Create a video task
Seedance uses an asynchronous task model: submit a creation request to receive a task UUID, then poll the query endpoint for task status and the video URL.
This page covers Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini, and Seedance 2.5. Parameters, input constraints, and examples are documented per version; platform task IDs, default parameters, query rules, and callback behavior on this platform follow this page.
Authentication
Authorization: Required. UseBearer <YOUR_API_KEY>.Content-Type: Required. Useapplication/json.oss-id: Optional. UUID of a configured and enabled custom storage repository; only used when custom storage is enabled for the service.oss-path: Optional. Storage path prefix used together withoss-id; has no effect whenoss-idis not provided. Up to 128 characters, using letters, digits, underscores, hyphens, dots, and/..., standalone dot directories, and empty path segments are not allowed; leading and trailing/are stripped.
Models and capabilities
| Model | Model ID | Duration | Resolution | Output format |
|---|---|---|---|---|
| Seedance 2.0 | doubao-seedance-2-0-260128 | 4–15 s | 480p, 720p, 1080p, 4k | mp4 |
| Seedance 2.0 Fast | doubao-seedance-2-0-fast-260128 | 4–15 s | 480p, 720p | mp4 |
| Seedance 2.0 Mini | doubao-seedance-2-0-mini-260615 | 4–15 s | 480p, 720p | mp4 |
| Seedance 2.5 | doubao-seedance-2-5-260628 | 4–30 s | 480p, 720p, 1080p | mp4, mov |
Both versions support text-to-video, first-frame / first-last-frame-to-video, image or video reference, video editing, video extension, audio generation, and watermarks.
Seedance 2.5 additionally supports up to 50 multimodal reference assets, audio-only reference, adaptive ratio, duration: -1, and omni_reference_task_type.
Models and task types
Seedance 2.0 series
- Text-to-video: Submit a text prompt.
- First-frame-to-video: Submit 1 first-frame image, optionally with a text prompt.
- First-last-frame-to-video: Submit 1 first-frame and 1 last-frame image, optionally with a text prompt.
- Multimodal reference: Submit reference images, videos, and audio, optionally with a prompt. Audio reference must be accompanied by at least one reference image or video; audio alone is not supported.
Seedance 2.5
- Text-to-video: Submit a text prompt; explicitly specify output ratio and duration.
- Reference-to-video: Submit reference images, videos, or audio, optionally with a prompt; audio-only input is allowed. It is recommended to explicitly set
omni_reference_task_type: "reference"and specify output ratio and duration. - Draft mode: Set
draft: trueon supported input combinations to generate a480ppreview; a successful draft can be referenced viadraft_taskto create a formal video task.
Current platform scope
Call only the model capabilities activated in your console. This platform does not currently support ratio=adaptive or duration=-1, so you cannot directly use Volcano Engine's automatic ratio or smart duration examples. Seedance 2.5 first-frame, first-last-frame, video editing, and video extension tasks require adaptive and are not supported on this platform at this time. Seedance 2.0 first-frame and first-last-frame tasks can still be called with an explicit ratio. Do not specify output frame rate on any version.
Create task
POST /api/v3/contents/generations/tasks
Request fields
| Field | Type | Required | Description |
|---|---|---|---|
model | string | Yes | Model ID. seedance-2.0: standard; maps to doubao-seedance-2-0-260128. seedance-2.0-fast: fast; maps to doubao-seedance-2-0-fast-260128. seedance-2.0-mini: lightweight; maps to doubao-seedance-2-0-mini-260615. seedance-2.5: version 2.5; maps to doubao-seedance-2-5-260628. |
content | array<object> | Yes | Text prompts and reference assets |
content[].type | string | Yes | text, image_url, video_url, or audio_url |
content[].text | string | Cond. | Text prompt |
content[].image_url.url | string | Cond. | Image URL. Supports jpeg, png, webp, bmp, tiff, gif, heic, heif. Under 30 MB per image; width and height each 300–6000 px; aspect ratio 0.4–2.5. 2.0 series multimodal reference: up to 9 images; first-frame mode: 1; first-last-frame mode: 2. 2.5 reference-to-video: up to 30 images. |
content[].video_url.url | string | Cond. | Video URL. Use mp4 or mov; video encoding H.264/AVC or H.265/HEVC. Up to 200 MB per video; frame rate 24–60 fps; width and height each 300–6000 px; aspect ratio 0.4–2.5. Total pixel count between 409,600 and 8,295,044. 2.0 series: 2–15 s per video, up to 3 videos, total duration ≤ 15 s. 2.5 reference-to-video: 2–30 s per video, up to 10 videos, total duration ≤ 30 s. |
content[].audio_url.url | string | Cond. | Audio URL. Use wav or mp3; up to 15 MB per clip. 2.0 series: 2–15 s per clip, up to 3 clips, total duration ≤ 15 s. 2.5: 2–30 s per clip, up to 10 clips, total duration ≤ 30 s. |
content[].role | string | Cond. | reference_image, reference_video, reference_audio, first_frame, or last_frame |
