Submit a video generation task (OpenAI compatible)

Submit an asynchronous video generation task using the gateway's
OpenAI-compatible request shape.

The response returns a task ID. Poll GET /v1/video/generations/{task_id}
or GET /v1/videos/{task_id} for the result.

Request shape

Top-level: model, prompt, seconds, and an optional image URL.

Extension parameters (ratio, resolution, generate_audio, watermark,
content, ...) go inside the metadata object.

seconds is a string. "-1" lets the model choose the duration.

Supported scenarios

Scenario	Fields
Text to video	`model`, `prompt`
Image to video, first frame	`model`, `prompt`, `image`
Image to video, first and last frame	`model`, `prompt`, `metadata.content` with `role: first_frame` and `role: last_frame`
Multimodal reference	`model`, `prompt`, `metadata.content` with several reference items
Edit video	`model`, `prompt`, `metadata.content` with `video_url`, plus `omni_reference_task_type: edit`
Extend video	`model`, `prompt`, `metadata.content` with `video_url`, plus `omni_reference_task_type: extend`

Media references

image and every metadata.content entry take an HTTP(S) URL or a
Base64 data URI such as data:image/png;base64,.... Uploaded files are
rejected; send the reference as a value instead.

Billing

Duration drives the cost. When seconds is "-1" the model picks the length,
and the final duration is reported in the task's duration field.

application/json

Task accepted. The task is queued and status is queued.

Bodyapplication/json

{
    "id": "task_abc123def456",
    "object": "video",
    "model": "doubao-seedance-2-0-260128",
    "status": "queued",
    "progress": 0,
    "created_at": 1712345678
}

Submit a video generation task (OpenAI compatible)

请求参数

请求示例代码

返回响应