ByteDance · BytePlusVIDEOCore tiersNative audio

Seedance 2.5 on VisionX.

Live on the roster: one continuous take up to 30 seconds, with 30 reference images behind it.

1model on rosterfrom 16.2 VXper shotup to 30reference imagesCast-readyidentity locked

OTTER IS NOBODY · TRAILER 0:16 · MADE WITH SEEDANCE 2.5 · OPEN-SOURCED BY BYTEPLUS

01

What Seedance 2.5 is on the roster

THE MODEL

Seedance 2.5 is live on VisionX. It is built around a different shape of output from the 2.0 tiers: one continuous take up to 30 seconds — a whole story beat, unbroken — where 2.0 puts several shots inside a 15-second clip.

It runs in the same composer as every other engine, sees the same Cast, bills the same VX wallet, and quotes its price per shot before you run it. Route a board line to it the way you would route to Seedance 2.0 or a guest engine; nothing about your campaign changes except the length of the take.

A 30-second single take

One continuous shot with no stitching and no scene-cut splicing — double the 2.0 ceiling, long enough to carry a complete narrative beat with dialogue.

30 reference images in one call

More than three times the nine the 2.0 tiers take, alongside video and audio slots — each reference carrying the role you name. Deep reference budgets are what keep cast, wardrobe and product consistent across a long take.

Edit and extend by prompt

Hand it a finished clip and ask: remove the background music, replace the subject, extend the story forward. The task type is read from your intent, not from a separate tool.

02

Otter is Nobody, shot by shot.

OPEN-SOURCE FILM

A little otter on its first day as a courier is handed a delivery nobody else will take, and gets it done — through a city torn up by giant insects, a dogfight down a glass canyon, a flooded pipe, and a very small customer. BytePlus released the 3:21 film with ten of its shot prompts and reference packs public. The ten are below, cut at the gallery's own timecodes, each with the prompt that made it and the settings it ran at — on Dreamina-Seedance-2.5, the model this page is about.

Film
Otter is Nobody
Form
Animated short · 3:21
Director
BLUESKY
AIGC director
Cheng Kuan
Technical director
Lu Qi
Model
Dreamina-Seedance-2.5
Format
21:9 · 720p · native audio
Published
July 2026 · BytePlus ModelArk gallery, open-source
  1. Clip 01 · 0:04–0:07

    The flower on the skyline

    A high-angle shot looking downward. A building occupies the foreground while the distant giant flower on the skyline is out of focus. A giant insect suddenly bursts out of a lower window, blasting glass outward, and focus snaps violently from the distant flower to the erupting insect.

    Fixed · 15 s4 image refs
  2. Clip 02 · 0:19–0:25

    Just a courier

    An old bright-yellow city bus being slowly crushed and rolled by the beetle into an almost perfect sphere. Body panels bend, windows deform, and eventually the entire vehicle compresses into a flawless yellow ball. The location is an urban street canyon overtaken by giant spreading tree roots.

    Smart time · 30 s ceiling3 image refs
  3. Clip 03 · 0:32–0:47

    Do not puke on my feathers

    The overall feeling is a Top Gun-style aerial dogfight through a teal glass-tower city — extremely fast, composed of many short shots, hard cuts only, and no slow motion. … Hornbill side close-up (50mm): it glances back at the beetles, then at the otter, with sharp controlled movement. Dialogue: “Whatever you do — do NOT puke on my feathers.”

    Smart time · 30 s ceiling4 image refs1 audio ref
  4. Clip 04 · 1:02–1:10

    A tiny seed

    Dialogue must appear exactly, with no omissions, in English only, and must lip-sync to the audio reference. In the third shot, the otter says in a weak, gentle, cute breathy voice: “A tiny seed? So you're my first customer?” The film uses three shot / reverse-shot setups with different framing and angles, constant real-time speed … Emotional arc: pain and fear → confusion → curiosity.

