Overview

Seedance 2.0 provides high-quality video generation from text prompts, with multiple modes supporting text-to-video, first/last frame control, and multi-modal omni references (images / videos / audio). This document describes the available models, pricing, modes, usage guidelines, and the syntax for referencing media in your prompts.

Currently available models

Three tiers. The Pro and Fast tiers each have a standard (strict moderation) and a less-restriction variant; the Mini tier is currently a single strict-moderation entry:

Model	Tier	Content moderation
`seedance-2`	Pro, higher quality	Strict
`seedance-2-fast`	Fast, lower cost	Strict
`seedance-2-mini`	Mini, lowest cost (720p only, no video/audio references)	Strict
`seedance-2-less-restriction`	Pro	Less restrictive
`seedance-2-fast-less-restriction`	Fast	Less restrictive

The -less-restriction variants apply a more permissive content review. They accept exactly the same input schema as the strict variants but bill at a +25% markup (see Pricing). They are not retried on content rejection — see Content Moderation below.

seedance-2-mini does not currently have a -less-restriction variant, does not accept video or audio references, and outputs only 720p. Use it when cost matters more than the higher resolutions or reference-driven modes offered by the Pro / Fast tiers.

Deprecated task types (auto-redirected)

The following legacy task types continue to be accepted for backwards compatibility and are silently redirected to the strict formal models. Please migrate to the active models above:

Deprecated task type	Now equivalent to
`seedance-2-preview`	`seedance-2`
`seedance-2-fast-preview`	`seedance-2-fast`
`seedance-2-preview-vip`	`seedance-2`
`seedance-2-fast-preview-vip`	`seedance-2-fast`

Existing clients sending the deprecated task types will continue to work without code changes. Note that the formal models default to 480p resolution; if you previously relied on the preview-vip default of 720p, pass resolution: "720p" explicitly to preserve output quality.

Pricing

Model	Resolution	Price per second
seedance-2	480p	$0.10
seedance-2	720p	$0.20
seedance-2	1080p	$0.50
seedance-2-fast	480p	$0.08
seedance-2-fast	720p	$0.16
seedance-2-mini	720p	$0.15
seedance-2-less-restriction	480p	$0.11
seedance-2-less-restriction	720p	$0.22
seedance-2-less-restriction	1080p	$0.55
seedance-2-fast-less-restriction	480p	$0.088
seedance-2-fast-less-restriction	720p	$0.176

The -less-restriction variants are priced at a +10% markup over the matching strict tier and resolution.

seedance-2-fast / seedance-2-fast-less-restriction do not support 1080p output.

seedance-2-mini outputs only 720p; 480p and 1080p requests are rejected. Mini does not accept input videos, so the cost formula collapses to unit_price × output_duration.

Cost calculation:

Without input videos: credits = unit_price × output_duration

With input videos: credits = unit_price × output_duration + (unit_price / 2) × total_input_video_duration

Duration Limits

Maximum output duration: 15 seconds (API enforces integer 4–15).

Maximum total input video duration: 15.4 seconds (sum of all reference videos).

Modes

The mode parameter determines how references are used and which parameters take effect.

mode is optional — when omitted, it is automatically inferred from the input:

No references → text_to_video

1–2 images only (no videos, no audios) → first_last_frames

Any other combination of references → omni_reference

You may also pass mode explicitly to enforce strict validation.

text_to_video

Pure text-to-video generation. No references accepted.

duration: user-controlled (4–15s)

aspect_ratio: user-controlled (6 options)

first_last_frames

Use 1–2 images as the first and/or last frame of the generated video.

image_urls: 1–2 images required. Only images accepted — video and audio URLs are rejected.

duration: user-controlled (4–15s)

aspect_ratio: user-controlled (6 options), or auto to derive from the first image

Supported image formats: jpg, jpeg, png, webp, bmp

omni_reference

Use 1–9 mixed references (images, videos, audio) for flexible multi-modal generation.

image_urls: optional, images as style/content references

video_urls: optional, videos as motion/style references (formats: mp4, mov)

audio_urls: optional, audio as soundtrack (mp3/wav only, must be ≤15 seconds total)

duration: user-controlled (4–15s)

aspect_ratio: user-controlled (6 options)

Audio-only is not allowed — at least one image or video reference must be provided alongside audio.

Total references (images + videos + audios) must be between 1 and 9.

