Seedance 2.5 prompt guide: structure, timestamps, dialogue and sound

One continuous shot with no cuts. 0-8s: dawn over a small fishing harbour; the camera glides low over still water towards a red wooden boat. 8-16s: it rises past the mast as a fisherman in a yellow oilskin coils rope on deck and glances up at the sky. 16-24s: the camera drifts along the quay past stacked crates while gulls lift off the railings. 24-30s: it pulls back and settles on a wide shot of the whole harbour as the sun clears the hills. Soft gold light, thin mist on the water, realistic film look. Audio: gulls, creaking rope, water lapping against the hulls. No music. No subtitles.

Why it works: Four timed beats give the model a plan for all 30 seconds, while "one continuous shot with no cuts" keeps it from editing.

One smooth gimbal shot through a bright Scandinavian apartment. 0-5s: the front door opens onto a hallway with oak floors. 5-10s: the camera moves into an open living room with a pale linen sofa and tall windows. 10-15s: it turns left past a kitchen island with a bowl of lemons. 15-20s: it ends on the balcony doors as they open onto a city view at golden hour. Eye-level camera, steady walking pace, no people, natural daylight. Audio: soft room tone, distant traffic through the open doors. No music.

Why it works: A fixed eye level and walking pace stop the camera from wandering, and each room gets its own five seconds.

One continuous handheld shot in a small ramen bar at night. 0-6s: close on a steaming pot as a chef in a black headband lifts noodles with a strainer. 6-12s: the camera follows his hands as he folds the noodles into a bowl, ladles broth and lays slices of pork and a soft egg on top. 12-19s: it tracks the bowl as he slides it across the wooden counter. 19-25s: it rises to a medium shot of a young woman catching the bowl, smiling and breaking her chopsticks apart. Warm tungsten light, steam in the air. Audio: bubbling broth, the clatter of the kitchen, quiet chatter. No music.

Why it works: The bowl is the thread that ties every beat together, so the long take never loses its subject.

Vertical tracking shot following a trail runner in an orange vest along a ridge at sunrise. 0-5s: the camera follows from behind at shoulder height. 5-10s: it swings round to her side as she leaps a small rock. 10-15s: she slows to a walk at the summit cairn and looks out over a sea of clouds. Crisp morning light, slight handheld movement, realistic sports film look. Audio: footsteps on gravel, heavy breathing, wind. No music.

Why it works: The camera's position is named for each beat (behind, then beside her), which keeps the follow shot readable in 9:16.

A premium coffee commercial in three shots. Shot 1 | 0-4s: extreme close-up of dark beans tumbling into a grinder, slow motion, hard side light. Shot 2 | 4-9s: medium close-up of espresso pouring into a white cup on a marble counter, the crema swirling, camera slowly pushing in. Shot 3 | 9-15s: medium shot of a man in a grey jumper taking the first sip by a rainy window, eyes closing for a moment. Warm brown and cream palette, shallow depth of field. Audio: grinder whirr, the hiss of the machine, rain on glass, a soft piano note at the end. No subtitles.

Why it works: "Shot N | time" is the shot-list format BytePlus uses in its own examples, and each shot carries its own framing and move.

A warm live-action short film in six shots. Shot 1 | 0-5s: wide shot of a rainy city street at dusk; a bike courier in a green jacket brakes beside a small wet terrier sitting alone under a bus shelter. Shot 2 | 5-9s: close-up of the dog's collar tag. Shot 3 | 9-15s: medium shot; the courier crouches, wraps the dog in his scarf and tucks it into his delivery bag with its head poking out. Shot 4 | 15-21s: tracking shot of the bike weaving through traffic, the dog's ears flapping. Shot 5 | 21-26s: a front door opens and a girl in pyjamas gasps. Shot 6 | 26-30s: close-up of the dog licking her face while the courier rides off, smiling. Soft rain, neon reflections. Audio: rain, bike bell, traffic; the girl laughs. No subtitles.

Why it works: Six shots over 30 seconds gives each one four to six seconds, enough to read without crowding the timeline.

0-5s: macro shot of a silver coin spinning on a dark wooden table, light glinting off its edge, camera locked off. At the 5-second mark, match cut on the same circular motion to Earth spinning slowly in space, lit from the left by the sun. 5-10s: the camera pushes in slowly towards the planet. Cinematic, high contrast. Audio: the metallic ring of the coin fades into a low, quiet rumble. No music.

Why it works: BytePlus recommends naming both the moment of a transition and its method, so the cut lands where you want it.

