Meta adds live video chat to Muse, giving its AI agent a face and voice

Meta added live video chat and customizable real-time voice to Muse, its personal AI agent, with chief AI officer Alexandr Wang demonstrating the feature in a September 24th post on X. Meta's research team published a September 23rd technical account of Muse Realtime Avatar, reporting portrait video streamed at 25 frames per second and roughly 870 milliseconds from the end of a user's turn to the first byte of its synchronized voice-and-video response. The feature follows Muse's September 8th debut as a personal AI agent that works across connected apps, browses the web and continues tasks in the background with user approval for sensitive actions such as purchases and emails.

Meta adds live video chat to Muse, giving its AI agent a face and voice Alexandr Wang demonstrated the feature on September 24th, a day after Meta outlined the real-time avatar technology behind it. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: X https://x.com/alexandr wang/status/2102921739621806290 Why it matters Meta is making its personal AI agent more conversational as it asks users to trust it with connected apps and delegated tasks. Live video makes the interface feel more present; whether that improves usefulness enough to justify the added access remains unproven. Meta https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/ has added live video chat and customizable real-time voice to Muse, its personal AI agent, extending the product from a task-oriented chat interface into a conversation with an animated presence. Meta chief AI officer Alexandr Wang @alexandr wang https://x.com/alexandr wang demonstrated the feature in a September 24th post on X https://x.com/alexandr wang/status/2102921739621806290 , saying users can video chat with Muse and prompt how it sounds. https://x.com/alexandr wang/status/2102921739621806290 https://x.com/alexandr wang/status/2102921739621806290 Wang's post showed the product rather than laying out its technical details. He noted that the iPhone screen recording did not capture audio, and told users to try the voice themselves. Axios reported https://www.axios.com/2026/09/23/meta-muse-ai-hardware-handheld-glasses the day before that Meta had added live video chat and real-time voice conversations with custom voice options to Muse. Meta's research team published a technical account of the avatar system on September 23rd. The research post describes Muse Realtime Avatar as an extension of Muse Realtime Voice: speech and video are generated from a shared stream so the avatar's voice, lip movement and expression stay synchronized. Meta says the system can animate a reference image during a live conversation, from a photographic portrait to an illustration, animal or everyday object. The model generates video in short, continuing segments, carrying recent visual context forward to keep the character consistent as the exchange continues. Meta reports that its system streams portrait video at 25 frames per second, with approximately 870 milliseconds from the end of a user's turn to the first byte of its synchronized voice-and-video response. Those are company-reported system measurements, not a guarantee of the delay every user will experience. The same post says Meta redesigned its real-time inference stack, used model and serving optimizations, and worked with Nvidia to reduce the cost of generation. Meta also compared Muse Realtime Avatar with Runway Characters and HeyGen https://runtimewire.com/models/fal/heygen-v2-translate-precision LiveAvatar in two-to-three-minute conversations. The company's raters preferred Muse overall and across the evaluated dimensions, though Meta said its mannerism result against Runway was not statistically distinguishable from parity. The comparison was conducted by Meta, and the research post cautions that its demonstrations do not all represent avatars https://runtimewire.com/models/fal/veed-avatars-1 available in the Muse app. That leaves a distinction between the underlying technology Meta has shown and the exact set of choices users can make in the product. The video feature follows Muse's September 8th debut as a personal AI agent. Meta introduced Muse as software that can work across connected apps, browse the web and continue tasks in the background, with user approval for sensitive actions such as purchases and emails. Its initial pitch was practical delegation: give Muse a task, then review what it proposes to do. Live video adds a more social interface to that agent, while leaving its core job - acting on a user's behalf - in place. That combination sharpens the product's central tradeoff. To do useful work, Muse needs access to personal services and context; to feel like a conversational partner, it needs to hold attention and respond naturally. Meta says users control which apps they connect and can review or change permissions. A lifelike, responsive avatar does not itself make an agent more capable, and the demonstration does not establish how well video chat helps users complete tasks. It does give Meta another way to make an assistant built around delegated work feel present during the interaction. For Wang, the post is a product demonstration during Muse's early rollout, not a separate model announcement. Meta's research release supplies the more consequential technical detail: the company is building voice and generated video as one real-time system, rather than simply placing a prerecorded or separately animated character over a chatbot. Whether that added presence makes people more willing to use Muse - and grant it access to more of their digital lives - is a product question the feature itself cannot answer.