
@Image1 is the storyboard sheet: follow its shot order, framing and camera notes. @Image2 is the main character and @Image3 is the forest location. Render it as a warm hand-painted 2D animation with soft grain, gentle wind and birdsong.
@Video1 is a textureless clay render that sets the camera path, character positions and blocking. @Image1 sets the materials, color and lighting. Render the clay layout as a photoreal night market with lanterns, steam from food stalls and a crowd moving through, keeping the camera path exactly as in @Video1.
@Image1 is the hiking jacket, exact yellow fabric and black zips; @Image2 to @Image4 are three different hikers. Three beats of eight seconds: each hiker wears the jacket on a different trail, desert, then forest, then a snowy ridge. Keep the jacket identical in every beat and end on all three at a summit cairn, with wind and footsteps and no music.
by ByteDance
Clips of 4 to 15 seconds steered by image, video and audio references
From 16 credits per video
by Black Forest Labs
Clips up to 20 seconds with sound, from a prompt, frames or a clip
From 90 credits per video
by Kuaishou
Multi-shot clips with spoken dialogue, motion transfer and a 4K mode
From 27 credits per video
Seedance 2.5 next to other models: price, maker and plan side by side.
Start free, create a project, and bring your brand’s references together. Or book a demo to explore Endless with your team.