A tense action teaser. 0-4s: low-angle wide shot of a woman in a long black coat walking through an empty underground car park, fluorescent lights flickering. 4-8s: she stops and turns her head; a car's headlights snap on behind her. 8-12s: speed ramp: the car lunges forward, slow motion as she dives sideways, then back to normal speed. 12-16s: close-up of her rolling to her feet and switching on a torch. At the 16-second mark the frame freezes for 1 second on her face lit by the beam. 17-20s: cut to black. Audio: echoing footsteps, engine roar, a deep percussive hit on the freeze, then silence. No subtitles.

Why it works: Point-in-time instructions ("at the 16-second mark", "freezes for 1 second") give the edit its rhythm.

A small, warmly lit kitchen at night. Medium two-shot, locked-off camera. A woman in her sixties with silver hair dries a plate; her adult son leans on the counter. 0-3s: the son's line (hesitant): {Mum, I took the job in Lisbon.} 3-6s: she stops drying and looks at him; her line (quietly): {When?} 6-9s: the son's line (gently): {Next month.} 9-12s: she puts the plate down and smiles with wet eyes; her line (warmly): {Then we'd better start packing.} Natural performances, realistic film look. Audio: the dialogue, a ticking clock, the fridge hum. No music. No subtitles.

Why it works: "Character's line (emotion): {line}" is the format BytePlus suggests when different lines need different delivery.

Vertical selfie video. A young man with curly hair sits in a parked car, phone held at arm's length, natural daylight through the windscreen. He speaks in English with a light Irish accent, relaxed and amused: {I've tried every alarm app going. This is the first one that actually got me out of bed.} He laughs and shakes his head. Slight handheld wobble, phone camera look. Audio: his voice, faint traffic outside. No music. No subtitles.

Why it works: Language, accent and delivery are stated before the line, and nothing is repeated after it (repeated words can trigger subtitles).

Nature documentary. 0-7s: aerial shot gliding over a misty pine forest at sunrise. 7-15s: slow push-in on a red fox stepping out of the trees onto frosted grass, its breath visible. Narrator (V.O.), a calm, deep British male voice: {Every winter morning begins the same way. With hunger, and with patience.} The narrator is never on screen. Audio: the narration, birdsong, frost crunching under paws. Very soft strings underneath, never masking the voice. No subtitles.

Why it works: "(V.O.)" and "never on screen" stop the model from inventing someone to lip-sync the narration.

A busy street market in Mexico City at midday, bright sun and colourful awnings. Medium close-up of a cheerful fruit seller in a straw hat holding a mango up to the camera. She speaks in Mexican Spanish, fast and friendly: {¡Pruébalo, está dulcísimo!} Handheld camera at chest height. Audio: her voice, market chatter, a passing scooter. No music. No subtitles.

Why it works: BytePlus lists native speech in more than 10 languages; naming the language and regional accent first steers the voice.

Audio policy for the whole clip: natural ambience only. No music, background music, BGM, score, instrumental, melody, synth, ambient pad or drone at any point. A quiet library reading room at night, green desk lamps, rain streaking the tall windows. Slow dolly along the empty desks towards a single open book. Audio: rain on glass, a clock ticking, a page turning by itself at the end. Repeat: no music of any kind.

Why it works: BytePlus's fix for unwanted music: list every word for music, and state the rule at both the start and the end.

Vertical skate video at a sunny concrete skatepark. Music: upbeat lo-fi hip hop with a dusty drum loop and vinyl crackle. 0-5s: a skater rolls in from the top of a bowl, camera following low behind the board. At the 5-second mark the beat drops: cut to a slow-motion kickflip against the blue sky. 9-15s: back to normal speed as he lands, rolls away and fist-bumps a friend. Board sounds stay audible over the music. No subtitles.

Why it works: Describing the music's genre and texture, plus where the drop lands, ties the cut to the track.

Macro product shot of a mechanical wristwatch on dark slate. 0-3s: extreme close-up of the second hand sweeping. 3-6s: the camera slides round to the open case back, gears turning. 6-8s: a hand lifts the watch out of frame. Soft top light, deep shadows. Audio: a crisp, close tick every second, the faint whirr of the gears, a soft click as it is picked up. No music. No voice.

Why it works: Specific, close sounds tied to visible actions read better than a general "satisfying sounds".

@Image1 is the presenter; she speaks with the voice timbre from @Audio1. She stands in a bright modern office, looking into the camera, medium shot, slow push-in. She says, warm and confident: {Three tools, one subscription, and your Monday morning just got shorter.} She finishes with a small nod. Keep her face, hair and clothing exactly as in @Image1. Audio: her voice and soft office ambience. No music. No subtitles.

Why it works: Each reference is bound to one job (the face from the image, the voice from the audio), the mapping BytePlus recommends.

