Video Generation

Video generation is asynchronous: submit a task to get a task_id, poll the task status, then read the video URL after completion.

Per-Model Pages

This page covers the shared submit, fetch, and polling flow. For each video family's capability matrix, request fields, per-capability examples, and parameter tables, see the dedicated pages:

• SeeDance Video Generation — Volcengine Ark Doubao Seedance (text/image-to-video, first/last frame, video reference/continuation)
• HappyHorse Video Generation — Alibaba Cloud Bailian DashScope (text/image/reference-to-video, video editing)

Endpoints

For new Seedance/Doubao integrations, prefer /v1/video/generations.

Request Body

POST /v1/video/generations

Correct and Incorrect Examples

Incorrect: sending the upstream top-level content[] body directly to TENSORAXIS.

{
  "model": "doubao-seedance-2-0-260128",
  "content": [
    {
      "type": "text",
      "text": "An old man wearing a hat smiles and walks forward"
    }
  ]
}

Correct: use TENSORAXIS request fields and let the relay translate them.

{
  "model": "doubao-seedance-2-0-260128",
  "prompt": "An old man wearing a hat smiles and walks forward",
  "images": ["https://example.com/reference.jpg"],
  "metadata": {
    "resolution": "1080p",
    "ratio": "16:9",
    "duration": 5,
    "camera_fixed": true,
    "watermark": false
  }
}

Submit a Task

curl https://api.tensoraxis.ai/v1/video/generations \
  -H "Authorization: Bearer $TENSORAXIS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "prompt": "An old man wearing a hat smiles and walks forward",
    "images": ["https://example.com/reference.jpg"],
    "metadata": {
      "resolution": "1080p",
      "ratio": "16:9",
      "duration": 5,
      "camera_fixed": true,
      "watermark": false
    }
  }'

A successful submit response returns a public task ID. Fields can vary slightly by video channel, but usually include:

{
  "id": "task_xxxxx",
  "task_id": "task_xxxxx",
  "object": "video",
  "model": "doubao-seedance-2-0-260128",
  "status": "queued",
  "progress": 0,
  "created_at": 1760000000
}

Save id or task_id for polling.

Fetch a Task

If you submit through the common video task endpoint, poll with:

curl https://api.tensoraxis.ai/v1/video/generations/task_xxxxx \
  -H "Authorization: Bearer $TENSORAXIS_API_KEY"

The common fetch response uses the task wrapper shape:

{
  "code": "success",
  "message": "",
  "data": {
    "task_id": "task_xxxxx",
    "status": "SUCCESS",
    "progress": "100%",
    "result_url": "https://example.com/video.mp4",
    "fail_reason": ""
  }
}

Common status meanings:

The OpenAI/Sora-style fetch path is:

curl https://api.tensoraxis.ai/v1/videos/task_xxxxx \
  -H "Authorization: Bearer $TENSORAXIS_API_KEY"

That path returns an object: "video" response and, on success, exposes the video URL as url, video_url, and metadata.url.

Seedance/Doubao metadata

For Doubao/Seedance channels, TENSORAXIS converts prompt, images, and metadata into the Volcengine content generation task shape:

Notes:

• metadata.model is removed and cannot override the billed model.
• metadata fields that do not map to the current adaptor structure are usually not forwarded upstream.
• metadata.content is an advanced internal compatibility field and may override the content list generated from images; public integrations should not use it.
• Valid values, enums, and actual upstream behavior are defined by the official Volcengine documentation for the selected model. This page only documents how the current TENSORAXIS relay receives and forwards fields.

Polling Recommendations

• Wait 2-5 seconds before the first poll after submission.
• Poll every 5-10 seconds while the task is generating.
• Do not use high-frequency polling as a replacement for callbacks; queue large batches in your application.
• If you receive 429, reduce concurrency and retry with exponential backoff.