Article 1 ===

Gemini 3.8 Live Avatar: Google Gives Its AI a Talking Face

Google's Gemini assistant now has a face — and it talks back. On September 24, 2026, Google introduced Gemini 3.8 Live with Live Avatar, an animated AI persona that holds real-time voice conversations, moves its lips in sync with its words, and reacts with facial expressions as it speaks. The quick answer for most readers: yes, it is real; yes, it is available today; and no, you cannot download it as a regular user — it is currently limited to Gemini Enterprise customers.

The announcement was published on Google's official blog by the Gemini Audio Team, written by research scientist Shuo-yiin Chang and software engineer CJ Zheng. Google's availability statement is unusually direct: "Starting today, Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise."

There is no free tier in this announcement and no consumer app update. Individual Gemini users will not wake up to a talking avatar on their phones. The release still matters for everyone, because it previews where voice AI is going — and because it landed just one day after Meta unveiled its own avatar technology at its Connect conference. Here is what launched, who can use it, and what actually changes.

Source: Google Blog

What Live Avatar actually does

Live Avatar is not a looping video of a digital human pasted on top of a chatbot. It is a live animation generated as the model speaks, built into the Gemini 3.8 Live experience. According to Google, the system "listens, sees, and speaks" — you can talk to it by voice while showing it something, and it responds with fluid turn-taking, meaning you can interrupt it naturally the way you would with a person.

Two capabilities deserve special attention. First, the avatar can pull information up on the screen while it is still talking, as The Verge's coverage pointed out. Second, Google highlighted asynchronous tool calling: the avatar can call tools and fetch data in the background while the conversation keeps flowing, with no awkward pause. Google's official demo shows a hotel check-in, where the avatar welcomes an arriving guest and chats naturally while it quietly retrieves the reservation details.

The feature builds on the Gemini 3.8 Live model with Extended Thinking that Google revealed on September 15, 2026 — nine days before this announcement. We covered that launch in detail, including how the model weaves longer reasoning into live voice conversations.

Source: Google Blog

Speaking 97 languages with one face

The most striking technical claim is a number: 97. Google says the avatar "dynamically adapts its lip-sync and expressions and can seamlessly transition across 97 languages without degrading video fidelity or introducing visual drift."

In plain English, the face never breaks character. If a conversation switches from one language to another, the mouth movements, the expressions, and the video quality all stay consistent. The Verge's demo shows the avatar moving between English and Japanese mid-conversation, with the mouth animation matching the spoken language in both.

This matters well beyond a polished demo. Most real service conversations in international business are multilingual — a caller starts in English, then drifts into their mother tongue. Voice assistants have usually handled this poorly. An avatar that follows the switch without visual glitches removes one more reason for companies to force customers onto a single language.

Google did not list all 97 languages in the announcement itself. However, when the underlying Gemini 3.8 Live model launched on September 15, Arabic was confirmed among the supported languages. For bilingual support teams — for instance a service desk handling English and Arabic callers in the Gulf, or a French-English desk in Brussels — that language coverage is the difference between a demo and a deployable tool.

How companies get their own avatar

Google offers two routes. The first is a library of preset avatars: ready-made characters that an organization can deploy quickly. The second is more ambitious. A company can create a custom avatar from a single high-quality reference image, "preserving reference likeness, brand styling, or character identity" — so the avatar can match a spokesperson, a mascot, or a brand's visual identity.

The custom route has a catch. It is currently available only through enterprise allowlisting. In practice, that means a company works directly with Google to get its custom avatar approved, rather than uploading a photo through a self-service tool. That is a managed process — a sensible guardrail for technology that can replicate a human likeness, but also a barrier for smaller teams.

For developers, the starting point is the Gemini Enterprise platform and the API documentation Google links from the announcement. Google has also published a model card, which describes the system's behavior and limits — useful reading before any customer-facing deployment.

The invisible SynthID watermark

Every second of Live Avatar output carries SynthID, Google's invisible watermark. Google says it is "woven directly into the audio and video output," which means both the voice and the moving face remain identifiable as AI-generated, even after the content is recorded and reshared.

This detail is easy to skip, but it may be the most important one. In an era of deepfakes, a company that deploys a digital spokesperson needs a way to prove what is synthetic. A watermark baked into audio and video gives enterprises — and regulators — a technical hook for disclosure. Google also describes safety protections designed to "respect identity," alongside the published model card.

Quick comparison: Live Avatar vs the voice assistants you use today

How does this stack up against what you already use? Here is an honest side-by-side, based only on what each company has actually announced.