@Image1 and @Image2 show the same water bottle from the front and the back. The bottle stands on a wet black rock beside a mountain stream. 0-5s: slow orbit from the front label round to the back, water droplets running down the steel. 5-10s: a hand lifts it and twists off the cap as the camera tilts up to a pine forest. Keep the shape, colour and label exactly as in the references. Crisp morning light. Audio: running water, the cap twisting. No music. No on-screen text.

Why it works: Seedance 2.5 accepts multi-view images of one subject, which BytePlus says 2.0 did not handle well.

Strictly follow the dance moves and the camera movement in @Video1, in the same order. The dancer is the woman from @Image1, wearing the outfit from the photo. The setting is a rooftop at sunset with the city behind her. Warm golden light, a lens flare at the end. Audio: keep the music from @Video1. No subtitles.

Why it works: When a reference already shows the motion, BytePlus advises saying "follow it" instead of describing every move again.

@Image1 is the first frame and @Image2 is the last frame. A paper boat floats down a rain gutter in a quiet street, gathers speed, tips over a small drop at the drain and lands upright in a puddle, where a child's boot steps in beside it. Continuous shot at street level, soft overcast light. Audio: rain, trickling water, a splash. No music.

Why it works: Naming the frames in the prompt keeps the aspect ratio you picked; a locked first-frame task would follow the image's shape.

Use @Image1 to @Image4 in order as keyframes. A hot-air balloon lifts off a frosty field at dawn, rises through low cloud, drifts over a snowy mountain ridge and descends towards a lakeside village as the sun comes up. The camera stays alongside the basket the whole way. Smooth, continuous motion between keyframes, soft pastel light. Audio: burner bursts, wind, distant church bells at the end. No music.

Why it works: BytePlus's rule for strict storyboards: the keyframe instruction goes in the first sentence.

@mara and @leo sit on a park bench in autumn, orange leaves falling. Shot 1 | 0-5s: wide shot; @leo offers @mara a paper cup of coffee. Shot 2 | 5-10s: close-up; @mara takes it and says, teasing: {You remembered the oat milk.} Shot 3 | 10-15s: two-shot; @leo shrugs and says: {I remember everything.} She laughs. Soft afternoon light, realistic film look. Audio: their voices, rustling leaves, children playing in the distance. No music. No subtitles.

Why it works: Saved characters stay bound to their @handle in every shot, which helps keep two people from swapping faces.

A detective stands at the far end of a dim corridor. Rack focus: the focus shifts smoothly from a brass door handle, sharp in the foreground, which blurs, to the detective in the background, who becomes clear as he turns towards the camera. Static camera, cold blue light, film noir mood. Audio: a buzzing strip light, slow footsteps. No music.

Why it works: BytePlus advises "term + explanation" for technical moves; the explanation tells the model what changes in frame.

Dolly zoom on a young woman gripping the rail at the top of a spiral stairwell, looking down. She stays the same size in frame while the stairwell below stretches away and deepens. Overhead angle, cold daylight from a skylight, tense mood. Audio: a low rising hum, her sharp breath. No music.

Why it works: "Dolly zoom" is on BytePlus's list of techniques that work as written, and the effect is described in one clause.

FPV drone shot. 0-4s: racing low over a jungle river. 4-8s: it banks hard right and flies straight through a curtain of waterfall spray. 8-12s: it bursts out into a sunlit valley full of mist and pulls up into the sky. Fast and smooth, slight motion blur, saturated greens. Audio: rushing wind, the roar of the water as it passes through.

Why it works: Each timed beat is one manoeuvre, so the speed stays high without the path becoming random.

A street dancer jumps into a mid-air split in a graffiti alley. At the 3-second mark time freezes: the camera orbits 180 degrees around her while dust and a dropped cap hang in the air. At the 6-second mark time snaps back and she lands. Bright afternoon light. Audio: a whoosh into the freeze, near silence during it, then a hard landing thud.

Why it works: Timestamps on the freeze and the release give bullet time a clear start and end.

The courier rides round the corner onto a quiet bridge as the rain stops. He slows down and lets the terrier look out over the river, and the camera rises to a wide shot of the city lights coming on. Same characters, same jacket, same evening light. Audio: dripping water, the bike's freewheel ticking, distant traffic. No music. No subtitles.

Why it works: Extend mode adds the "extend the video" instruction for you, so the prompt only needs the next beat and the continuity.

The woman at the counter looks up from her phone, straight into the camera, and says with a small smile: {Same time tomorrow?} Hold on her face for two seconds. Keep her voice, clothes and the cafe lighting the same. Audio: her voice and the cafe ambience from the clip. No music. No subtitles.

Why it works: Extensions carry the source's look; asking for the same voice and light keeps the join seamless.