Motion-transfer footage can look smooth because a short beat disappeared from the copied action. A dance cue or tutorial gesture becomes inaccurate if that movement matters. Test with clearly separable reference beats and a stable camera before grading whether the result carries the expected performance.
Build the test input
Use a permitted performer tapping a desk with the index finger, lifting the palm, and rotating the wrist in one small sequence. Annotate the start and end of each distinct motion in a note. Select a fictional target with a plain sleeve and no detail near the wrist clip edge.
Run the proposed comparison
Run the same performer file and target still in comparable motion-reference operations. Establish which version accepts the required video input first. Compare frame rate and trim alignment. Watch once at delivery speed, then count the finger tap and palm action from the export at the key event moments.
Decide what passes
Pass when the three distinct actions remain present and in the original order. Smooth video that combines the tap and lift in one nonspecific motion must still fail if the source action needs reproduction. Mark ordinary differences in intermediate pose as less important than the missed beat.
Act on the outcome
When the reference includes more action than the target output handles, produce a calmer reference with pauses, or split the transfer into shorter shots. Try the simplest variation before raising resolution. Allocate retries based on repeatable adherence to the required sequence, not independent qualities of the actor image.