    Smart time · 30 s ceiling4 image refs1 audio ref
  5. Clip 05 · 1:12–1:23

    Kitchen chase

    The action is a one-direction comedic chase within a single environment, using hard cuts, 21:9 framing, total runtime 30 seconds, entirely inside the ruined kitchen, progressing left to right. It uses an escalating repeated beat: otter closes in and pounces on the seed, getting closer each time → fleeing seed grabs an object and throws it backward without looking → the object interrupts the capture at the last possible instant → seed regains distance → repeat. No dialogue.

    Smart time · 30 s ceiling3 image refs
  6. Clip 06 · 1:27–1:33

    Multiverse lunge

    A comedic “style-hopping” chase short titled “MULTIVERSE LUNGE.” … During the middle slow-motion section, the camera must orbit the two characters for one full complete revolution, an uninterrupted 360° circle that returns all the way to its starting position. … Each style lasts only one or two frames, about eight styles per second. … Style order: 1. Monet-like Impressionist oil painting, broken color and soft chromatic shadows. 2. Van Gogh-like treatment, thick swirling impasto, cobalt blue against chrome yellow.

    Smart time · 30 s ceiling3 image refs
  7. Clip 07 · 1:36–1:40

    The pipe

    Low over-the-shoulder view from the seed toward the fan throat: exposed blades fill the tunnel with no grille. The vortex pulls the seed inward, black-dot eyes filled with terror. From a branch pipe, the otter cuts at high speed into the main tunnel and sees the danger. … Fully streamlined sprint-swim, powerful tail oscillation, wet fur stripping off a sheath of microbubbles, closing the distance at about three times current speed as pipe walls streak by.

    Smart time · 30 s ceiling3 image refs
  8. Clip 08 · 1:40–1:57

    Need a hand, kid?

    Camera work uses aggressive handheld faux-documentary style: violent shake, frequent whip pans, visible loss and recovery of focus, and unstable caught-on-the-spot framing. … During this chaos, the petroleum-blue spacecraft finally drifts slowly into frame from the upper rear and matches their speed. Inside, the Malayan tapir boss lounges calmly, piloting one-handed and casually sipping from the tiny white coffee cup. … Dialogue consists only of the boss saying “Need a hand, kid?”

    Smart time · 30 s ceiling8 image refs1 audio ref
  9. Clip 09 · 2:08–2:18

    The dive

    Shot 1, approximately 0–3s, establishes the layered composition with a steep top-down angle and Dutch tilt. An extreme-high-altitude wide view, sharply tilted, with the spacecraft in the near foreground corner, clear and sharp, trailing blue exhaust. Looking downward past the spacecraft: it is positioned extremely far above the giant flower, with a dizzying vertical distance between them. Clearly show three layers of atmospheric perspective.

    Smart time · 30 s ceiling11 image refs
  10. Clip 10 · 2:50–3:01

    Fern slide

    The overall tone is that of a healing fantasy picture book — soft, semi-translucent, dreamlike, and joyful, like a wildly celebratory sci-fi aerial theme park crossed with a roller coaster. … Giant elevated fern leaves hanging from above naturally act as slides / roller-coaster tracks. Semi-transparent stems glow from within. Golden-orange morning mist and floating dust fill the air. … A semi-transparent petal platform carries everyone through continuous flight and slowly rises toward the exit in the finale.

    Smart time · 30 s ceiling5 image refs2 audio refs

* Every frame generated on Dreamina-Seedance-2.5 by BLUESKY, published open-source by BytePlus in the ModelArk gallery and shown here as published. Prompts quoted verbatim from that release. Not rendered on VisionX — the same model runs on the roster here.

03

Wonder Boy, shot by shot.

OPEN-SOURCE FILM

A boy on a Chester street in 2006 watches Michael Owen on a pub television, then plays the same moves in the street — every touch flowing into a moment from Owen’s career — until, in 2026, he is the one wearing the number 10 shirt at kickoff. BytePlus released the 1:44 film with seven of its shot prompts and reference packs public. The seven are below, cut at the gallery's own timecodes, each with the prompt that made it and the settings it ran at — on Dreamina-Seedance-2.5, the model this page is about.

