New modelSeedance 2.5 · One-take audio-video model

Seedance 2.530 seconds of story, generated in one take

Seedance 2.5 is a flagship audio-video model. It writes picture, motion and sound in the same pass — so a full scene with a setup, a turn and a payoff arrives as one continuous clip instead of a pile of stitched fragments.

Up to 30s in one passNative synced audio1080p output50 multimodal referencesMulti-shot in one clipNo install, runs in browser
Images
Add end frame
0/1
Return Last Frame
0/5000
5s
What is Seedance 2.5

A production model, not a clip toy

Seedance 2.5 treats a clip as finished work, not a fragment. Text, images, video and audio go in as direction — a graded, scored, lip-synced clip comes out.

The headline capability is length: one continuous 30-second take, long enough for a scene to open, develop and resolve without identity or lighting drift. Bring up to 50 references to lock the brief, then fix a single detail with region-level editing instead of regenerating everything around it.

One continuous take

30 seconds in a single pass, extendable in further rounds.

Everything in one pass

Picture, dialogue, effects and score arrive already in sync.

Revise, don't restart

Region-level edits change one element, not the whole frame.

Capabilities at a glance

Clip duration
4–30 seconds
Single-pass generation
Full 30s in one context
Resolution
480p · 720p · 1080p
Frame rate
24 fps
Aspect ratios
16:9 · 9:16 · 4:3 · 3:4 · 21:9 · 1:1
Audio
Native stereo, generated in-pass
Lip sync
Phoneme-level, 8 languages
References
30 images · 10 videos · 10 audio
Modes
Text-to-video · Image-to-video · Reference-to-video
Multi-shot
Natural-language shot labeling
Editing
Region-level · Extension

Options reflect what Vioart supports today.

Features

What Seedance 2.5 does differently

Six capabilities that change how a shot gets made — each one removing a step you used to do by hand in post.

01One-take storytelling

A full 30-second scene, generated as one continuous shot

Ask for a story and you get a story. Inside 30 seconds Seedance 2.5 organises multiple logically connected beats — setup, development, turn, resolution — instead of stretching one moment thin. Because it never stitches internally, there is no join where the lighting jumps or the face quietly changes.

  • Stable character identity and lighting across the full window
  • Camera logic that carries from the first frame to the last
  • Multi-shot cuts inside a single generation via natural-language shot labels
Capability30s single pass
Sample output

30s · 21:9 · one pass

02Multimodal referencing

Up to 50 references, all read together before the first frame

Bring 30 images, 10 video clips and 10 audio files into one request. Character sheets and wardrobe stills lock identity, reference footage carries camera language and pacing, and audio clips set voice character and musical tone. Everything is reasoned over jointly — not mixed in afterwards.

  • Images for identity, costume, environment, props and visual style
  • Video for camera movement, motion rhythm and pacing
  • Audio for voice character, ambience and score direction
Capability30 + 10 + 10 assets
Sample output

Reference-to-video · character locked

03Native audio

Dialogue, effects, ambience and score written with the picture

Audio is not a layer added later. Seedance 2.5 generates a stereo mix in the same pass as the image, so footsteps land on the footstep, room tone matches the room, and phoneme-level lip sync holds across eight languages. What you download is already a finished cut.

  • Dialogue with phoneme-level lip sync in 8 languages
  • Sound effects and ambience tied to on-screen action
  • Background score that follows the emotional arc of the shot
CapabilityIn-pass stereo mix
Sample output

Sound on · generated in-pass

04Precision editing

Change one detail without rebuilding the shot

When a take is right except for one thing, target that thing. Region-level editing swaps a prop, a garment or a background element while camera movement, lighting direction and motion continuity stay untouched.

  • Region-level edits that preserve the surrounding frame
  • Targeted changes to props, garments or backgrounds
  • Reference-based editing to adopt the treatment of another clip
CapabilityRevise, not regenerate
Sample output

Region edit · rest of frame preserved

05Camera direction

Direct the lens instead of describing it and hoping

Set camera angle, height and orientation as parameters. Prompt adherence is roughly 20% better than the previous generation, so the shot you asked for is the shot you get more often.

  • Camera perspective control for predictable composition
  • Prompt-guided camera motion with stronger instruction adherence
  • First and last frame anchoring in image-to-video mode
Capability~20% better adherence
Sample output

Prompt-guided camera move

06Beyond 30 seconds

Keep the story moving beyond a single generation

For anything longer than one generation, extend forward or backward in multiple rounds. Seedance 2.5 carries visual and audio continuity across each boundary, helping every new segment feel like part of the same take.

