The renderer is only one layer
A character can look good and still feel dead if the app has no structured listening, thinking and speaking states.
Lip sync needs a lifecycle
Speech start, cancel and interruption need to control the mouth reliably. A frozen open mouth is a state bug, not an art problem.
Expression should be structured
Expressions work better when driven by Companion state/events than by brittle word spotting in the chat text.
GPU sharing is part of the feature
The avatar cannot assume it owns the GPU when the same machine may also run the LLM and image generation.
This article describes AVRE’s current product direction and general technical reasoning. Details can change as development and testing continue.