WONDER BOY · TRAILER 0:21 · MADE WITH SEEDANCE 2.5 · OPEN-SOURCED BY BYTEPLUS

Film
Wonder Boy
Form
Concept film · 1:44
Presented by
BytePlus
Director
Zhang Weidong
AIGC directors
Pan Deng, Zheng Zekun, Zhang Weidong
Art director
Guo Zhihan
Executive producers
Lai Xin, Zhou Peng
Producer
Guo Jiafei
Model
Dreamina-Seedance-2.5
Format
21:9 · 1080p · native audio
Published
July 2026 · BytePlus ModelArk gallery, open-source
  1. Clip 01 · 0:00–0:13

    In front of the television

    Create a premium, strongly commercial football TVC set on the streets of Chester in 2006. Maintain a consistent mid-2000s atmosphere throughout: an old television, glass shopfronts, vintage buses, street pubs, newspaper stands, brick walls, overcast natural light, and wet stone pavement. … Shot 4: Frame a frontal upper-body close-up of Owen with slight handheld movement, rapidly push in, then stop. Owen stands near midfield, lowers his head slightly, and takes a deep breath without visible condensation.

    Fixed · 15 s6 image refs
  2. Clip 02 · 0:13–0:25

    The street becomes the stadium

    Matching the previous ball trajectory, the ball flies past the camera. A British elderly man in a sweater holds a cup of tea and looks down at the street; the tea trembles, he raises an eyebrow, and smiles in amusement. Shot 13: In a semi-interior space such as a pub entrance or residential doorway, begin with a handheld medium shot and push rapidly toward the feet. The ball rolls to a passerby, who reacts instantly, gives it a light juggle, and kicks it cleanly back into the street as an effortless street-football relay.

    Fixed · 15 s6 image refs
  3. Clip 03 · 0:31–0:41

    Under the bench

    The scene takes place inside a cafe. A football rolls through the interior as the boy dribbles at speed, followed loosely by approximately ten pedestrians. … Place the camera inside a park bench and film from a low angle. The boy dribbles toward the bench. The football rolls beneath it first; the boy makes a light jump around the side, then regains possession on the far side.

    Smart time · 10 s ceiling3 image refs
  4. Clip 04 · 0:41–0:56

    Knocked down

    Begin with a wide side view and transition briefly into slow motion. Owen instinctively jumps over an opposing player sliding toward him. His leg is lightly clipped in mid-air, shifting his center of gravity forward and to the side. He lands on one foot, his boot slips on the grass, and he throws out both arms to regain balance. The camera pans toward the direction of his fall with a slight delay. … The room seems to fall silent, leaving only the match audio, urgent commentary, and restrained breathing.

    Smart time · 15 s ceiling2 image refs
  5. Clip 05 · 0:56–1:05

    Between two buses

    The football is already moving between two stationary buses as an approximately eighteen-year-old girl dribbles from right to left. The bus windows reflect blurred images of her and the pursuing players. Begin by filming the action through the reflection on the side of a bus. As she approaches, move rapidly away from the reflection and transition into a waist-height lateral tracking shot parallel to her movement. A defender reaches a foot in from behind.

    Fixed · 10 s1 image ref
  6. Clip 06 · 1:05–1:17

    Back on his feet

    Use a handheld, low-angle side close-up of the boy’s legs without revealing his face. He is lying on muddy ground, then plants only his right foot and rises with determination. Show the complete process of standing up while keeping the framing limited to his legs. … The setting is a worn South American city with dense, complex buildings and a very small football court. A slight boy in worn clothing runs into frame from behind the camera, remains back-facing throughout, takes his run-up, and strikes a free kick.

