Both position a limb. They differ in what you control.
Forward kinematics
Rotate each joint in turn from the root outward. Shoulder, then elbow, then wrist.
- Natural for arcs. A swinging arm, a throw, anything where the motion of the whole limb matters.
- Predictable. Each joint does exactly what you set.
- Bad at contact. Making a hand land precisely on a surface means adjusting three joints iteratively until it happens.
Inverse kinematics
Place the end — the hand, the foot — and the solver works out the joint rotations to reach it.
- Natural for contact. Feet on uneven ground, a hand on a doorknob, a weapon held in two hands.
- Stable targets. The foot stays planted while the body moves over it.
- Less predictable in the middle. The solver chooses joint angles, and its choice may not be the one you wanted — which is what pole vectors are for, steering the elbow or knee direction.
Using both
Most rigs offer both with a blend, because an animation frequently needs each at different moments. A walk uses IK for the planted foot and FK for the swinging one; the switch between them is animated as a blend so the transition does not pop.
In engines
IK at runtime handles what an authored animation cannot know: ground that is not flat, targets that move, characters of different heights sharing an animation. Foot IK adjusting to terrain is nearly universal in third-person games.
Where it relates to generated characters
An auto-rigged character gets a skeleton, not a rig — bones and weights, without IK controls, pole vectors or a control hierarchy. Those are authored on top, either in a DCC tool or through the engine's own IK systems. That distinction is worth being clear about: rigged and animatable are not the same state.