Express avatars can be trained with roughly one minute of video and replicate users’ head movements during speech. Premium+ is the company’s marquee model, which might need at least a few minutes of recording for training, but can reproduce an AI avatar for videos with hands and torso. The company’s idea is that with hands and torso movements, the avatar feels more human-like in interaction with other users.

The company said that the Premium+ models are capable of real-time interaction that might be good for use cases like webinar and translation.