ByteDance video family · Standard tier · up to 4K 10-bit

Seedance 2.0 Video Generator

Generate 4 to 15 second clips from text or keyframes on the standard tier of the Seedance 2.0 family, the only tier in that family that reaches 4K.

Try Seedance 2.0
01

Create with Seedance 2.0

Choose text, one image, or first and last frames, then set generated audio, duration, resolution, and aspect ratio before reviewing the point estimate.

LIVE
2s
2s15s

Results

Creation history

Generated content will appear here
02 / Overview

Seedance 2.0 at a glance

Facts reviewed:

Developer
ByteDance / Volcano Engine
Official model ID
doubao-seedance-2-0-260128
Family and tier
Seedance 2.0 series · Standard
Official release date
February 12, 2026
Official clip length
4–15 seconds
Official resolution
480p, 720p, 1080p at 8-bit; 4K at 10-bit
Official frame rate
24 fps
Official container
MP4
Platform route
seedance-2.0

Official model capabilities versus this workspace

The official tables describe the model. This page runs one platform route. The two boundaries are not identical, so they are listed separately rather than blended into one list.

OFFICIAL

Documented for the official model

  • Text, image, video, and audio input in one request
  • Up to 9 images, 3 videos, and 3 audio clips as references
  • Text-to-video, first-frame image-to-video, and first-and-last-frame video
  • Multimodal reference, video editing, and video extension
  • Generated audio with two-channel output
  • Web search tool and return of the clip’s last frame
  • 24 fps, 4–15 seconds, MP4, six aspect ratios
  • 480p, 720p, and 1080p at 8-bit, plus 4K at 10-bit
  • The only tier in the Seedance 2.0 series that reaches 4K; no draft mode and no offline inference tier
LIVE PLATFORM

Exposed by this workspace today

  • Text-to-video, image-to-video with one image, and first-and-last-frame with two images
  • A 2–15 second selector driven by the current provider route
  • 480p, 720p, and 1080p output
  • Six fixed aspect ratios plus an adaptive option
  • A generated-audio switch, off by default
  • Points quoted before submission on the live estimate

Two combinations are explicitly unsupported by the official model: text plus audio, and audio on its own. 4K is not exposed on this page, so the official 4K ceiling is a model capability rather than a control here, and reference-image upload is not available in the reference mode of this workspace either.

Model brief

What is Seedance 2.0?

Seedance 2.0 is the standard tier of ByteDance’s Seedance 2.0 family. ByteDance Seed launched it on February 12, 2026, and Volcano Engine registers it in the Ark model list as doubao-seedance-2-0-260128, alongside doubao-seedance-2-0-fast-260128 and doubao-seedance-2-0-mini-260615. The official launch describes a unified multimodal audio-video joint generation architecture that accepts text, images, audio, and video, and the project page states that the model carries the broadest set of multimodal reference and editing capabilities the team has published.

The clearest documented difference inside this family is resolution, and it runs in one direction only. The official tier table records 480p, 720p, and 1080p at 8-bit plus 4K at 10-bit for the standard model. Seedance 2.0 Fast and Seedance 2.0 Mini stop at 720p, and even Seedance 2.5, the later generation, stops at 1080p. Within the Seedance 2.0 series, the standard tier is the only one that reaches 4K, and it is also the only model in the wider line whose published resolution list includes 4K at all.

The official capability list is broader than a short-clip generator, and this part is widely under-reported. The series accepts mixed modal input, with up to 9 images, 3 videos, and 3 audio clips alongside a natural-language instruction, and it supports multimodal reference, video editing, video extension, generated audio with two-channel output, a web search tool, and return of the generated clip’s last frame. Exactly two combinations are ruled out: text plus audio, and audio alone. The launch announcement frames the same capability set as replacing complex post-production and live shooting with generated footage.

Prompting follows a published method rather than guesswork. The official prompt guide gives a base formula per task type and an advanced formula that sequences eight elements: precise subject, action detail, scene environment, light and colour, camera movement, visual style, image quality, and constraints. It also fixes a small symbol vocabulary, round brackets for music, angle brackets for sound effects, and braces for dialogue, which lets one prompt separate dialogue from score from effects. Three further rules change results more than vocabulary does. Define each subject with two or three stable static traits and reuse the same label; organise a complex prompt as numbered shots in event order and do not force exact timecodes, because the model handles precise timing poorly; and specify one camera movement per shot instead of stacking push, pull, pan, and tilt in the same line.

