On November 6, Zuckerberg announced on Threads that the personal AI assistant Muse has officially launched real-time voice conversations and avatars. The feature is first available to users in the United States, covering mobile, web, and Mac apps. Users can tap the phone icon in the upper right corner of the app to chat with it as if making a phone call, and can interject to interrupt mid-sentence at any time. This means Muse is no longer just text in a typing box, but a conversational partner that can respond and talk back.
One Set of Models Handles Both Mouth and Face
Behind it are two models, Muse Realtime Voice and Muse Realtime Avatar. The former generates both "what to say" and "how to say it" simultaneously, while the latter follows the same data stream to add expressions, lip movements, and gestures in real time, keeping audio and visuals in sync. The system can stream output vertical video at 448×768 and 25 frames per second, with roughly 870 milliseconds from when the user finishes speaking to receiving the first audio-visual response. For interaction, users can go voice-only with the avatar staying beside the chat, or switch to fullscreen "face-to-face"; the default cartoon character Jolly can be swapped for people or animals, and there are hundreds of voice styles, from "New Yorker energy" to "pirate accent."
Targeting "Hands-Free" Scenarios
Meta positions it as a more natural personal agent: while walking or with hands busy, you can assign tasks by voice, and Muse carries them out in the background; it can also be used for brainstorming, trip planning, or practicing interviews and speeches. The company also demonstrated making the avatar dance, then sharing it with one tap to Instagram, Facebook, and WhatsApp. Once AI grows a face and a voice, the companionship-chat business will probably have to compete along the "real-person feel" line.