    Fixed · 15 s2 image refs
  7. Clip 07 · 1:24–1:41

    Kickoff, 2026

    Edit [the reference video] as a continuous one-take sequence. Preserve the original camera path, timing, shot sizes, composition, pacing, lighting, the seamless transition through the television screen, the pub action, and all picture-in-picture content. Replace only the two principal characters: replace both the young player at the stadium and the adult man in the pub with the same person shown in [the reference image]. … Begin moments before kickoff at a football stadium in 2026.

    Smart time · 18 s ceiling1 image ref1 video ref

* Every frame generated on Dreamina-Seedance-2.5 by Zhang Weidong and BytePlus, published open-source by BytePlus in the ModelArk gallery and shown here as published. Prompts quoted verbatim from that release. Not rendered on VisionX — the same model runs on the roster here. Michael Owen appears under BytePlus’s own arrangement with him, exactly as in the released film.

04

What Seedance 2.5 actually does

CAPABILITIES

What routing a shot to Seedance 2.5 buys you over the 2.0 tiers.

One take, up to 30 seconds

A single continuous shot with no stitching and no scene-cut splicing — double the 2.0 ceiling. Long enough to carry a complete story beat, a real script, or a full product move.

Single take

30 reference images, each with a job

More than three times the nine the 2.0 tiers accept, alongside video and audio slots. Faces, wardrobe, product, set, lighting — a long take stays consistent because every element has an anchor, and each reference carries the role you name.

Reference budget

Edit a finished video by prompt

Hand it a clip and say what changes: remove everyone except the protagonist, swap the music, repair the frame. The output keeps the source clip’s aspect and length automatically.

Video editing

Extend a story forward or back

Give it a finished take and ask it to continue — the story, not just the motion. The camera can travel into a second referenced clip and keep going.

Extension

Direction by timestamp

Write the shot as a schedule — [0–10s] macro, [10–22s] pull back, [22–30s] wide — and the model blocks the take against your clock instead of improvising the pacing.

Timestamped prompts

Dialogue in 11 languages

Sound is generated in the same pass as the picture, and spoken lines lip-sync natively across languages from English and Mandarin to Arabic, Thai and Portuguese.

Multilingual speech
05

Every variant, exactly as the studio runs it

SPEC SHEET

These rows are read live from the engine registry and the pricing engine. What you see here is what the composer quotes.

ModelModesResolutionsDurationsVX
Seedance 2.5Seedance 2.5 tierT2V · I2V · V2V480p · 720p · 1080p5s / 10s / 15s / 30s16.2 VX · 5s T2V at 720p

Self-serve rates run as low as $0.085 per VX at scale. The composer quotes the exact VX for your shot, references included, before you run it.

06

How to write for Seedance 2.5

PROMPTING

Seedance 2.5 reads the task from your intent — reference, edit, or extend — and takes direction by timestamp. Being explicit about both is the whole craft.

BRAND FILM · ONE TAKE · 30S · 16:9
Use [Image1] as the bottle. Keep the label sharp through every move.
Use [Image2] for the set: wet black stone, one hard key light.

[0–10s] Macro on condensation, slow orbit around the bottle.
[10–22s] Pull back — a hand lifts it, the presenter says "This is the one."
[22–30s] Settle wide in golden-hour light, label centred, logo resolves.

Block the take against the clock

Segment the prompt with [0–10s]-style timestamps. A 30-second take without a schedule drifts; with one, the pacing is yours.

Name the task when a clip is involved

On a source video, words like "edit", "remove", "replace" or "extend", "continue" pick the task type. An editing task keeps the clip’s aspect and length automatically — that is the model’s rule, not a bug.

Anchored frames set the aspect

On a first-frame or first-and-last-frame shot, the output inherits the anchor image’s aspect ratio. Compose the anchor at the ratio you want delivered.

Give every reference a role, not a mood

"Use [Image1] as the character’s face" beats "in the style of [Image1]" — same rule as the 2.0 tiers, thirty slots instead of nine, and it matters more the longer the take.

07

What teams route to Seedance 2.5

USE CASES

The board lines that route to a 30-second single take.