Not available on seedance-2-mini — Mini supports only text_to_video and first_last_frames; any request containing video_urls or audio_urls is rejected.

Important Notes & Special Cases

Parameter behavior by mode

Parameter	text_to_video	first_last_frames	omni_reference
duration	User input (4–15)	User input (4–15)	User input (4–15)
aspect_ratio	User input (6 options)	User input or `auto` (from image)	User input (6 options)
image_urls	Not allowed	1–2 required	optional (shared 9 max)
video_urls	Not allowed	Not allowed	optional (shared 9 max)
audio_urls	Not allowed	Not allowed	optional (shared 9 max)

On seedance-2-mini, the omni_reference column does not apply — Mini only handles text_to_video and first_last_frames.

Aspect ratio

Supported values: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. The value auto is only available in first_last_frames mode (derives from the first image).

Default: 16:9 (auto becomes 16:9 if no image is provided).

Duration

Range: 4 to 15 seconds (any integer).

Default: 5 seconds (when omitted or out of range).

Resolution

Supported values: 480p (default), 720p, 1080p.

1080p is not supported on the seedance-2-fast / seedance-2-fast-less-restriction tiers.

seedance-2-mini supports only 720p. The resolution defaults to 720p when omitted; passing any other value returns a 4xx error (the rejection echoes the offending value).

Audio constraints

Audio references are only accepted in omni_reference mode.

Supported formats: mp3, wav only.

Maximum total audio duration: 15 seconds. The upstream silently rejects longer audio (the upload will time out with no explicit error message); the API pre-validates duration to surface this early.

Audio-only submissions are rejected — at least one image or video must accompany any audio references.

Content moderation

All models apply content moderation. Submissions found to contain disallowed content are rejected and your credits are refunded.

seedance-2 / seedance-2-fast (strict): on a content rejection, the task may be retried once via an alternate review path that typically resolves prompts flagged for style / advertising / character intent.

seedance-2-less-restriction / seedance-2-fast-less-restriction: the primary review is itself less restrictive, and there is no further retry on rejection — a rejection is final, your credits are refunded, and the task ends with a content-violation error.

seedance-2-mini: single review path with no retry on rejection — a content rejection refunds credits and ends the task.

Peak hours

From 09:00 to 15:00 GMT, Seedance experiences high traffic. Queue times may extend to several hours during this period.

Using References in Prompts

When you include references, you can mention them in the prompt using @-syntax.

Syntax

@image1, @image2, … — reference images by position in image_urls (1-based)

@video1, @video2, … — reference videos by position in video_urls (1-based)

@audio1, @audio2, … — reference audios by position in audio_urls (1-based)

Chinese format also supported: @图片1, @视频1, @音频1 (with or without _ between the type and the number)

Examples

Single image

The cat in @image1 walks through a garden

Image + video + audio

Apply @image1 artistic style with @video1 motion and @audio1 soundtrack

Morphing

@image1 transforms into @image2

Validation

If your prompt references a material that was not provided (e.g. @image3 but only 2 images supplied), the request is rejected with a descriptive error message. The same applies to @videoN and @audioN.

Using Private Assets

The Private Asset Library lets you reference uploaded characters, products, scenes, or audio across tasks instead of re-fetching the same URL every time. It also unlocks a one-call submission flow where the service uploads your inputs for you.

Available on seedance-2-less-restriction and seedance-2-fast-less-restriction only. Strict variants do not accept asset references — they will reject the request with HTTP 422.

Two modes

Mode	How you reference inputs	When to use
Manual	Upload to the library once (see Private Asset Library guide), then pass `asset://<asset_id>` strings in `image_urls` / `video_urls` / `audio_urls`.	A stable cast / product library reused across many tasks. The asset is pre-vetted once and referenced for free thereafter.
Auto-upload	Set `auto_upload_assets: true` and pass raw URLs as you would normally. The service uploads each non-`asset://` URL as an ephemeral asset, runs the task, and cleans up afterwards.	Each task uses fresh references (on-demand user content, throwaway avatars). No manual upload / delete bookkeeping.

Both modes can be mixed in the same request. Explicit asset:// items pass through untouched while raw URLs are auto-ingested.

Quick reference — auto-upload

{
  "model": "seedance",
  "task_type": "seedance-2-fast-less-restriction",
  "input": {
    "prompt": "A cinematic 5-second scene blending Image 1 and Image 2",
    "image_urls": [
      "https://your-cdn.com/character-a.png",
      "https://your-cdn.com/character-b.png"
    ],
    "auto_upload_assets": true,
    "asset_retention_hours": 3,
    "aspect_ratio": "16:9",
    "duration": 5,
    "resolution": "720p"
  }
}

The task starts with a brief ingestion phase (typically 30–90 s for the first reference) before generation begins.

Quick reference — manual + mixed

{
  "model": "seedance",
  "task_type": "seedance-2-fast-less-restriction",
  "input": {
    "prompt": "Image 1 walks through Image 2 at sunset",
    "image_urls": [
      "asset://asset-20260608154633-rv7g2",
      "https://your-cdn.com/scene-today.png"
    ],
    "auto_upload_assets": true,
    "aspect_ratio": "16:9",
    "duration": 5
  }
}

The asset:// item is referenced as-is; the raw URL becomes an ephemeral asset for this task only.

Things to know

The two ephemeral parameters (auto_upload_assets, asset_retention_hours) are ignored on strict task types — strict requests with raw URLs continue to behave exactly as before.

Strict task types reject any asset:// reference (HTTP 422). Switch to a -less-restriction variant if you need asset references.

Ephemeral assets do not consume your manual per-plan quota — they use a separate per-plan concurrency cap. See the Private Asset Library guide for the full lifecycle, quotas, and error codes.

Tips

Use publicly accessible URLs (e.g. hosted on a CDN or cloud storage). Signed / expiring URLs may fail.

Keep input videos short (5–15 seconds) for best results.

Describe the desired output clearly in your prompt.

application/json

Bodyapplication/json

Examples

Task created (pending)Completed

{
    "code": 200,
    "data": {
        "task_id": "07ed43de-465d-43b1-b62c-44249b14d818",
        "model": "seedance",
        "task_type": "seedance-2-fast",
        "status": "pending",
        "config": {
            "service_mode": "",
            "webhook_config": {
                "endpoint": "",
                "secret": ""
            }
        },
        "input": {
            "prompt": "A woman sings and strums her guitar",
            "mode": "text_to_video",
            "duration": 5,
            "aspect_ratio": "16:9"
        },
        "output": null,
        "meta": {
            "created_at": "2026-04-05T13:33:26.10346684Z",
            "started_at": "0001-01-01T00:00:00Z",
            "ended_at": "0001-01-01T00:00:00Z",
            "usage": {
                "type": "point",
                "frozen": 4000000,
                "consume": 0
            },
            "is_using_private_pool": false
        },
        "detail": null,
        "logs": [],
        "error": {
            "code": 0,
            "raw_message": "",
            "message": "",
            "detail": null
        }
    },
    "message": "success"
}

Seedance 2

Overview

Currently available models

Deprecated task types (auto-redirected)

Pricing

Duration Limits

Modes

text_to_video

first_last_frames

omni_reference

Important Notes & Special Cases

Parameter behavior by mode

Aspect ratio

Duration

Resolution

Audio constraints

Content moderation

Peak hours

Using References in Prompts

Syntax

Examples

Single image

Image + video + audio

Morphing

Validation

Using Private Assets

Two modes

Quick reference — auto-upload

Quick reference — manual + mixed

Things to know

Tips

Request

Request Code Samples

Responses

Seedance 2

Overview#

Currently available models#

Deprecated task types (auto-redirected)#

Pricing#

Duration Limits#

Modes#

text_to_video#

first_last_frames#

omni_reference#

Important Notes & Special Cases#

Parameter behavior by mode#

Aspect ratio#

Duration#

Resolution#

Audio constraints#

Content moderation#

Peak hours#

Using References in Prompts#

Syntax#

Examples#

Single image#

Image + video + audio#

Morphing#

Validation#

Using Private Assets#

Two modes#

Quick reference — auto-upload#

Quick reference — manual + mixed#

Things to know#

Tips#

Request

Request Code Samples

Responses

Overview

Currently available models

Deprecated task types (auto-redirected)

Pricing

Duration Limits

Modes

text_to_video

first_last_frames

omni_reference

Important Notes & Special Cases

Parameter behavior by mode

Aspect ratio

Duration

Resolution

Audio constraints

Content moderation

Peak hours

Using References in Prompts

Syntax

Examples

Single image

Image + video + audio

Morphing

Validation

Using Private Assets

Two modes

Quick reference — auto-upload

Quick reference — manual + mixed

Things to know

Tips