Kling Video Models: Features and Capabilities

Last updated: March 24, 2026

Available Kling Models

Runware supports several Kling video generation models:

Kling 3.0 Standard and Pro

The latest generation of Kling video models, built for longer, more stable clips with enhanced features:

  • Text-to-video (T2V) and Image-to-video (I2V) generation

  • Multi-shot generation in manual or automatic modes

  • Cross-shot consistency for characters and voices

  • Multi-language audio with lip sync

  • Up to 15-second clips with scene cuts and reduced drift

  • Native audio support

AIR IDs:

  • Standard: klingai:kling-video@3-standard

  • Pro: klingai:kling-video@3-pro

Kling O3 Standard

Also available with similar capabilities to Kling 3.0.

Motion Control

Motion control is supported for Kling 3.0 models, allowing fine-grained control over motion paths and trajectories. This feature uses reference images and videos along with character orientation settings.

Sample payload for motion control:

[
  {
    "taskType": "videoInference",
    "taskUUID": "abcdb1fc-0124-44fd-93c7-f6d6fe3daabb",
    "model": "klingai:kling-video@3-standard",
    "positivePrompt": "A person dancing gracefully in a sunlit room.",
    "deliveryMethod": "async",
    "numberResults": 1,
    "outputType": "URL",
    "outputFormat": "mp4",
    "includeCost": true,
    "providerSettings": {
      "klingai": {
        "keepOriginalSound": true,
        "characterOrientation": "video"
      }
    },
    "inputs": {
      "referenceImages": [
        "https://example.com/reference-image.jpg"
      ],
      "referenceVideos": [
        "https://example.com/reference-video.mp4"
      ]
    }
  }
]

Elements Support

Elements support is currently not available for Kling models but is under review by the team. Updates will be provided when this feature becomes available.

ComfyUI Support

Both Kling 3.0 and Kling O3 models are supported via ComfyUI integration.

Documentation

For complete API documentation and examples, visit:

You can also test these models directly in the Runware playground: