Meta has launched Muse Voice in the United States, bringing real-time spoken conversations with an animated AI avatar to mobile devices, the web and Mac.

The features began rolling out to U.S. users on Thursday across Muse’s mobile app, website and Mac app.

Users can talk with Muse using voice alone, place an avatar beside the conversation or switch to a full-screen, face-to-face view; Meta says hundreds of voices are available.

The default avatar, “Jolly,” uses expressions and gestures that respond to the conversation, and users can customize it as a person, animal or other character, although the AI cannot see them.

Muse Realtime Voice responds immediately, lets users interrupt and can keep delegated tasks running in the background during a conversation, including trip planning, developing ideas and rehearsing presentations or interviews.

The launch builds on Muse’s messaging and task capabilities, introduced on September 8, 2026, which let the agent handle tasks such as sending emails or booking travel, continue working after the app closed and request approval before sensitive actions.

Meta reports that the system generates portrait video at 448 by 768 pixels and 25 frames per second, with the first synchronized response arriving about 870 milliseconds after a user finishes speaking; serving optimizations reportedly allow 12 concurrent sessions on one GB200.

The avatar model creates short video segments using the speech stream, reference imagery and recent video context; Meta says distillation cut the process from 120 guided evaluations to two unguided evaluations per segment while closely preserving quality.

Generated avatar videos carry an invisible Meta Video Seal watermark, and Muse is restricted to users aged 18 and older; Meta says some research examples do not represent avatars available in the app.

Muse surpassed five million app downloads after launching in September, according to the article.

Meta announced the rollout in a blog post, and Alexandr Wang shared a demonstration of his Muse avatar, a fluffy white dog.

The Verge’s Jay Peters found Muse generally quick to respond, though it sometimes paused to look up information; he said the avatar could feel uncanny, and its voice could not be made to sound exactly like Mark Zuckerberg.