Meta Unveils Audiovisual AI to Power Realistic Avatar Conversations

Details

• Meta's Fundamental AI Research (FAIR) team introduced audiovisual behavioral motion models designed to generate natural conversational dynamics for digital avatars.
• This development was a collaborative effort between the FAIR group, Codec Avatars lab, and Core AI lab at Meta.
• The models leverage both audio and visual cues to produce lifelike facial expressions and body gestures, enriching interactions across 2D and 3D Codec Avatars in virtual and augmented reality.
• The system is built on the Seamless Interaction Dataset, comprising more than 4,000 hours of two-person conversation recordings.
• The dataset is the largest of its kind, featuring diverse interactions and incorporating robust privacy protections, such as watermarking and content safeguards.

Impact

This breakthrough paves the way for more lifelike, emotionally nuanced experiences in VR and AR, potentially accelerating adoption of social and telepresence technologies. With the largest open dataset in the field, Meta is setting a new benchmark for conversational AI and inviting wider innovation. The move comes as competitors have yet to launch similar capabilities, giving Meta a clear edge in avatar-driven digital communication.