Kling motion control reference video requirements

A Kling motion control reference video should show one person in one continuous shot, last 3 to 30 seconds, and have a short edge of at least 340 px and a long edge of at most 3,850 px. Kling's guide also asks for steady, moderate movement and no cuts or camera moves.

As of 2026-09-29, those numbers come from Kling's own Motion Control User Guide, where they sit in the Video 2.6 section; the 3.0 section does not repeat them. What Sume's endpoint enforces comes from the request schema in the Sume API reference, described in the API reference docs.

What does Kling ask of the motion reference video?

The guide's rules for the clip that supplies the movement, in the wording of its 2.6 section:

What does Kling ask of the character image?

The image is the other half of the input. Kling says to match the character's full-body or half-body framing in the image with the framing in the motion reference, and to leave enough space in the image for the character to move freely when the motion is large.

Kling's own tip for the motion clip: pick one with a wide range of motion, moderate speed, and minimal displacement.

What does Sume check on the driving video?

On Sume the motion reference is the motion_video_url of POST /v1/kling/3.0/motion-control, and the schema is short on rules. The URL must be a fetchable public HTTPS video of at most 30 seconds, and duration_seconds must be a number from 1 to 30. The schema lists no pixel-size, shot-count or people rules, so Kling's list above is guidance for a good result, not a set of documented Sume checks.

Sume also accepts exactly one visual source: image_url, or a ready avatar_id / avatar_handle, and the API reference marks them mutually exclusive.

How do I send a clip that meets the list?

Set duration_seconds to the clip's length. Sume uses it to reserve the price at submit, and the output length follows the driving video.

curl -X POST https://api.sume.com/v1/kling/3.0/motion-control \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: motion-ref-001" \
  -d '{
    "image_url": "https://example.com/character.png",
    "motion_video_url": "https://example.com/reference-8s.mp4",
    "duration_seconds": 8
  }'

What happens if the clip breaks a rule?

Kling warns that a clip with cuts or camera movement may be truncated, and that very complex or fast action can produce a result shorter than the upload, because the model keeps only the valid continuous segments. The Sume endpoint's fields and price are in Motion control API: animate an image with a driving video.