Underneath the skin it is four systems stacked: a skeleton that holds the shape, actuators that supply force, linkages that carry that force to the surface, and a controller replaying motion that a human authored frame by frame.
A cutaway of the most common subject. The thing most people get wrong: the motors are almost never at the joint they move. They sit in the body, and the motion is routed up to the face.
Motion is not generated live. It is authored ahead of time on a timeline — one track per channel — then played back locked to the audio, so the mouth lands exactly on the recorded words.
Three families, chosen by what the movement needs. Most large figures mix all three in one body.
A motor with a built-in position sensor. You send it an angle and it holds that angle. Cheap, controllable, and easy to stop anywhere mid-stroke — which is why faces, eyes and fingers are almost always servos.
Compressed air pushing a piston. Explosively quick and strong for its weight, which is why theme-park figures use it for jaw snaps and lunges. Air compresses, so holding a precise mid-stroke position is genuinely hard.
Oil under pressure. Oil barely compresses, so unlike air it holds position under load and moves smoothly. This is what drives figures the size of a dinosaur. Heavy, expensive, and it leaks eventually.
Push rods, bellcranks and Bowden cables — the same idea as a bicycle brake. They convert push into rotate, change direction, and let a heavy motor sit in the chest while the thing it moves is in the eyebrow.
Expression is not one movement. It is a dozen small ones arriving in the right order, at slightly different times — which is exactly what makes it hard.