generate_audio | boolean | No | Whether to generate audio |
resolution | string | No | 480p, 720p, 1080p; 2.0 standard also supports 4k |
ratio | string | No | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16; 2.5 also supports adaptive |
duration | integer | No | 2.0: 4–15 s, -1 not supported. 2.5 regular tasks and first-frame/first-last-frame tasks: 4–30 s or -1. 2.5 video editing must be -1; video extension supports 4–30 s or -1 |
output_format | string | No | Only configurable on 2.5; mp4 or mov. 2.0 output format is fixed to mp4 |
watermark | boolean | No | Whether to add an "AI-generated" watermark |
return_last_frame | boolean | No | Whether to return the last frame image of the video |
draft | boolean | No | Draft mode (Seedance 2.5 only) |
omni_reference_task_type | string | No | 2.5: auto, reference, edit, extend |
tools | array | No | 2.0 web search: [{"type":"web_search"}]; text-only input only |
Assets can be passed via URL, Base64, or asset URI: images support all three methods; videos support URL or asset URI; audio supports URL, Base64, or asset URI. Asset library IDs must be written as asset://<asset_id>; do not pass bare IDs. When referencing assets in prompts, use names like "image 1", "video 1", "audio 1"; numbering starts at 1 for each asset type in the request body. Do not use Asset IDs directly in prompts.
Output specifications
Standard generation requires a fixed resolution, aspect ratio, and duration. When these fields are omitted, the model's configured defaults are used; when no default is configured, they must be explicitly provided. It is recommended to fix these specs in your request.
resolution:string, output resolution. Supported specs:seedance-2.0supports480p,720p,1080p,4k;seedance-2.0-fastandseedance-2.0-minisupport480p,720p.seedance-2.5supports480p,720p,1080p; draft mode only supports480p. Asset combinations and output specs must still be supported by the selected model.ratio:string, output aspect ratio. Options:16:9,4:3,1:1,3:4,9:16,21:9.adaptiveautomatic ratio is not supported.duration:integer, output duration in seconds. 2.0 series range is 4–15 seconds; 2.5 is 4–30 seconds. This platform does not support-1smart duration.
Both 2.0 and 2.5 requests use duration; do not submit frames or specify framespersecond on your own. Do not infer frames support from frame-count conversion rules.
Optional generation parameters
When parameters are omitted, platform configuration or model default behavior may apply. Do not carry over options and defaults from one model to another.
generate_audio:boolean.truerequests audio synchronized with the video;falserequests a silent video.return_last_frame:boolean.truerequests returning the last frame URL.watermark:boolean. Whether to request a watermark.execution_expires_after:integer, task execution validity period in seconds. Model-allowed range is 3600–259200. For typical integrations, it is recommended to omit this and use platform configuration; do not treat Volcano Engine's 172800-second default as this platform's fixed default. This parameter represents allowed execution time, not generation completion time or artifact link validity.priority:integer, range 0–9. Higher values mean higher priority. Only affects ordering of pending tasks within the same inference queue; does not interrupt running tasks and does not guarantee priority across queues.callback_url:string, optional public HTTPS callback URL, up to 2048 characters.
Seedance 2.5 exclusive parameters
The following fields are for 2.5 only; omit them for the 2.0 series:
omni_reference_task_type:string, reference task type. This platform usesreferencefor reference-to-video. The officialeditandextendtypes depend on automatic ratio or smart duration and are not applicable at this time. When omitted, the model automatically determines the task type; if the prompt expresses editing or extension intent, errors may still occur after acceptance.draft:boolean.trueenables draft mode, generating a preview video to confirm scene structure, camera movement, subject action, and prompt intent.- When draft mode is enabled, only 480p Draft video is supported (other resolutions will return an error).
falsedisables draft mode and generates a normal video.
output_format:string, output container format. This platform usesmp4; MOV output is not in scope of this document.content[].draft_task.id:string, platform UUID of a successful draft that can be referenced. See draft reuse.
Fields that should not be submitted
Both 2.0 and 2.5 creation requests should omit service_tier, seed, camera_fixed, and frames. 2.0 should additionally omit draft, output_format, and omni_reference_task_type, and must not submit draft_task content elements. Even if the platform endpoint accepts a field, it does not mean the target model accepts it. Do not use empty strings in place of omission. The task list's filter.service_tier is a query filter parameter and should not be copied into the creation request body.
Null values and data types
When JSON fields are passed as null, empty strings, whitespace-only strings, or the string "null", they are treated as unset. 0, false, empty arrays, and empty objects still participate in normal validation. Booleans and numbers should use native JSON types. For example, duration uses 5, not "5"; generate_audio uses true, not "true".
Examples
Text-to-video (Seedance 2.5)
curl -X POST /api/v3/contents/generations/tasks \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "doubao-seedance-2-5-260628",
"content": [{"type": "text", "text": "A premium, highly cinematic 30-second 3D motion graphics sequence in steampunk and vintage miniature landscape style."}],
"generate_audio": true,
"ratio": "16:9",
"duration": 30,
"output_format": "mp4"
}'Multimodal reference (Seedance 2.5)
{
"model": "doubao-seedance-2-5-260628",
"content": [
{"type": "text", "text": "Maintain video 1's first-person perspective composition throughout, and use audio 1 as background music throughout."},
{"type": "image_url", "role": "reference_image", "image_url": {"url": "https://example.com/image.jpg"}},
{"type": "video_url", "role": "reference_video", "video_url": {"url": "https://example.com/video.mp4"}},
{"type": "audio_url", "role": "reference_audio", "audio_url": {"url": "https://example.com/audio.mp3"}}
],
"generate_audio": true,
"ratio": "16:9",
"duration": 11,
"omni_reference_task_type": "reference"
}First-frame / first-last-frame-to-video
First-frame tasks must include an image with role first_frame in content; first-last-frame tasks must include both first_frame and last_frame images. For 2.5, both task types must set ratio to adaptive.
{
"model": "doubao-seedance-2-5-260628",
"content": [
{"type": "text", "text": "The camera starts from the first frame scene and smoothly transitions to the last frame scene."},
{"type": "image_url", "role": "first_frame", "image_url": {"url": "https://example.com/first.jpg"}},
{"type": "image_url", "role": "last_frame", "image_url": {"url": "https://example.com/last.jpg"}}
],
"ratio": "adaptive",
"duration": 10
}Video editing and extension (Seedance 2.5)
Video editing requires the reference video to be 4–30 seconds long, and uses ratio: "adaptive", duration: -1, omni_reference_task_type: "edit". Video extension uses omni_reference_task_type: "extend", ratio must be adaptive, and duration can be 4–30 or -1.
{
"model": "doubao-seedance-2-5-260628",
"content": [
{"type": "text", "text": "Video editing: remove all people from video 1 except the protagonist."},
{"type": "video_url", "role": "reference_video", "video_url": {"url": "https://example.com/source.mov"}}
],
"generate_audio": true,
"ratio": "adaptive",
"duration": -1,
"omni_reference_task_type": "edit",
"output_format": "mov"
}Draft example (Seedance 2.5)
Drafts are for preview; resolution is fixed at 480p. The following generates a text-to-video draft with explicit ratio and duration.
{
"model": "doubao-seedance-2-5-260628",
"content": [
{
"type": "text",
"text": "At dusk on the beach, an orange cat runs slowly along the shore, camera following steadily."
}
],
"resolution": "480p",
"ratio": "16:9",
"duration": 5,
"draft": true
}Create response
{
"id": "550e8400-e29b-41d4-a716-446655440000"
}task_id is used for subsequent queries; currently the id in the creation response is the same as task_id.
Task types and parameter constraints
| Task type | Trigger condition | Parameter requirements |
|---|---|---|
| Text-to-video | Text prompt only | No special constraints on ratio or duration |
| First-frame task | Must include one first_frame image | ratio must be adaptive |
| First-last-frame task | Must include both first_frame and last_frame images | ratio must be adaptive |
| Reference-to-video | At least one reference_image, reference_video, or reference_audio | No special constraints on ratio or duration |
| Video editing | reference_video plus an editing-intent prompt | ratio: "adaptive", duration: -1; video duration 4–30 s |
| Video extension | reference_video plus an extension-intent prompt | ratio: "adaptive"; duration 4–30 or -1 |
When using multimodal reference tasks, you can specify auto, reference, edit, or extend via omni_reference_task_type. When unspecified or set to auto, the model automatically determines the task type based on the prompt and assets.
Output specifications
| Parameter | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
resolution | Standard supports 480p, 720p, 1080p, 4k; Fast/Mini support 480p, 720p | 480p, 720p, 1080p |
ratio | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | Above ratios plus adaptive |
duration | 4–15 s, -1 not supported | Regular tasks, first-frame/first-last-frame, and video extension support 4–30 s or -1; video editing must be -1 |
output_format | Fixed mp4 | mp4 or mov |
watermark | true / false | true / false |
Seedance 2.5 video editing, video extension, and first-frame/first-last-frame tasks must use ratio: "adaptive". Seedance 2.0 standard 4K output uses 10-bit/H.265 encoding, which may not be compatible with some players; Seedance 2.5 1080p also uses 10-bit/H.265. Seedance 2.5 mov is better suited for video editing and extension, using professional encoding that may not be compatible with some players. watermark: true adds an "AI-generated" label in the bottom-right corner of the video.
