Meta equips Muse with live video chat and customizable voice

Live Video and Voice Features

Meta has expanded its personal AI agent Muse to support live video chat and real‑time voice that users can customize.

Chief AI officer Alexandr Wang announced the capability in a September 24 post on X, showing a demo where “users can video chat with Muse and prompt how it sounds.”

Wang noted that the iPhone screen recording used for the demo did not capture audio, inviting users to try the voice themselves.

The addition follows Muse’s September 8 launch as a task‑oriented assistant that can browse the web, connect to apps, and act on user‑approved requests.

By animating an avatar during a conversation, Meta aims to make the interaction feel more present while preserving Muse’s core function of delegated work.

Technical Performance and Comparisons

Meta’s research team released a technical account on September 23 describing Muse Realtime Avatar as an extension of Muse Realtime Voice, synchronizing speech, lip movement, and expression from a shared stream.

The system can animate any reference image, ranging from a photographic portrait to an illustration, animal, or everyday object.

Video generation occurs in short, continuous segments that retain recent visual context to keep the character consistent throughout the exchange.

Meta reports streaming portrait video at 25 frames per second with roughly 870 milliseconds from the end of a user’s turn to the first byte of the synchronized voice‑and‑video response.

These latency figures are company‑reported measurements and may differ for individual users.

To achieve the performance, Meta redesigned its real‑time inference stack, applied model and serving optimizations, and collaborated with Nvidia to lower generation costs.

In internal comparisons, Muse Realtime Avatar was evaluated against Runway Characters and HeyGen LiveAvatar in two‑ to three‑minute conversations.

Meta’s raters preferred Muse overall and across evaluated dimensions, though the mannerism score against Runway was not statistically distinguishable from parity.

The research note cautions that the demonstrations do not necessarily reflect all avatar options currently available in the Muse app.

Implications for Delegated AI Assistants

Muse’s live video feature introduces a more social interface without changing its fundamental role of acting on behalf of the user.

To be useful, the assistant must access personal services and context; to feel conversational, it must hold attention and respond naturally.

Meta emphasizes that users retain control over which apps are connected and can review or modify permissions at any time.

The presence of a lifelike avatar does not automatically improve task completion, and the demo does not quantify how video chat impacts productivity.

Nevertheless, the addition offers Meta another pathway to make a delegated‑work assistant feel present during interactions.

Wang framed the announcement as a product demonstration during Muse’s early rollout rather than a separate model release.

Meta’s research release supplies the technical details that underpin the new experience, highlighting the engineering effort behind real‑time avatar generation.

By integrating visual and auditory cues, Muse moves toward a more immersive conversational partner while still relying on user‑approved actions for tasks such as purchases or email drafting.

Observers will watch how the avatar’s responsiveness and customization options influence adoption as the feature rolls out to a broader audience.

Why This Matters: The live video and voice upgrade gives Muse a tangible presence that could shape how users evaluate AI assistants that blend task delegation with conversational engagement.