Google launches Live Avatar in Gemini Enterprise: conversational AI with real-time lip-synced faces

Google launches Live Avatar in Gemini Enterprise: conversational AI with real-time lip-synced faces

On September 24, Google launched Gemini 3.8 Live with Live Avatar in general availability for Gemini Enterprise customers — a conversational AI with an animated persona that lip-syncs and displays facial expressions in real time. The feature was first shown at Google Cloud Next 2026, and arrives one day after the company's new text-to-speech models, in a week where Google is consolidating its audio and video offerings into a single enterprise product.

What was launched

According to The Verge, users with the Gemini 3.8 Live update can hold conversations with the model while an animated AI persona responds in real time. The "Live Avatar" lip-syncs and displays various facial expressions during the conversation. For now, the feature is available only to Gemini Enterprise customers — consumers do not yet get access.

Android Headlines, which also covers the launch, describes the feature as combining real-time video generation directly with speech models across websites, mobile apps, and interactive self-service kiosks — touchpoints where businesses today typically rely on text chat or physical service points.

How it works

The most technically concrete details come from Android Headlines, and should be read with the caveat that they are not independently confirmed: the Live Avatar engine is said to process visual and voice inputs simultaneously, allowing the avatar to see the user's camera feed or screen share and hold "natural conversations" based on what it sees. The system is also said to support asynchronous tool calls — backend tasks that run while the conversation is ongoing, so the avatar can look up information in a company's systems without freezing the conversation. The report also ties the feature to Google's Agent Development Kit (ADK), the framework the company offers for building agents.

It is worth emphasizing the distinction: that the avatar lip-syncs and displays facial expressions in real time is reported by The Verge with reference to Google. The details on camera input, asynchronous tool calls, and ADK integration appear in Android Headlines alone, and are not confirmed by other sources in this evidence base.

Languages and quality — Google's claim

Google claims that Live Avatar can switch among the 97 languages it supports "without degrading video fidelity or introducing visual drift" — meaning the avatar's appearance and image quality are supposed to remain stable across language switches. This is a vendor claim, cited by The Verge, and not independently verified. For businesses operating across languages, this is precisely the kind of consistency that matters — but we do not know how the performance actually behaves in practice.

Custom avatars and watermarking

Google offers a library of preset avatars that businesses can choose from, but organizations can also create their own. According to Android Headlines, only a single reference photo and one audio sample are required to create a custom persona — but creation is subject to strict approval via an allowlist. How large this list is, or who is on it, has not been disclosed.

All output is said, according to Google, to carry the company's invisible SynthID watermark, along with safety measures to "respect identity." The latter is an obvious response to the risk posed by technology that can animate faces from a single photo and one audio sample: that the same capability can be misused for deepfakes. How the watermarking works in practice, and how robust it is, is not documented in the available evidence.

The context: a busy week and competitive pressure

Live Avatar lands one day after Google introduced Gemini 3.8 TTS and Gemini 3.8 Flash-Lite TTS (September 23), which the company positions as its most capable audio generation models ever, according to TechTarget. The two launches in the same week appear as a package that assembles audio and video into one product.

Analyst Bradley Shimmin of Futurum Group characterized the TTS work this way to TechTarget (the quote is translated from English): "What we're seeing here is a refinement of your text-to-speech, with some heavy targeting to specific use cases." In other words: refinement, not breakthrough — and the main benefit for businesses is that everything is now gathered in one model family.

Android Headlines frames the launch in terms of competitive pressure: the company's shares are reported to have fallen around 20 percent in recent months following reported delays with Gemini 3.5 Pro, while competitors OpenAI and Anthropic have launched newer competing models, and a Google DeepMind executive is reported to have teased Gemini 4. These are details from a single source that are not confirmed by other sources here, and should be read with that caveat.

Open questions

Several key details remain missing: Google's own primary documentation (an official blog post, release notes, or Enterprise documentation) is not part of the evidence base for this story, so all capability claims rest on secondary coverage that itself cites Google. Pricing is not known. The specific "enterprise compliance guarantees" mentioned are not specified. It is unclear whether and when Live Avatar will become available outside Gemini Enterprise, for example to consumers in the Gemini app. And we do not know how SynthID detection actually works in practice for this type of video output.

What is clear, however, is the direction: realistic, speaking, and seeing AI faces are no longer a demo program from a conference stage, but a product businesses can deploy today — with the technical, commercial, and ethical questions that entails.