Motion transfer is the underlying ML technique behind Motion Control. It is the process of extracting motion data — body pose, gesture, camera path, lip movement — from a source video and applying it to a different character or image. The character's identity (face, outfit, lighting) is preserved from the target image; the motion is fully inherited from the source.
Motion transfer is reference-driven, which makes it deterministic in a way prompt-based generation isn't. The same source clip produces the same motion every time. This is useful when motion fidelity matters (clone-video, dance trends, choreography) and unnecessary when motion is generic (talking heads, ambient B-roll).
The leading consumer-facing implementation is Kling 2.6 Motion Control, served via fal.ai. Inside UGC Copilot, motion transfer is exposed as the Motion Control toggle in the clone-video render queue.