Unbroken social hero films

One continuous take for vertical placements, where a cut is the thing that breaks the illusion.

Full-length product spots from a packshot

A bottle or box carried through a complete 30-second spot without the mid-way identity drift that stitching invites.

Presenters who can carry a script

A portrait held long enough for a real pitch, not a single line — with lip-sync generated against the speech, not dubbed over it.

Branded story films

A complete narrative beat in one unbroken shot rather than a sequence assembled in an editor afterwards.

Fix-it-in-the-prompt revisions

The take was right and one thing was wrong. Edit the clip by prompt — remove the extra, change the music — instead of re-rolling the whole shot.

Multi-character scenes

Thirty reference slots are what keep several cast members, their wardrobe and the set consistent across a long take.

08

Seedance 2.5 against the alternatives

COMPARISON

The short version against the two engines a long-take board line would otherwise go to. All three run on the same VisionX board, so you can settle it on your own footage.

Seedance 2.5 capability comparison against Seedance 2.0 and Kling 3.0
CapabilitySeedance 2.5Seedance 2.0Kling 3.0
Max single take30s15s15s
Reference images per call309Text, image
Timestamped shot directionYes — [0–10s] segmentsNumbered shotsNo
Edit / extend by promptYesMulti-clip fusionNo
Audio generationSame pass as the pictureSame pass as the pictureSeparate sync step
Max resolution1080pNative 4KVendor-announced 4K

VisionX rows are read from the engine registry. The Kling column describes the vendor’s published capabilities as of August 2026 — vendors ship fast, so re-check before quoting these in a deck.

FAQ

Seedance 2.5, answered

Can I use Seedance 2.5 on VisionX today?
Yes. Seedance 2.5 is live on the roster: pick it in the composer like any other engine, and the exact VX cost is quoted per shot before you run it. It sees the same Cast, references and wallet as everything else on the board.
How is Seedance 2.5 different from Seedance 2.0?
Shape of output, mostly. 2.5 is built around one continuous take up to 30 seconds, where 2.0 puts several shots inside a 15-second clip. It also takes 30 reference images per call against 2.0’s nine, takes direction by timestamp, and it can edit or extend a finished video from a plain-language prompt. The 2.0 engines remain the draft-to-master ladder — and Seedance 2.0 remains the 4K mastering engine.
The announcement said 4K at 10-bit. Where is it?
Not yet, so not on this page. ByteDance’s June keynote promised 4K at 10-bit; the release VisionX runs opened 1080p in August and has not shipped 4K — so 1080p is the number we publish, and 4K masters keep routing to Seedance 2.0, which renders native 4K today. The spec sheet above tracks the provider: it said 720p at launch, it says 1080p now, and it will say 4K the day we can actually serve it.
How do video editing and extension work?
Feed a finished clip as a reference and state the intent: "remove the background music", "replace the character with [Image1]", "extend the story forward". The model classifies the task from your wording. Edits keep the source clip’s aspect ratio and length automatically; extensions keep the aspect and take the duration you ask for.
Does Seedance 2.5 generate audio and lip-sync?
Yes — sound is generated in the same pass as the picture, so dialogue, ambience and effects are timed to the frames rather than dubbed on afterwards. Spoken lines lip-sync natively in 11 languages, including English, Mandarin, Spanish, Japanese, Korean, Arabic, Thai and Portuguese. Put the line in quotation marks in your prompt.
What does a 30-second take cost?
The composer quotes the exact VX figure for your duration, resolution and references before you commit — the same per-shot quote as every engine on the roster, and the spec sheet above shows the starting rate. A 30-second single take generally prices below stitching and re-rolling several shorter shots to get one continuous story.
Do my Seedance 2.0 prompts and Cast carry over?
Yes. It is the same model family and the same prompt shape — reference tags, quoted dialogue, shot direction all carry. Identity lives in your Cast rather than in any one model, so pointing a campaign at 2.5 is a routing choice, not a migration.

Try Seedance 2.5 on your Cast.

20 VX free on signup, no card required.