  • Multi-round extension, forward or backward from your clip
  • Visual continuity held across extension boundaries
  • Audio continuity maintained as the story grows
CapabilityMulti-round extension
Sample output

Extended take · continuity held

How to use

From prompt to finished cut in four steps

No timeline, no plugins, no render queue. Everything happens in the generator above.

  1. 1

    Open the Seedance 2.5 studio

    Choose Seedance 2.5 in the model picker and pick your mode — start from text, or upload an image to anchor the first frame.

    Text-to-video or image-to-video
  2. 2

    Write the scene and add references

    Describe the shots in order, name the subject, the motion, the light and the sound. Then attach up to 50 reference images, clips and audio files to lock identity, camera style and tone.

    Up to 50 reference assets
  3. 3

    Set duration, ratio and audio

    Pick anything from 4 to 30 seconds, choose from six aspect ratios, step up to 1080p when the clip is going out for real, and leave audio on so dialogue and score are generated with the picture.

    4–30s · up to 1080p
  4. 4

    Generate, then refine or extend

    Review the take. Fix a single element with a region edit, push the story further with an extension round, or download the finished clip with its audio already mixed.

    Edit · extend · download

Prompts that get better takes

  • Label your shots — "Shot 1: wide, dusk. Shot 2: close on hands." — so cuts land where you want them.
  • Curate references for coherence, not volume: five aligned images beat fifty conflicting ones.
  • Write the sound as deliberately as the picture: name the dialogue, the ambience and the score.
  • Direct the camera explicitly — lens, height, movement — rather than leaving composition to chance.
  • Iterate in small edits; change one variable per pass so you can tell what actually helped.
Use cases

Where a 30-second take changes the job

The work that used to need an edit bay, a sound pass and three rounds of stitching.

Advertising

Campaign spots and brand films

A full spot with an opening, a product moment and a closing beat generated and reviewed as one unit — then corrected with region edits instead of a reshoot.

Commerce

Product launches and e-commerce

Hold the exact product across every variant with reference images, then generate vertical, square and cinematic cuts of the same scene for each channel.

Film

Storyboarding and short-form narrative

Map out a scene with clear shot direction and visual references, then turn the sequence into a polished, scored concept before anyone books a crew.

Music

Performance and music video

Reference audio sets the voice and the tone, phoneme-level lip sync keeps the vocal on the mouth, and one take can carry a whole verse.

Social

Always-on social content

Turn one creative direction into a batch of on-brand clips across six aspect ratios, with audio finished in-pass so nothing waits on an editor.

Long-form

Extended stories and episodic content

Continue a strong clip forward or backward across multiple rounds while keeping the look, motion and sound consistent from one segment to the next.

Testimonials

Loved by creators around the world

Filmmakers, brand storytellers and solo creators across every timezone on what changed once a whole scene fit into one generation.

4.9 / 5 average rating from creators generating with Seedance on Vioart
The stitch problem is just gone. We used to burn a day matching lighting between two 15-second clips — now the whole spot comes out of one generation with the grade already consistent.
Mara WhitfieldMara WhitfieldCreative Director · Independent studio
Fifty references sounds like marketing until you have a brief with a fixed character, a lookbook wardrobe and a camera style pulled from a reference reel. All of it fits in one call now.
Diego SalasDiego SalasBrand Content Lead · DTC skincare
Region editing changed our review cycle more than the length did. Client wants a different jacket? We change the jacket. Everything else stays exactly as approved.
Priya RamanPriya RamanPost Supervisor · Agency in-house
I stopped opening my audio editor. The dialogue lands on the mouth, the room tone matches the room, and the score already follows the beat I wrote in the prompt.
Tobias LenzTobias LenzSolo Filmmaker · YouTube documentary
We define the shots with clear prompts and visual references. What comes back is polished enough that we now use it to pitch the direction.
Hana OkabeHana OkabeStoryboard Artist · Commercial production
Six aspect ratios from the same direction means one approval covers every placement. That alone cut a week out of our launch calendar.
Elena NovakElena NovakHead of Social · Consumer electronics
FAQ

Seedance 2.5, answered

The questions creators ask before their first generation.

Start creating

Your next scene is one prompt away

Thirty seconds of story, sound included, generated in a single take. Open the Seedance 2.5 studio and see what one pass gets you.

Runs in your browser · no install · credits scale with duration and resolution