← Documentation home

Quick Agent Access

Kling motion control

How to call Kling motion control

Use the authenticated video task endpoint POST /v1/video/generations with a public model that is mapped to native kling-v3 on a Kling channel. This is character motion transfer: one appearance image plus one motion-reference video. It is not text-to-video, camera movement, or a drawn mask.

{
  "model": "YOUR_PUBLIC_MODEL",
  "image": "https://YOUR_MEDIA_HOST/person.png",
  "metadata": {
    "video_url": "https://YOUR_MEDIA_HOST/motion.mp4",
    "character_orientation": "video"
  },
  "media_options": {
    "input_mode": "action_control",
    "resolution": "1080p",
    "audio_enabled": false
  }
}
  • character_orientation is required: video allows 3–30 seconds; image allows 3–10 seconds.
  • Do not send duration, seconds, aspect ratio, output size, frame rate, or voice controls. Output follows the reference media.
  • Supported resolutions are 720p and 1080p (default 720p). Omit audio to keep the reference audio; false disables it.
  • The reference video must be an HTTP(S) URL the platform can fetch. Keep that URL unchanged for the life of the task.
  • Use the public model name, then confirm the task in task logs. Prices follow the administrator media policy; this page does not publish a retail rate.