Introduction

A week after Gemini 3.8 Live landed as a speech-first dialogue model, Google gave it a face. On Sep 24, 2026, the Gemini Audio team introduced Gemini 3.8 Live with Live Avatar — near real-time streaming video paired with live speech so enterprise agents can listen, look, speak, and emote in one continuous conversation. It is generally available in Gemini Enterprise starting that day. Think of it as Google’s answer to the live-avatar wave Meta kicked with Muse Realtime Avatar — this time glued to Gemini’s tool-calling brain and shipped for company deployments, not a research demo alone.

What Ships Today

Live Avatar natively couples Gemini’s live dialogue stack with low-latency generated video. The pitch is precise lip-sync, natural expressions, and fluid turn-taking so customer-service bots, interactive walkthroughs, and brand agents feel present rather than like a voice behind a still photo.

Headline product claims from Google’s announcement:

  • Near real-time audiovisual replies that take in what the avatar sees and hears
  • Multimodal conversation — audio, images, video, and text in; audio, video, and text out
  • Asynchronous tool execution while the avatar keeps talking — hotel check-ins and similar chores can run in the background without freezing the face
  • 97-language speech-to-speech with lip-sync and expression adapting mid-conversation, without the visual drift that usually kills multilingual talking heads
  • Preset avatar library plus custom avatars from a high-quality reference image (custom creation is enterprise-allowlisted today)
  • SynthID watermarking woven into both audio and video output

Secondary coverage echoing Google Cloud notes adds practical deployment detail: US and EU endpoints under provisioned capacity, console tryouts for Enterprise customers, and sales-led paths for dedicated capacity and custom-avatar allowlisting. Model-card chatter around Gemini 3.8 Live points to a Gemini 3 Pro–family base, roughly 128K input tokens and about 24K output tokens, with Live Avatar emphasizing full-head rotational dynamics and phonetic alignment for a few minutes of continuous engagement rather than hours-long sessions.

Why Creators and Product Teams Should Care

Live talking heads are no longer a novelty slide. Meta’s Muse Realtime Avatar framed the research case for audio-locked DiT faces. Google’s drop is the enterprise packaging: same category, but with multilingual lip-sync at global scale, tool calls that do not kill presence, brand-fit custom faces, and SynthID as a trust default. If you build concierge, retail, training, or support experiences, the interesting question is less “can AI animate a face?” and more “can the face stay useful while the agent actually does work?” Live Avatar’s async tool story is aimed squarely at that.

It is not a consumer Grok Imagine / Flux playground model and it is not on ArtRealmAI Gen. Treat it as a Gemini Enterprise product lane — API and console — rather than a weekend local checkpoint.

How It Fits the Avatar Race

  • Meta Muse Realtime Avatar — research/product push for audio-driven DiT faces synced to Muse Realtime Voice (~subsecond A/V), preference-tested against Runway Characters and HeyGen LiveAvatar in Meta’s own study.
  • Gemini 3.8 Live with Live Avatar — Google’s live-dialogue model plus streaming video persona, 97 languages, enterprise GA, custom reference faces, SynthID.
  • LemonSlice CWM-1 and similar interactive character models — adjacent controllable avatar video, different stacks and distribution.

Same hunger (presence in conversation), different doors. Google’s door opens through Gemini Enterprise and the Live API docs.

Original Source

https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-with-live-avatar/

Google Blog, Sep 24, 2026 — Introducing Gemini 3.8 Live with Live Avatar (Shuo-yiin Chang, CJ Zheng / Gemini Audio Team). Secondary enterprise framing: The News International’s Sep 24 write-up summarizing Google Cloud availability notes. For the speech-first precursor, see Google’s earlier Gemini 3.8 Live post.

Conclusion

Gemini just made its live agent look at you while it talks — and keep talking while tools run in the background. For enterprises that need branded presence across dozens of languages with watermarked A/V, Live Avatar is the clearest Google-shaped invitation yet into conversational video. Pair it with the Live API docs, watch the allowlist for custom faces, and remember SynthID is on every frame by design.

—Aurelia ♡