Kling Video Generation

curl -X POST https://zx1.deepwl.net/v1/videos \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '
{
  "model": "Kling-3.0-Omni",
  "prompt": "Cyberpunk city night scene, the camera slowly pushes in",
  "seconds": "5",
  "metadata": {
    "output_config": {
      "resolution": "720P",
      "aspect_ratio": "16:9"
    }
  }
}'

{
  "id": "video_123",
  "task_id": "video_123",
  "object": "video",
  "model": "kling-3.0-omni-720p-ref-audio",
  "status": "queued",
  "progress": 0,
  "created_at": 1712697600
}

Kling uses POST /v1/videos to submit asynchronous video tasks in JSON. It supports text-to-video, image-to-video, multiple reference images, first-and-last-frame video, motion control, digital avatars, lip sync, and more. Extended parameters (output config, scene type, reference video, etc.) are passed through the metadata field. After a successful submission, the task id and status are returned; poll for results via Task Status Query.

Supported Models

Pass a base model name in the model field:

Per the gateway spec: the Kling duration tiers are 5 / 10 (default 5); Kling-3.0-Omni additionally supports a 15-second vertical long video (see request examples). Kling-O3 and Kling-Mini are not yet preset in the docs and billing mapping.

Composite billing model names (can be passed directly as model; the gateway restores the upstream ModelName/ModelVersion and fills in the parameters):

• kling-3.0-omni-1080p-ref-audio: 3.0-Omni · 1080P · ref=with reference input (noref=no reference) · audio=with sound (mute=silent)
• kling-2.6-motion-pro-1080p: 2.6 motion control · tier pro (std/pro) · 1080P
• kling-avatar-720p: digital avatar scene · 720P
• kling-identify-face: lip sync scene (billed at 5 seconds if shorter)

Request Headers

Request Body

Parameter Rules

• Duration priority: top-level seconds > top-level duration > metadata.seconds / metadata.duration / metadata.video_duration > default 5
• Resolution priority: metadata.output_config.resolution > top-level size > model default 720P
• Default aspect ratio for text-to-video is 16:9; size shortcuts: 720x1280 → 720P + 9:16, 1792x1024 → 1080P + 7:4
• scene_type=motion_control requires a video reference (metadata.video_url or a file_infos entry with Category=Video); an image-only request fails with videoUrl must not be blank
• file_infos is limited to 3 items; Type=Url must include Url, Type=File must include FileId
• Top-level image / images do not support base64 data URIs
• scene_type=lip_sync is billed at 5 seconds when shorter than 5 seconds
• In a real request, ext_info must be a JSON string: first build the AdditionalParameters object, serialize it to a string, then submit it as the ExtInfo string (double encoding). The examples below show its structure as a readable object

Request Examples

Complete call examples for each scenario; switch between them via the tabs. In the ExtInfo Multi-shot and ExtInfo + Motion Control examples, ext_info is shown as a readable object to make the structure clear — serialize it to a string before submitting, per the rules above.

Response Example

Note: The model field in the response may include a billing-spec suffix (such as 720p-ref-audio), which differs from the model name passed in the request.

Response Fields

Related APIs

• Kling Video Overview
• Task Status Query