Capability	Gemini 3.8 Live Avatar	Standard Gemini voice	ChatGPT voice mode	Meta Muse avatar plans
Animated face on screen	Yes, with lip-sync and expressions	No, audio only	No, audio only	Realtime avatar shown at Connect
Mid-conversation language switching	97 languages, no visual drift	Many languages, no face to animate	Many languages, no face to animate	Not detailed
Custom avatar from a photo	Yes, via enterprise allowlisting	No	No	Not detailed
Background data fetching while talking	Yes, asynchronous tool calling	Limited	Limited	Not detailed
Invisible watermark	SynthID in audio and video	SynthID on Gemini content	None announced	Not detailed
Who can use it today	Gemini Enterprise customers	Gemini app users	ChatGPT users	Not launched as a product yet

A pattern jumps out: Google is aiming at businesses first, while Meta is aiming at consumers first. The two companies revealed their avatar projects within 48 hours of each other — Meta on September 23 at Connect, Google on September 24. Read our full rundown of Meta's announcements, including the pocket-sized Muse Charm device, to see the other side of this race. And if tracking this industry is part of your study or career plan, the open programs listed on Truescho are worth a scroll while the news is fresh.

What this means for you

If you are an individual user: nothing changes today. Live Avatar has no free tier and no announced price, and there is no consumer release date. The voice features in the free Gemini app continue as they are. The realistic alternative today is what you already have: standard Gemini voice, ChatGPT voice, or Meta AI.

If you represent a business or a student organization: the adoption path is Gemini Enterprise, not the app store. Start with a preset avatar for a pilot, and treat the custom-avatar allowlist as a later step requiring a direct relationship with Google. The built-in watermark helps with disclosure policies — a growing legal requirement in several countries for synthetic media.

If you are outside the US: availability follows enterprise contracts, not your app store region, and Google did not publish a country list in the announcement. The 97-language claim — which includes Arabic according to the September 15 model launch — makes this genuinely multilingual rather than an English-first feature with translated menus. Gemini is also spreading across Google's wider ecosystem, from study tools to the Gemini-powered laptops we have covered, so the underlying voice model is likely to reach consumers through many doors.

What Google has not said yet

Honesty is more useful than hype, so here is the open-items list. Google has not published pricing, a consumer timeline, or the full list of 97 languages. The company has not explained how the allowlisting process works or which organizations qualify. Privacy questions — call recording, data retention, and how likeness consent is verified for custom avatars — are described only at a high level. And there is no committed date for anything beyond the enterprise launch that happened this week.

Questions people are asking

How much does Gemini Live Avatar cost?

Google did not publish a price in its announcement. Live Avatar ships exclusively inside Gemini Enterprise, a business product, so the cost depends on the contract each organization negotiates with Google. There is no standalone subscription or one-time fee announced for individuals. Treat any specific price you see online as unconfirmed until Google publishes official pricing.

Is Gemini Live Avatar free?

No. The announcement makes it available in Gemini Enterprise only, with no mention of a free tier or consumer release. Free Gemini app users keep the existing voice features but do not receive the animated avatar in this rollout. If Google later brings Live Avatar to consumers, expect pricing details to be published at that time.

How do I get Gemini Live Avatar?

Individuals cannot sign up directly. Access runs through a Gemini Enterprise subscription, and Google points developers to the API documentation to get started. Organizations that want a custom avatar built from a reference image also need enterprise allowlisting, which means coordinating with Google rather than flipping a switch in an app.

Is Gemini Live Avatar available outside the US?

Google did not publish a country list. Because Live Avatar is part of Gemini Enterprise rather than a regional consumer feature, availability follows your organization's access to Google's enterprise products, not your app store region. The 97-language claim, which includes Arabic according to the September 15 model launch, suggests the target audience is global.

What languages does Gemini Live Avatar support?

Google says the avatar transitions across 97 languages while adapting its lip-sync and expressions, without degrading video fidelity or introducing visual drift. The announcement does not name every language. A demo published by The Verge shows a mid-conversation switch between English and Japanese, with mouth movements matching speech in both.

Can companies create a custom Live Avatar from a photo?

Yes. Google says a custom avatar can be generated from a single high-quality reference image, preserving reference likeness, brand styling, or character identity. The catch is access: custom avatars are currently available only through enterprise allowlisting, so companies coordinate with Google instead of using a self-service upload tool.

What is the SynthID watermark on Gemini output?

SynthID is Google's invisible watermark. In Live Avatar it is woven directly into both the audio and the video output, so generated speech and animation remain identifiable as AI-produced even after recording or sharing. Google also published a model card and describes identity protections for the people and brands avatars may resemble.

Does Gemini Live Avatar work for customer service?

Yes, and that is the clearest use case in Google's own demo. The avatar handles a hotel check-in while fetching reservation details in the background using asynchronous tool calling, so the conversation never stalls. Multilingual support, on-screen information, and an always-on watermark make it aimed squarely at enterprise customer-facing roles.

The bottom line

Gemini 3.8 Live with Live Avatar is the most convincing sign yet that voice AI is becoming face AI. The technology is real, it ships today for enterprises, and the 97-language lip-sync claim sets a new technical bar. But it is also a closed door for now: no price, no consumer version, and no self-service path. Watch this space — and if you are building the skills to work with tools like these, explore the scholarships and programs on Truescho or compare programs in our university rankings to find where AI talent is being trained.