Most real-time AI avatars I've seen are framed from the shoulders up. That framing is a workaround. A face on a plain background is about the easiest thing a video model can keep consistent. So the crop quietly picked the use cases. Real-time avatars went into support, tutoring and sales, work that's mostly conversation. Jobs that needed hands stayed out of reach. Hands and the objects they hold are where AI video still slips, like the mug that turns into a vase mid-shot. Vivix's new A1 model…