Gaze¶
Where a character looks, and how its body gets there. Two engines with different aims, three scopes of how much of the body joins in, and the idle behaviours that keep a resting face alive.
Gaze is the part people notice without knowing they are noticing it. A character whose eyes are dead reads as dead however good the mouth is.
Two Engines¶
Both aim a character at a target. They disagree about what should happen in between.
Biomechanical is a closed loop. It keeps what the character is trying to look at separate from what its body has managed so far, plans each look against how far the character can comfortably reach, relieves a hold that has become strained, and drives the chain through the physics of a thing with weight.
Because the intent and the body are separate, looking away does not erase the intent. The character still knows where the target is, so looking back is a movement, not a snap to a remembered value.
Cascade walks the chain from the head down. Each bone takes its share of the angle, clamped to its own limit, and passes the remainder on. It aims straight at the target with limits and smoothing over the result.
Which One to Use¶
Biomechanical for characters people look at. Dialogue, close-ups, anything where the character holds attention. It costs more and it is the one that produces the behaviour people describe as lifelike.
Cascade when you want predictability. It does what you would expect, it is easy to reason about, and it is the one to reach for on a rig where the closed loop misbehaves, which does happen on unusual skeletons.
The honest summary is that Biomechanical trades predictability for realism and Cascade trades realism for predictability, and both are correct answers to different questions.
How Much of the Body Joins In¶
A look is not only the eyes. How far down the body it carries is a separate choice.
Eyes alone leaves the head where your animation put it. This is the setting to use when a character's head motion is authored and has to stay exact.
Eyes with head and neck is the usual choice, and the one that reads as a person looking at something.
Carried through the torso is for the large turns, where a character turns to face something instead of glancing at it.
Taking the body out is a real requirement and not a limitation. A cutscene with hand-animated head motion should keep it, and letting gaze move the head would fight the animator.
Look Targets¶
Point the character at something and it looks. The target can move, and the character keeps up with it.
A look can be marked as a surprise, which changes how it is approached: a startled look does not ease in the way a deliberate one does.
A target beyond where the character can comfortably reach is handled, never ignored. Under Biomechanical the character reaches as far as it comfortably can, holds, and the strain of holding is itself relieved over time, which is what a person does when something is at the edge of their vision.
Hand IK Compensation¶
When the body turns to carry a look, a character reaching for something would otherwise take their hand with them.
FaceCue compensates, so the hand keeps reaching for what it was reaching for while the torso turns underneath it. Without this, any gaze that reaches the torso breaks every hand placement in the scene.
Blinks and Lids¶
Blinks keep their own schedule, drawn from the character's own randomness so two characters do not blink in unison.
The rate is not constant. Talking raises it. Concentration lowers it, the way it does in someone working something out. Listening deliberately does neither, because attention does not change how often people blink.
Lids trail fast vertical movement. Look sharply up or down and the lid follows a moment behind the eye, which is what real lids do and what their absence makes look wrong.
The Eyes at Rest¶
A resting eye is never still, and three separate behaviours keep it alive.
Drift is the slow wander a held gaze has.
Small movements hold a target instead of locking onto it, the tiny corrections an eye makes constantly.
Larger movements break away to something else worth looking at, then come back. How often depends on what the character is doing: a talker's gaze breaks away considerably more often than a listener's, which is one of the clearest differences between the two states.
Idle Head Motion¶
A slow sway runs underneath everything, and it is deliberately small while a character is talking, because the emphasis-driven head motion is already moving the head and the two would compete.
Posture shifts move the head and neck as a unit while the eyes hold their target, which is how a person changes position without losing what they were looking at.
The head rights itself against the body under locomotion, so a character walking does not carry their head at the angle their spine happens to be at.
Listening gates all of this right down. People hold still when they are paying attention, and a listening character that keeps swaying reads as distracted.
Limits¶
Neither engine decides what to look at. They aim at a target you set. Choosing the target is your game's job, and there is no attention model here picking one for you.
Nothing here averts gaze deliberately. A character does not look away because it is uncomfortable or thinking. What changes with state is how often the eyes break away and how much the body settles, not a decision to avoid a look. Deliberate aversion is on the roadmap.
A look stops where the body stops. Both engines respect joint limits, so a target behind the character is approached as far as the character can turn and no further.