Video generation

---
title: Video generation
description: Create a Seedance task, poll its status, and download the completed video.
lastUpdated: 2026-07-27
---

Video generation is asynchronous. Create a task, poll its status, and download content only after it reaches `completed`.

## Endpoints

| Action | Method and path |
|---|---|
| Create | `POST /v1/video/generations` |
| Status | `GET /v1/video/generations/{task_id}` |
| Download | `GET /v1/videos/{task_id}/content` |

A create request requires `model` and `prompt`. Support for `duration`, `seconds`, `image(s)`, `size`, and `metadata` depends on the selected model.

```bash
curl https://api.nexinfer.com/v1/video/generations \
  -H "Authorization: Bearer $NEXINFER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "prompt": "A calm ocean at sunrise",
    "duration": 4,
    "metadata": {"resolution": "720P"}
  }'
```

The returned `id` is the public NexInfer task ID. States include `queued`, `in_progress`, `completed`, and `failed`. Use bounded polling and never retry a failed paid task indefinitely.

## Network interruption and settlement

- Once a task ID is returned, a client disconnect does not cancel the asynchronous task. Resume polling the same ID.
- If the create connection ends before the client receives an ID, the outcome is unknown. Keep the request ID and check usage/support before submitting the same paid job again.
- Successful tasks settle from the model provider's reported `total_tokens`. A terminal `failed` task or platform task timeout refunds its pre-consumed quota.
- A local polling timeout or temporary network failure is not itself a failed model task and does not trigger a refund; keep polling within your operational limit.

Video input must be placed in `metadata.content` with `role: "reference_video"`:

```json
{
  "model": "doubao-seedance-2-0-260128",
  "prompt": "Continue this video naturally",
  "duration": 4,
  "metadata": {
    "resolution": "720P",
    "content": [{
      "type": "video_url",
      "role": "reference_video",
      "video_url": {"url": "https://example.com/input.mp4"}
    }]
  }
}
```

:::caution[Current verified boundary]
Valid MP4 output and six-tier token billing have been verified for 480P/720P/1080P/4K without video input and 720P/1080P/4K with video input. A requested duration of 4 seconds currently produces about 5 seconds. Mixed image/audio references, video edits, and extensions remain unverified.
:::

Video generation is asynchronous. Create a task, poll its status, and download content only after it reaches completed.

Endpoints

Action	Method and path
Create	`POST /v1/video/generations`
Status	`GET /v1/video/generations/{task_id}`
Download	`GET /v1/videos/{task_id}/content`

A create request requires model and prompt. Support for duration, seconds, image(s), size, and metadata depends on the selected model.

curl https://api.nexinfer.com/v1/video/generations \
  -H "Authorization: Bearer $NEXINFER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "prompt": "A calm ocean at sunrise",
    "duration": 4,
    "metadata": {"resolution": "720P"}
  }'

The returned id is the public NexInfer task ID. States include queued, in_progress, completed, and failed. Use bounded polling and never retry a failed paid task indefinitely.

Network interruption and settlement

Once a task ID is returned, a client disconnect does not cancel the asynchronous task. Resume polling the same ID.
If the create connection ends before the client receives an ID, the outcome is unknown. Keep the request ID and check usage/support before submitting the same paid job again.
Successful tasks settle from the model provider’s reported total_tokens. A terminal failed task or platform task timeout refunds its pre-consumed quota.
A local polling timeout or temporary network failure is not itself a failed model task and does not trigger a refund; keep polling within your operational limit.

Video input must be placed in metadata.content with role: "reference_video":

{
  "model": "doubao-seedance-2-0-260128",
  "prompt": "Continue this video naturally",
  "duration": 4,
  "metadata": {
    "resolution": "720P",
    "content": [{
      "type": "video_url",
      "role": "reference_video",
      "video_url": {"url": "https://example.com/input.mp4"}
    }]
  }
}