Model Rover

Seedance Video Tasks

Create Seedance text, frame, reference, edit, extension, and draft video tasks.

POST /v1/contents/generations/tasks

Creates a Volcengine-compatible Seedance task. The response id and any returned model are rewritten to platform values.

POST /v1/contents/generations/tasks
Authorization: Bearer $API_KEY
Content-Type: application/json

Request body

FieldTypeRequiredDescription
modelstringYesComplete public Seedance model ID.
contentarrayYesNon-empty text and media item list.
resolutionstringModel-dependent480p, 720p, 1080p, or 4k where supported.
ratiostringNo16:9, 4:3, 1:1, 3:4, 9:16, 21:9, or adaptive.
durationintegerNoOutput seconds or -1 for supported automatic-duration modes.
framesintegerNoOverrides duration; supported models accept 29–289 in the form 25 + 4n.
generate_audiobooleanNoSynchronized audio; defaults to true on supported models.
watermarkbooleanNoAdd the provider's AI-generated watermark; defaults to false.
output_formatmp4 or movNoOutput container; mov is supported only by applicable models.
seedintegerNoRandom seed from -1 to 2,147,483,647 on supported models.
camera_fixedbooleanNoRequest a fixed camera on supported non-reference modes.
return_last_framebooleanNoInclude a PNG last-frame URL in the terminal result.
draftbooleanNoCreate a 480p draft on supported Seedance 1.5 Pro models.
omni_reference_task_typestringNoauto, reference, edit, or extend on supported profiles.
execution_expires_afterintegerNoUpstream task timeout from 3,600–259,200 seconds; platform lifetime can end it earlier.
safety_identifierstringNoStable privacy-safe end-user identifier, up to 64 characters.
toolsarrayNoSupported tools such as { "type": "web_search" }.
callback_urlstringNoPublic terminal callback URL handled by the platform.

Content item types include:

TypeImportant fieldsPurpose
texttextPrompt or instruction.
image_urlimage_url.url, roleFirst frame, last frame, or reference image.
video_urlvideo_url.url, roleReference video for edit or extension.
audio_urlaudio_url.url, roleReference audio on supported models.
draft_taskdraft_task.idPromote a compatible succeeded draft task.

Use first_frame, last_frame, or reference_image for image roles; reference video and audio roles are reference_video and reference_audio. Seedance 2.5 can accept audio-only reference input, while Seedance 2.0 requires at least one reference image or video. Media count limits remain model-specific.

callback_url is consumed by the platform and is not forwarded as your address to the provider. Only the platform's terminal callback contract described in Tasks and Callbacks applies.

Examples

curl "http://127.0.0.1:11113/v1/contents/generations/tasks" -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" -d '{
  "model": "volcengine/seedance-2.0",
  "content": [{"type": "text", "text": "A train crossing a snowy valley"}],
  "resolution": "1080p",
  "ratio": "16:9",
  "duration": 8
}'

Mode rules

The gateway infers text-to-video, image-to-video, first/last-frame, reference, edit, extension, or draft promotion from content and omni_reference_task_type.

  • At most one first frame and one last frame are accepted; a last frame requires exactly one first frame.
  • Frame inputs cannot be mixed with reference image, video, or audio inputs.
  • A draft task cannot be mixed with media inputs.
  • Explicit edit or extension requires at least one reference video and ratio: "adaptive". Seedance 2.5 edit additionally requires a 4–30 second source and duration: -1.
  • Seedance 2.5 accepts 4–30 seconds or -1 and defaults to -1; Seedance 2.0 accepts 4–15 seconds or -1; Seedance 1.5 Pro accepts 4–12 seconds or -1. Older 1.0 variants accept 2–12 seconds.
  • Exact resolution, duration, reference-count, audio, and output-format combinations depend on the selected model profile. Unsupported capabilities fail before an upstream task is created.

priority is platform-managed and rejected. Client routing and paid service-tier controls are also rejected.

Response and result

{
  "id": "9dcf8ab2-c4b7-42be-a516-597223897b99"
}

Creation returns only the platform task ID. Status, model, output, and usage are available from the task query after the creation response.

Use either GET /v1/contents/generations/tasks/{taskId} or the common GET /v1/tasks/{taskId}. Terminal video and optional last-frame URLs are transferred to platform storage before the success snapshot is published.

Seedance additionally supports DELETE /v1/contents/generations/tasks/{taskId}. See Tasks and callbacks for lookup, cancellation, terminal deletion, and callback semantics.

Help us improve this page

Found something unclear, outdated, or incorrect?

Last updated on