Understand the two-layer pipeline where Attention selects focus targets and Gaze drives procedural eye tracking, head rotation, and saccades.
The Gaze & Attention utility adds natural eye contact and head tracking to AI characters. It runs entirely inside Unity — no data is sent to Convai — and operates independently of the session lifecycle. Two cooperating layers handle what the character looks at and how it looks there.
The Attention layer decides what the character looks at. It evaluates all registered IFocusTargetProvider components each frame, selecting a winner using priority, distance-based relevance, and an interest budget that prevents indefinite fixation on a single target.
The Gaze layer decides how the character looks at the attention target. It translates the active attention reading into eye rotation, head movement, procedural saccades, blinks, and eyelid follow. Gaze authority is weighted by the current dialogue state — the character looks more assertively during speech and more softly during idle.
Each layer is independently configurable through ScriptableObject profiles. You can tune attention persistence, eye tracking sharpness, head range, saccade frequency, and per-dialogue-state gaze authority — all without code.
ConvaiAttentionController
Selects the active focus target each frame
ConvaiGazeCoordinator
Blends attention output with dialogue state to produce GazeIntent
ConvaiEyeGazeActuator
Rotates eye bones and drives blendshapes from GazeIntent
ConvaiHeadLookActuator
Rotates neck and head bones from GazeIntent
AnimationRiggingGazeBridge
Optional: drives Unity Animation Rigging constraints instead of procedural bone writes
Last updated
Was this helpful?
Was this helpful?
