Kling Motion Control prompt: appearance only, motion is the video

In Sume's Kling 3.0 Motion Control, the prompt steers how the character looks. It does not choose the movement: motion_video_url drives all output motion. If the clip moves wrong, change the motion video, not the prompt.

What each field controls

Each field has one job in the Sume contract for POST /v1/kling/3.0/motion-control.

Duration and price

The provider has no duration setting: the length of the motion video sets the output length. Sume only uses duration_seconds to price the job at admission, as ceil(duration_seconds) x $0.1575 (fal list $0.126 x 1.25). If you declare 10 s but upload a 6 s motion video, the clip is 6 s long while the hold was for 10.

What the vendor schema says

fal's own schema calls the prompt an optional description of the action or scene, with an example of a man dancing (fal page, read 2026-10-05). Sume's doc is stricter: it positions the prompt as appearance guidance. Write your prompt to describe the person and style, and let the reference video carry the action.

Strict body

This route is strict. It rejects Video 1.0 vocabulary such as reference_image_urls, reference_video_urls, model and provider endpoint fields. That keeps it from becoming a back door into the kling-3 text-to-video reference inputs, which do not exist on that row either.

Example

A body that sets appearance in the prompt and leaves motion to the video.

{
  "image_url": "https://example.com/character.png",
  "motion_video_url": "https://example.com/walk-6s.mp4",
  "duration_seconds": 6,
  "prompt": "Same person, navy raincoat, soft overcast light"
}