Seedance 2.0 is not the newest model in the line, and this page does not claim that it is. ByteDance’s own model navigation currently lists Seedance 2.5 rather than Seedance 2.0, and the Ark model list places doubao-seedance-2-5-260628 ahead of the 2.0 tiers. What the same Ark list does show is that doubao-seedance-2-0-260128 is still offered and carries no end-of-life notice, unlike doubao-seedance-1-5-pro-251215. Read together, the position is specific rather than promotional: the standard 2.0 tier is the resolution ceiling of the family, and 2.5 is a separate decision about clip length and delivery quality.

This page runs the platform route seedance-2.0 rather than a direct account on Volcano Engine, and the two service boundaries differ in ways worth stating plainly. The controls here span 2 to 15 seconds, while the official endpoint documents 4 to 15 seconds. Output here is 480p, 720p, or 1080p, so the official 4K option is not available on this page. Six fixed aspect ratios plus an adaptive option are exposed, generated audio is a switch that is off by default, and reference-image upload is not offered in the reference mode, which is why the official nine-image ceiling is described above as a model capability rather than a control here.

03 / Capabilities

Why use Seedance 2.0?

Choose the standard tier when the delivery target needs the highest resolution this family can produce, or when the work needs the broadest documented capability list.

01

The only Seedance 2.0 tier that reaches 4K

The official tier table lists 4K at 10-bit for the standard model, and stops Fast and Mini at 720p. Seedance 2.5 reaches 1080p at 10-bit and no further. If the target resolution is 4K, this is the tier the official list points to.

02

Text to video

Turn a chronological production brief into a short clip covering subject, action, camera movement, lighting, and sound cues in a single instruction.

03

Image to video from a first frame

Animate one approved still while preserving its composition and directing movement, performance, timing, and ambience.

04

First-and-last-frame video

Supply two compatible images to define the opening and the destination of a continuous move. The official guide recommends this mode when the endpoints must match the supplied images exactly.

05

Multimodal reference

The official model takes up to 9 images, 3 videos, and 3 audio clips together, so a shot can inherit a character, a location, a camera rhythm, and a voice from separate references. The guide suggests four to five assets rather than filling the ceiling.

06

Video editing and video extension

Official capabilities for adding, replacing, or removing an element inside existing footage and for continuing a clip forward or backward while holding style, subject, and narrative steady.

07

Generated audio with two-channel output

Dialogue, effects, music, and ambience are generated alongside the picture, which the launch announcement describes as two-channel audio for the 2.0 family.

08

Web search tool and last-frame return

Both are documented for all three 2.0 tiers. The search tool lets a task pull external information, and the last-frame return hands back the closing frame of the generated clip as a usable still.

04 / Scene ideas

Best uses for Seedance 2.0

Choose the standard tier when the delivery target needs the highest resolution this family can produce, or when the work needs the broadest documented capability list.

Work whose specification names 4K

The standard tier is the only Seedance 2.0 model registered for 4K output. Note that this workspace currently stops at 1080p, so a 4K delivery pass has to run on the official endpoint rather than here.

Client delivery at 1080p 8-bit

Produce review-ready and delivery-ready output at the top resolution this workspace exposes, at 24 fps and up to 15 seconds per clip.

Multi-shot narrative work

Plan a compact sequence of connected shots with dialogue, ambience, and music cued by the official bracket conventions rather than stacked into one flat paragraph.

Product and brand motion

Test controlled movement, lighting transitions, sound cues, and compact storytelling around approved assets before committing to a longer production pass.

Character-led social clips

Animate an approved character or portrait for horizontal, square, or vertical campaign concepts, keeping identity and wardrobe stable across shots.

Previsualisation with a known resolution budget

Lock framing, timing, camera direction, and sound at 1080p, then decide whether the survivor needs the 4K tier, a 2.5 pass for longer clips, or a cheaper draft tier for the next round.

05 / Prompt lab

Seedance 2.0 prompt examples

Each prompt follows the official structure: named subjects, numbered shots, one camera movement per shot, and constraints at the end rather than exact timecodes. Round brackets carry music, angle brackets carry sound effects, and braces carry dialogue.

SCENE 01text to video

Two-shot dialogue scene

Uses the official brace convention for spoken lines so the model separates dialogue from effects and score.

View prompt
Shot 1: medium shot of a woman in a red coat standing under a station clock, slow push in, rain on the platform. She looks down and says {you are late}. Shot 2: cut to a wider shot as a second figure steps in from the left, steady lateral track, she looks up and says {I know}. Keep both faces stable and unchanged between shots. Warm station lighting, realistic rain texture, <distant train rumble>, (low string drone). No subtitles, no logos.
SCENE 02image to video

Portrait performance test

Checks whether an approved portrait, a subtle performance, and one camera move hold before a longer production pass.

View prompt
Preserve the uploaded person’s identity, hairstyle, wardrobe, lighting direction, and background layout. Add one natural breath, a small shift of the eyes toward camera, a brief confident smile, and a slow five-percent push in. Keep facial proportions stable and skin texture realistic. <quiet room ambience>, no added objects, no text.
SCENE 03start end to video

Opening-to-ending transition

Validates the motion path between two prepared keyframes before refining the final version.

View prompt
Connect the uploaded opening and ending frames in one continuous move. Keep the central object aligned while the camera arcs gently clockwise, the background changes from a clean studio to a nighttime storefront, and reflections appear gradually on the ground. No cuts, no shape drift, physically coherent light change, <soft city ambience>, (sparse piano).
SCENE 04text to video

Four-shot product sequence with cued sound

Tests a compact product story with a numbered camera plan and sound cued by symbol rather than by timestamp.

View prompt
Shot 1: macro side light on a matte black travel mug, condensation on the lid, locked-off camera, <water dripping>. Shot 2: a hand lifts the mug from a wooden desk, steady lateral track left, <ceramic scrape on wood>. Shot 3: match cut to the same mug on a stone ledge at dawn, slow push in, <distant wind>, (warm synth pad). Shot 4: the mug is replaced on the desk, camera holds. Keep the mug shape, colour, and lid geometry unchanged in every shot. Cinematic and restrained, no logos, no captions.
SCENE 05text to video

Single camera move study

Isolates one camera movement to learn how the tier renders geometry and motion before combining moves.

View prompt
One continuous shot, no cuts. A ceramic bowl of cherries on a linen tablecloth, locked-off camera at table height. Light enters from the left and moves slowly across the surface as the frame holds. Keep the cherry count, bowl shape, cloth folds, and colour unchanged for the full clip. <soft room tone>, no hands, no camera movement other than the light, no text.
SCENE 06text to video

Atmosphere-first ASMR close-up

Tests fine material detail and delicate effects, the area the official release highlights for immersive sound design.

View prompt
Immersive first-person close-up of a pair of hands in warm amber light. The hands move slowly and gently, pressing a sheet of bubble wrap, <bubble wrap popping>, then brushing a folded wool scarf, <wool fibres rubbing>, then tapping an acrylic panel, <light plastic tap>. No background music, no dialogue, shallow depth of field, steady camera, no text, no logos.
COMPARE

Seedance 2.0 vs Seedance 2.0 Fast vs Seedance 2.0 Mini vs Seedance 2.5

Every row comes from one of two official tables that cover all four models at once: the Ark model list and the Seedance 2.0 series guide, with token rates from the published Ark price page.

Seedance 2.0 vs Seedance 2.0 Fast vs Seedance 2.0 Mini vs Seedance 2.5
ComparisonSeedance 2.0Seedance 2.0 FastSeedance 2.0 MiniSeedance 2.5
Official model IDdoubao-seedance-2-0-260128doubao-seedance-2-0-fast-260128doubao-seedance-2-0-mini-260615doubao-seedance-2-5-260628
Official positioningHighest generation qualityBalance cost and generation speedLowest costNot stated in the 2.0 series guide
Output resolution480p, 720p, 1080p (8-bit); 4K (10-bit)480p, 720p (8-bit)480p, 720p (8-bit)480p, 720p (8-bit); 1080p (10-bit)
Frame rate24 fps24 fps24 fps24 fps
Clip duration4–15 seconds4–15 seconds4–15 seconds4–30 seconds
Aspect ratios21:9, 16:9, 4:3, 1:1, 3:4, 9:1621:9, 16:9, 4:3, 1:1, 3:4, 9:1621:9, 16:9, 4:3, 1:1, 3:4, 9:1621:9, 16:9, 4:3, 1:1, 3:4, 9:16
ContainerMP4MP4MP4MP4, MOV
Token rate, 480p/720p, no video input¥46 per million tokens¥37 per million tokens (list)¥23 per million tokens (list)¥70 per million tokens
Token rate, 1080p¥51 per million tokensNot supportedNot supported¥77 per million tokens (list)
Token rate, 4K¥26 per million tokensNot supportedNot supportedNot supported
Reference, editing, and extensionSupportedSupportedSupportedSupported
Search tool and last-frame returnSupportedSupportedSupportedSupported
Draft mode and offline inferenceNot supportedNot supportedNot supportedNot supported
LIVE PRICING

What Seedance 2.0 costs

Volcano Engine prices this tier by tokens. Without video input, 480p and 720p output is ¥46 per million tokens, 1080p is ¥51, and 4K is ¥26. Official per-clip examples for a 5-second 16:9 output list ¥2.31 at 480p, ¥4.97 at 720p, ¥12.39 at 1080p, and ¥25.27 at 4K. The 4K token rate is the lowest of the three while the per-clip example is the highest, because the two tables measure different things: Volcano Engine states that a clip’s price is set by the token rate and the token volume together, and that higher resolutions cost more per clip under otherwise equal conditions. As published on September 12, 2026, the promotional list on that price page covers Seedance 2.5, Seedance 2.0 Mini, and Seedance 2.0 Fast but not the standard Seedance 2.0 tier, so the figures here are list prices. This site charges platform points instead, so the live estimate in the workspace is the price that applies here.

Open the workspace to see the current estimate
01

Platform points

The workspace quotes points before you submit. Points are set by this site and are not the official token rate.

02

Resolution and duration

480p, 720p, and 1080p from 2 to 15 seconds on the current provider route, against an official endpoint range of 4 to 15 seconds. No 4K option is exposed here.

06 / Workflow

How to use Seedance 2.0

  1. 01

    Choose a mode

    Start from text, one approved still, or two compatible keyframes. Reference, editing, and extension are documented model capabilities rather than separate controls in this workspace.

  2. 02

    Write the prompt in the official order

    Precise subject, action detail, scene environment, light and colour, camera movement, visual style, image quality, then constraints.

  3. 03

    Define every subject explicitly

    Give each subject two or three stable static traits and reuse the same label each time it appears, so a multi-image reference does not blur into one ambiguous character.

  4. 04

    Sequence shots, and cue sound by symbol

    Number the shots in event order, keep one camera movement per shot, and mark music with round brackets, sound effects with angle brackets, and dialogue with braces instead of writing timestamps.

  5. 05

    Review at target resolution, then decide

    Compare candidates at the resolution you intend to deliver within this workspace, keep the prompt and settings of the strongest one, and move to the official 4K route only when the specification genuinely requires it.

Review before publishing

Seedance 2.0 limits, pricing, and responsible use

Generation can still produce identity drift, unstable hands or objects, unintended cuts, inaccurate on-screen text, or audio that drifts from the picture. The official launch states that the model still needs work on detail stability, realism, and dynamic liveliness, that multi-person lip synchronisation and occasional audio distortion are unresolved, and that multi-subject consistency, text reproduction, and complex editing still have room to improve. Treat every draft as review material and inspect the full clip and its complete soundtrack before publication.

Two service boundaries apply on this page and they are not interchangeable. Volcano Engine documents the official endpoint at 4 to 15 seconds with 480p, 720p, 1080p, and 4K, and with no draft mode and no offline inference tier; this workspace exposes 2 to 15 seconds through a provider route and stops at 1080p. Platform points, availability, and the preview rate can change, so the live estimate shown before submission is authoritative for this site.

Upload rules are narrower than for the video family as a whole. The official series guide states that the Seedance 2.0 models do not accept reference images or videos that contain a real human face, and points to a separate licensing workflow for portrait work. For that reason the official demonstration clips that use human subjects are described as AI-generated or licensed. Do not attempt to work around this by editing a face into a reference asset.

4K output at 10-bit is not universally playable. The official documentation records 10-bit and H.265/HEVC playback as conditionally supported on Windows browsers, and recommends Safari or Chrome on macOS and Edge or Chrome on Windows, with VLC or mpv for local playback. Because 4K is where the 10-bit path applies, a 4K deliverable should be checked on the playback target before it is handed over.

Price and availability statements here are dated. The token rates and per-clip examples come from the Ark price page reviewed on September 12, 2026, and that page’s promotional list does not include the standard Seedance 2.0 tier. If the official rates or the promotion list change, this section needs re-checking.

Upload only material you are authorised to use. Do not impersonate people, mislead viewers, violate privacy or intellectual-property rights, or omit disclosures required for synthetic media. The official announcement applies the same standard to its own demonstration material.

Sources

Sources and fact review

Official model identity, capabilities, limits, and pricing are separated from this platform’s controls, points, and availability.

