What makes a real-time AI avatar feel real
Gaze, blinks, breathing and timing shape a live avatar’s expression. How Facelet approaches animation in its native Mac app and upcoming Windows PC version.
Mac and Windows: Facelet is built natively for both. The Mac app is available now; the Windows PC version is coming soon, after signing and release testing. See Windows requirements and release status.
Facelet’s live 3D characters combine facial animation with gaze, breathing and small movements. These details shape how a conversation feels in the native Mac app and the Windows PC version coming soon. Here is how we approach them.
Gaze
Constant eye contact can make an animated face look rigid. Facelet varies the character’s gaze and head position between listening, thinking and speaking. The aim is to give each part of the conversation a visible state without making every movement demand attention.
Blinks, breath and the shoulders
Speech animation is only part of the character. Facelet also animates blinks, the head and neck, breathing and shoulder sway. These movements keep the face and upper body active between replies. Too much motion would be distracting, so their scale matters as much as their presence.
Listening and thinking
Facelet has distinct listening and thinking states and moves between them as the conversation changes. A head tilt or a change of gaze gives visual feedback while you wait. Those are animation cues; they do not change how ChatGPT produces an answer.
The mouth, and the clock
Facelet analyzes the reply audio on your computer and uses it to animate the mouth and face. The aim is to follow speech sounds as well as volume. The mouth can trail the voice by a fraction of a second, and timing depends on the audio and rendering pipeline. A live avatar does not guarantee an instant ChatGPT response.
Stillness on request
Facelet’s hold-still setting pauses idle motion. A nod or a wink adds a deliberate gesture. Having control over movement lets you choose whether the character should be expressive or quieter while you work.
Framing and light
Facelet frames its characters as a face and shoulders, with lighting that makes the eyes and expression visible. Both native versions draw the character on the computer’s graphics hardware. The released Mac app requires Apple silicon, M1 or newer. The Windows PC version is coming soon; its hardware and release qualification are still in progress.
You can watch Facelet in motion to see how these details work together, or read about local avatar rendering and its hardware requirements.
