Skip to main content
Seedance 2.5
Use official ByteDance model ID dreamina-seedance-2-5-260628. Keep the official JSON body and switch the video base URL and API key to TensorAI. For asset URL and AK/SK settings, see migration settings. Select Try it to create a task. After creation, Query task prefills the task ID and API key. You can also query a video task to retrieve status and results.

Create a task

HTTP 200 response:
Save the returned id to check progress and retrieve the video.

Request parameters

The JSON body limit is 64 MiB. Use JSON booleans true / false. Omitted optional values are left to the model; supplied values must meet the field and model requirements.

2.5 parameter combinations

First-frame / first-last-frame tasks and edit / extend tasks require ratio=adaptive. Editing and extension require reference_video; if duration is supplied for editing, it must be -1. Use {"type":"draft_task","draft_task":{"id":"<DRAFT_TASK_ID>"}} to reference a successful 2.5 draft task owned by the current account. Use the same model as the draft. Final-video parameters must also meet the actual model requirements, including resolution.

Media and content

Use public HTTP(S) URLs or asset://<ASSET_ID> from the current account. Images and audio also accept Base64 Data URLs (decoded limits: 30 MiB and 15 MiB). Inline video Base64 is unsupported. URLs must not target local or private addresses. For a reference image, replace the creation request’s content with:
Try it starts with one text item and one image item. Select each item’s type and fill its matching fields. Use Add item for more media, or delete items you do not need. Existing assets undergo ownership and readiness checks. See asset management for manual registration and visual verification.

Retrieve results

Try it: query a video task. Seedance 2.0 and 2.5 share the same query endpoint.
Example success response (optional metadata varies):
Only tasks created by the current account are accessible. The primary video URL is the platform’s transferred output. Read content.last_frame_url when a last frame was requested and returned, and use the returned duration for automatic-duration tasks.

Official Python SDK

This example uses the official volcenginesdkarkruntime SDK. See the official SDK installation guide.
Set PLATFORM_API_KEY to your TensorAI API key: