Kling 4.0 stable motion vs Sume's Kling 3.0 Motion Control

Kling 4.0's page says it has enhanced motion stability for action, sports and dance (read 2026-10-02), but that describes text and reference generation, not copying a specific move. To copy a given movement onto a still on Sume, call Kling 3.0 Motion Control (kling/3.0/motion-control) with a driving video of up to 30 seconds. Sume lists no Kling 4.0 id.

What does Kling 4.0 say about motion?

The Kling 4.0 feature page lists stable dynamic motion for action sequences, sports, dance and extreme sports, 30-second single-pass clips, up to ten keyframes, and up to 15 combined references. Flash is limited to 20 seconds and 720p, and is in a limited rollout.

Those are properties of a generative model steered by prompts and references. The page does not say you can hand it a video and have that motion transferred onto a photo.

How does Sume's Motion Control work?

You give a public HTTPS still (or a ready avatar), a public HTTPS motion video, and duration_seconds. The output length follows the driving video. character_orientation decides whose framing wins: video (default) or image. keep_original_sound keeps the driving audio by default.

The optional prompt is up to 2000 characters and, per the schema, only steers appearance details; motion and framing follow the video.

Which should I choose?

If you can film or find the exact move, Motion Control is the controlled route, and it exists on Sume today. If you want a model to invent the motion from a prompt, use the video models Sume lists and check the video catalog for ids.

Read Jobs and results for the poll-and-fetch loop that every job uses.