Joint Audio and Video
Sound generated with the picture rather than added after it.
Seedance 2.0 AI Video Generator
ByteDance's Seedance 2.0 produces picture and sound in one pass, from a first and last frame or from reference images, at sizes from 480P to native 4K.
- Seedance 2.0
作成を始めるには、デスクトップのブラウザでこのページを開いてください。
Sound and Picture Made Together, Not Bolted On
Seedance 2.0 is ByteDance's video model, and its defining property is that audio is not a separate stage.
Most generated video arrives silent and gets scored afterwards, which is why so much of it feels like footage with music placed over it. Seedance 2.0 generates joint audio and video, so ambience, movement sounds, and rhythm are timed against what is actually happening on screen rather than approximately matched to it later. Virse carries the model as four entries, splitting frame-driven generation from reference-driven generation and offering a Fast variant of each.
Use Seedance 2.0 when the clip has to sound like the place it depicts. Build advertising cuts, product motion, animated sequences, character-led series, and social video where the audio is part of the shot rather than a decision you postpone.
What Is Seedance 2.0?
Seedance 2.0 is an AI video generation model developed by ByteDance. It generates video with synchronised dual-channel audio, following written instruction about subjects, action, camera behaviour, pacing, and sound.
The model is built around three major strengths:
- Joint audio and video generation in a single pass
- Multimodal reference input covering images, video clips, and audio
- Native 4K output alongside lower sizes for drafting
Generations run from four to fifteen seconds, with aspect ratios spanning 21:9 through 9:16. Virse lists four entries: Seedance 2.0 and Seedance 2.0 Fast take a first frame, a last frame, or both, while Seedance 2.0 Reference and Seedance 2.0 Fast Reference work from supplied reference material instead.
Seedance 2.0 Specs at a Glance
Four to Fifteen Seconds
Single-pass generation length, extendable by continuing from a previous result.
Native 4K
True 4K output rather than an upscale, with 480P, 720P, and 1080P below it.
Multimodal References
Images, video clips, and audio can all be supplied as input.
Four Entries
Frame-driven and reference-driven paths, each with a Fast variant.
Wide Aspect Range
21:9 through 9:16 from the same written brief.
Key Features of the Seedance 2.0 AI Video Generator
Audio Timed to the Action
Because sound is generated alongside the picture, a footstep lands when the foot lands. That synchronisation is difficult to achieve by scoring a silent clip afterwards and is the main reason to choose a joint model.
References Beyond Images
Most video models accept pictures. This one also takes video clips and audio as reference material, so motion character and sonic texture can be shown rather than described.
Frame Control or Reference Control
The four entries split cleanly. Supply frames when you know what the shot opens and closes on; supply references when a subject has to stay recognisable and describing it is harder than showing it.
A Genuine Draft Tier
480P exists so that testing timing and motion does not consume the budget for the finished cut. The brief does not change when you move up.
Fast Variants for Iteration
Seedance 2.0 Fast and Fast Reference cover the lower sizes at reduced cost, which is where the exploratory rounds belong.
Stylised and Animated Motion
The model handles 2D cartoon motion, 3D animated sequences, and illustrated treatments as readily as live-action-style footage.
Build with Seedance 2.0 in Virse
Reference-driven video needs its references to exist first, and they rarely arrive as a tidy set. A character comes from one place, a product shot from another, a motion reference from a third. Virse assembles that material in the same space where the video gets made. Stills generated by any image model on the canvas can go straight into Seedance 2.0 as references or frames, without an export and re-upload in between.
Build the Reference Set on the Canvas
Generate character and product stills with an image model, then point Seedance 2.0 at them directly.
Draft Small, Deliver Large
Settle timing and motion at 480P, then re-run the identical brief at 1080P or 4K.
Access 30+ Creative Models
Move between Seedance 2.0 and 30+ other image and video models without leaving the canvas or rewriting the brief.
Keep the Series Together
Reuse the same reference set across every clip in a sequence so the whole run stays consistent.
What Can You Create with Seedance 2.0?
Advertising Cuts
Short branded sequences where the audio bed is generated with the picture.
Product Motion
Rotations, reveals, and detail passes that start and end on defined compositions.
2D and 3D Animation
Illustrated and animated sequences holding a consistent look across shots.
Character-Led Series
A run of clips in which one performer stays recognisable from shot to shot.
Social Video
Vertical and landscape cuts drafted at 480P and finished at 1080P.
Previz From Storyboards
Boards turned into moving reference before anything goes into production.
How to Use Seedance 2.0 in Virse
Choose the Entry
Frame-driven for defined compositions, reference-driven for continuity, Fast for exploration.
Supply Frames or References
Upload the opening and closing frames, or the reference material, and say what each contributes.
Write the Action and the Sound
Describe what moves, how the camera behaves, and what the clip should sound like.
Re-Run at Delivery Size
Take the approved brief up to 1080P or 4K without changing a word of it.
How to Write a Seedance 2.0 Prompt
A useful Seedance 2.0 prompt usually includes six elements:
- Subject
- Environment
- Action Sequence
- Camera
- Visual Style
- Audio Direction
こう書く代わりに
A product video of a watch, cinematic.
こう書く
Open on the watch face shown in Image 1, filling the frame under a single hard overhead light. Rotate slowly clockwise around the case as reflections travel across the polished bezel. At the halfway point, pull back to reveal the full watch resting on dark stone. End on the composition in Image 2. Keep the movement continuous with no cuts. Quiet room tone throughout, with one low metallic tone as the camera settles.
Seedance 2.0 Prompt Examples
Product Reveal, Frame to Frame
Begin on the closed perfume bottle shown in Image 1, centred against a deep burgundy backdrop. Push in slowly for the first third of the clip as the light shifts across the glass. Lift the stopper away in one continuous movement, keeping the bottle still and the camera steady. End on the composition in Image 2, with the stopper suspended above the bottle. Faint room ambience and one soft glass chime at the moment the stopper lifts.
Animated Character Shot
Animate the character from Image 1 in the illustrated style of Image 2. The character walks in from the left of a hand-painted forest clearing, stops at the centre, and looks upward as light breaks through the canopy. Hold the linework, colour palette, and proportions from the references throughout. Camera locked off. Layered forest ambience with birdsong, and a light rising musical phrase as the character looks up.
Reference-Driven Campaign Shot
Use the model from Image 1, the jacket from Image 2, and the rooftop location from Image 3. Start on a medium shot as the model turns toward the camera, then track slowly right as the city skyline comes into frame behind them. Hold the face, and the jacket's cut, colour, and texture, to what the references show. Overcast daylight. Wind, distant traffic, and a restrained electronic bed underneath.
Seedance 2.0 vs. Seedance 2.5
| 比較項目 | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
| Clip length | Four to fifteen seconds | Up to thirty seconds |
| Reference inputs | Images, video, and audio | Broader multimodal reference set |
| Story continuity | Single shots and short sequences | Multi-round extension for longer narratives |
| Editing control | Prompt-level across the clip | Timestamp-level on specific moments |
| Entries in Virse | Four, splitting frames from references | One |
| Where it fits | Defined shots at a chosen size | Longer structured narratives |
Tips for Better Seedance 2.0 Results
Write the Audio Direction
On a joint model, silence about sound is not neutral — it means the model decides. Name the mix even when it is simple.
Supply Both Frames When the Ending Matters
Constraining the closing composition removes most of the drift from a shot that has to land somewhere specific.
Give Every Reference a Stated Job
Naming what each image, clip, or audio file contributes is what stops the model averaging them together.
Draft at 480P
Settle framing, pacing, and camera movement at the lowest size and spend the delivery budget once.
Seedance 2.0 FAQ
What is Seedance 2.0?
Seedance 2.0 is a video generation model from ByteDance that produces video and synchronised audio together. Virse lists it as four entries covering frame-driven and reference-driven generation, each with a Fast variant.
How long can a Seedance 2.0 clip be?
ByteDance specifies four to fifteen seconds per generation. Longer sequences are built by continuing from a previous result.
Is Seedance 2.0 free, and what does it cost?
Seedance 2.0 is available on Virse's paid plans. Generation is billed by the second from a monthly credit allowance, and the higher plans make it unmetered within fair use. Current plan rates are listed on the Virse pricing page.
Can Seedance 2.0 generate audio?
Yes. Audio and video are produced in the same pass, with dual-channel sound timed to the action rather than laid over a finished clip.
What is the difference between Seedance 2.0 and Seedance 2.0 Reference?
The input. The standard entry works from a first and last frame; the Reference entry works from supplied reference material. Both produce the same kind of output.
What is Seedance 2.0 Fast for?
Drafting. The Fast variants cover the lower output sizes at reduced cost, which is where exploratory rounds belong before a brief is settled.
Can Seedance 2.0 produce 4K video?
Yes, natively rather than by upscaling, with 480P, 720P, and 1080P available below it.
How do I use Seedance 2.0 in Virse?
Select the entry that matches your input, upload frames or references, write a brief covering action and sound, and generate.
Start With a Frame, Finish With a Cut
Bring your opening frame, your references, and your sound direction into one workspace, and let the model handle the motion and the mix together.