A reference gesture reads differently when the hands fill the frame versus appearing far away. Before buying a set of motion-control retries, compare the same movement at reference and target scale. fal lists a Kling 2.6 Pro motion-control endpoint. Oakgen's Kling 3.0 page documents a separate version.
Build the test input
Film a permitted performer raising one forearm beside the face, then stepping back with the same arm visible. Create two crops of the same movement without changing timing. Design one original target character whose arm has a recognizable band near the wrist. Keep background and target framing fixed.
Run the proposed comparison
Where allowed by the operation, submit the full-frame and cropped reference as separate conditions. Save the effective crop of the actual output before assessing it. Match time windows across candidates and check the first visible raised arm, turnaround point, and the return into its resting position.
Decide what passes
Pass when the intended arm remains visible, elbow and wrist follow the same order, and the band stays on the correct side. Separate copied background movement from pose error. For example, a slight target camera pan may widen framing without any actual change to the arm's recovered motion.
Act on the outcome
If using a tight motion crop makes the result easier to follow, shoot that crop deliberately before buying further retries. If it produces frequent wrist disappearance, match coverage to the target frame and remove clutter. Comparative scores apply to the chosen movement and references rather than an entire version family.

