What motion control means in AI video
Motion control is conditioning a video model on the movement extracted from a reference clip, so the output follows that exact choreography instead of whatever the model would have invented from a description.
What gets extracted
A pose estimator runs over the reference clip and produces a skeleton per frame — joint positions, nothing else. No pixels of the reference person reach the output.
The generator then draws your subject into each skeleton in turn. That is the whole trick, and it is why the reference dancer's appearance never leaks into the result.
Why it beats a prompt for dancing
Repeatability. Run a template twice and you get the same routine twice, because the conditioning signal is identical. Run a prompt twice and you get two different clips.
Specificity. "Does the dance from that clip" is not expressible in a sentence. A skeleton per frame expresses it exactly.
What you give up
Freedom. You get the routines that exist in the library and nothing else. If you want your subject riding a horse through a supermarket, that is a prompt, not a template.
Framing. The reference clip fixes the camera. A full-body routine cannot be delivered as a close-up.