Reference to video

Request

{
  "model": "wan-3.0",
  "mode": "reference_to_video",
  "variant": "standard",
  "parameters": {
    "resolution": "720p",
    "duration_seconds": 8,
    "ratio": "9:16",
    "audio": true
  },
  "input": {
    "prompt": "Use the video for motion and the image for the character",
    "references": [
      {
        "media_kind": "video",
        "locator": {
          "type": "https_url",
          "url": "https://cdn.example.com/motion-reference.mp4"
        }
      },
      {
        "media_kind": "image",
        "locator": {
          "type": "https_url",
          "url": "https://cdn.example.com/character.png"
        }
      }
    ]
  }
}

Reference locators

Image, video, and audio references accept either credential-free HTTPS URLs or ready, project-owned assets. Documents and web pages remain HTTPS-only:

{
  "media_kind": "video",
  "locator": {
    "type": "media_asset",
    "media_asset_id": "7c74c900-8b68-4b0f-a337-32ed3a9a2020"
  }
}

A ready media_asset has server-verified metadata and a pinned immutable object version. In the closed alpha, Alibaba fetches caller-hosted HTTPS media directly, so that URL must stay reachable through provider processing. Client-declared duration is never a billing fact.

Reference limits

A document and web page cannot appear in the same request. Verified reference-video and reference-audio duration is limited to 15 seconds each. Until the isolated importer is enabled, a request may contain at most one caller-hosted HTTPS video and one caller-hosted HTTPS audio, without mixing either with a media asset of the same kind.