07 / FAQ

Seedance 2.0 FAQ

What is the difference between Seedance 2.0 and Seedance 2.5?
Seedance 2.5 is the later generation and reaches 1080p at 10-bit, extends clips to 30 seconds, adds MOV output, and costs ¥70 per million tokens at 480p and 720p against ¥46 for the standard 2.0 tier. What 2.5 does not add is 4K: its published resolution list stops at 1080p, so the standard Seedance 2.0 tier remains the only model in the two generations whose documented output includes 4K.
Does Seedance 2.0 support 4K?
Yes. The Ark model list records 480p, 720p, and 1080p at 8-bit plus 4K at 10-bit for doubao-seedance-2-0-260128, and the series guide shows the same four tiers. Fast and Mini stop at 720p, and Seedance 2.5 stops at 1080p.
Can I generate 4K on this page?
No. This workspace exposes 480p, 720p, and 1080p only, so 4K is a capability of the official model rather than something this page can produce. The two boundaries are listed separately in the capability section above rather than merged into one list.
How much does Seedance 2.0 cost?
Without video input, the published Ark token rate is ¥46 per million tokens at 480p and 720p, ¥51 at 1080p, and ¥26 at 4K. Official per-clip examples for a 5-second 16:9 output are ¥2.31, ¥4.97, ¥12.39, and ¥25.27 respectively. With video input the 480p and 720p rate drops to ¥28 per million tokens.
Is there a promotional rate for Seedance 2.0?
Not as published. The promotional list on the Ark price page, reviewed on September 12, 2026, covers Seedance 2.5, Seedance 2.0 Mini, and Seedance 2.0 Fast, and the standard Seedance 2.0 tier is not on it. The prices quoted here are therefore list prices, and the way this site charges is platform points rather than token rates.
How many reference assets can I supply?
The official model accepts up to 9 images, 3 videos, and 3 audio clips in one request, and the guide recommends staying below the ceiling, using roughly four to five assets split across character, scene, camera-reference, and audio roles. Two combinations are unsupported: text plus audio, and audio alone. This workspace does not currently expose reference-image upload in its reference mode.
Can I use a photo of a real person as a reference?
Not directly. The official series guide states that the Seedance 2.0 models do not accept reference images or videos containing a real human face, and that portrait work goes through a separate licensing workflow. The official demonstration clips with human subjects are labelled as AI-generated or licensed for the same reason.
Can Seedance 2.0 generate audio?
Yes. Audio-video joint generation is part of the family, and the official tables mark generated audio as supported for all three 2.0 tiers. The launch announcement describes the 2.0 family as producing two-channel audio. Two combinations are unsupported: text plus audio, and audio alone.
How long can a clip be, and at what frame rate?
The official endpoint documents 4 to 15 seconds at 24 fps. This workspace exposes 2 to 15 seconds on the current provider route, and the frame rate is 24 fps, the same as every other Seedance tier including 2.5.
Do I need to write exact timestamps in the prompt?
No, and the official guide advises against it. It states that the model handles precise timings such as zero to three seconds poorly and that forcing them can cause abnormal output. Number the shots in event order and let the model set the internal rhythm.
How are dialogue, music, and sound effects written?
With three official symbols in the same prompt: braces for dialogue such as {I know}, round brackets for music such as (low string drone), and angle brackets for sound effects such as <distant train rumble>. Keeping them separate is what lets the model place each layer against the right moment in the picture.
Can Seedance 2.0 edit or extend an existing video?
Both are official capabilities listed for all three 2.0 tiers. Editing covers adding, replacing, or removing an element inside existing footage, and extension continues a clip forward or backward while holding style, subject, and narrative consistent. One phrasing detail matters: the guide says to refer to the source as the video itself rather than as a reference, so that the task is not misread as plain reference generation.

Related video models and tools

01

Seedance 2.0 Fast

Compare the cost-oriented tier: the same feature list with a 720p ceiling and a lower token rate.

02

Seedance 2.5

Compare the later generation: 1080p at 10-bit, clips up to 30 seconds, and MOV output, but no 4K.

03

Vidu Q3

Compare Pro, Turbo, Pro Fast, synchronized audio, and optional off-peak processing.

04

Text to Video

Open the general prompt-led video workspace with model selection.

Render your next scene

Create your first Seedance 2.0 video

Return to the live workspace, choose the right input mode, and review the current point estimate before generating.

Open Seedance 2.0 workspace