Video API

Overview

The Video API submits video-generation requests task-style, sharing the same submit endpoint as the Image API: POST /api/generate/submit. The difference is in the model field and the input fields (video additionally accepts duration / seconds). Submit returns a task_id immediately — poll GET /api/generate/status/{task_id} until status=finished.

Authentication

Same Bearer token as the AI API: Authorization: Bearer sk-xxxxxx.

Shared Request Skeleton

{
  "model": "sora-2 | sora-2-pro | veo-3.0-generate-001 | veo-3.1-generate-preview | ...",
  "input": {
    "prompt": "video description",
    "size": "WxH pixels, e.g. 1280x720 / 1920x1080 / 720x1280",
    "duration": 4,
    "seconds": "4",
    "image": "first-frame reference URL (optional)",
    "images": ["first-frame reference URLs (optional)"],
    "metadata": { /* model-specific params */ }
  }
}

Only prompt / size / image / images / duration / seconds are allowed at the top of input; any other field (aspectRatio, resolution, negativePrompt, etc.) must go under input.metadata or the request returns 400.

file_type is currently always `image`

In the success response, data.files[].file_type is hardcoded to "image" regardless of the media type. Detect video vs image from the file_url extension or your call-site context.

Overview

Shared Request Skeleton

Models

sora-2

veo-3

On this page