Input constraints
Seedance 2.0
- Images 0–9, videos 0–3, audio 0–3; maximum 15 items total.
- "Text + audio" or audio-only input is not supported.
- Single video duration 4–15 s; 2.0 standard supports 4K.
Seedance 2.5
- Up to 30 images, 10 videos, 10 audio clips; maximum 50 items total.
- Total duration of video and audio reference assets does not exceed 30 seconds; audio-only reference is supported.
- Video editing reference video duration is 4–30 s; supports 4–30 s output and
duration: -1. - Images support jpeg, png, webp, bmp, tiff, gif, heic, heif; videos support mp4, mov; audio supports wav, mp3.
- Image aspect ratio must be 0.4–2.5, width and height 300–6000 px, under 30 MB per image.
- Videos support 480p/720p, 24–60 FPS, up to 200 MB each, aspect ratio 0.4–2.5; regular reference video duration 2–30 s, editing reference video duration 4–30 s.
- Audio clips 2–30 s each, up to 15 MB each; request body size up to 64 MB. Total video and audio reference duration does not exceed 30 s.
- Task records are kept for 7 days; video URLs are valid for 24 hours; Seedance 2.5 generated video URLs can be downloaded at most 100 times.
Web search (Seedance 2.0)
Web search is only applicable to text-only input:
{
"model": "doubao-seedance-2-0-260128",
"content": [{"type": "text", "text": "Generate a city promotional video based on today's weather."}],
"tools": [{"type": "web_search"}],
"ratio": "16:9",
"duration": 11
}The actual number of searches can be viewed in the query response's usage.tool_usage.web_search.
Rate limits
Reference online inference rate limits from the official source: for 2.0 and 2.5 non-4K tasks, enterprise users have a maximum RPM of 600 and concurrency of 10; individual users have a maximum RPM of 180 and concurrency of 3. For 2.0 standard 4K tasks, both enterprise and individual users have a maximum RPM of 15 and maximum concurrency of 1. Actual account limits are subject to server-side responses.
Python polling example
import time
import requests
BASE = "https://openapi.yhsapi.com"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}
payload = {
"model": "doubao-seedance-2-5-260628",
"content": [
{
"type": "text",
"text": "An orange cat runs across a keyboard, camera following.",
}
],
"ratio": "16:9",
"duration": 5,
"generate_audio": True,
}
task = requests.post(
f"{BASE}/api/v3/contents/generations/tasks",
headers=HEADERS,
json=payload,
timeout=30,
).json()
task_id = task.get("task_id") or task["id"]
while True:
result = requests.get(f"{BASE}/api/v3/contents/generations/tasks/{task_id}", headers=HEADERS, timeout=30).json()
if result["status"] == "succeeded":
print(result["content"]["video_url"])
break
if result["status"] == "failed":
raise RuntimeError(result.get("error"))
time.sleep(10)FAQ [#faq]
Why does the task stay in queued or running for a long time? Video generation is asynchronous. Poll at approximately 10–30 second intervals; official examples use 10-second or 30-second intervals, and the actual frequency should be adjusted based on account rate limits. Read content.video_url only after the task reaches succeeded.
Why can't the video URL be accessed long-term? Video URLs are valid for 24 hours; Seedance 2.5 generated video URLs can be downloaded at most 100 times; task records are kept for 7 days. Download or transfer the video as soon as the task succeeds.
How do I choose between 2.0 and 2.5? Both versions support multimodal reference with images, videos, and audio. Use Seedance 2.0 standard when you need 4K; use Seedance 2.5 when you need up to 30 seconds per generation, up to 50 reference assets, audio-only reference, duration: -1, or mov output.
Why does a task fail asynchronously after creation? Seedance 2.5 multimodal reference tasks and first-frame/first-last-frame tasks continue to validate task type and parameters after the task starts. Video editing must use ratio: "adaptive", duration: -1, and first-frame/first-last-frame tasks must use ratio: "adaptive"; constraint violations may return InvalidParameter.TaskTypeConstraint. When the task type is known, you can use omni_reference_task_type to validate early.
How long a video does duration: -1 generate? For video editing tasks, the model tries to keep the duration the same as the source video; for other 2.5 tasks, the model automatically selects a duration within the 4–30 second range based on the prompt and assets.
Protocol reference
Volcano Ark: Create video generation task. The official page covers multiple model versions; select 2.0 or 2.5 parameters according to each field's "model support" notes. Authentication, task IDs, endpoint URLs, and extended behavior on this platform follow this document.

