Google's Talking AI Avatars Go Live for Enterprise Customers

TL;DR

• Google has launched Live Avatar for Gemini Enterprise, allowing businesses to add a speaking, animated avatar to their AI assistants.
• The feature can be used on websites, in apps and at interactive kiosks. Businesses can choose a preset avatar; creating a custom one requires Google’s approval.
• Google says it synchronizes mouth movements with speech across 97 languages and embeds an invisible watermark in generated audio and video.
• Google’s published rates, checked on September 28, put video output alone at about $0.37 per minute while the avatar speaks. Audio and other processing cost extra.

What happened

Google has released a feature that lets businesses add an animated avatar to an AI assistant’s spoken answers. Called Live Avatar, it generates video with mouth movements and facial expressions synchronized to speech, according to the company’s announcement. It became available through Gemini Enterprise on September 24, 2026.

The feature builds on Gemini 3.8 Live, which already supported spoken conversations, visual input and background tasks. The addition is the generated avatar accompanying the spoken response.

Google says Live Avatar can switch between 97 languages during a conversation, adjusting its mouth movements and expressions. It also says generated audio and video carry SynthID, an invisible watermark intended to help identify AI-generated content.

Businesses can deploy the feature on websites, in mobile apps and at interactive kiosks. They can select a preset avatar or apply to create a custom one through an approval and verification process. Google’s demonstration uses instructions, a reference image and an audio sample to create a custom avatar.

Google’s launch material describes an Autotrader assistant that helps shoppers search for and compare vehicles through conversation. It also includes a Salesforce statement about integrating Gemini 3.8 Live with Agentforce, its AI agent platform. That statement does not specifically confirm a Live Avatar deployment.

As of September 28, Google’s pricing table lists avatar video output at $1 per million tokens for deployments using regional endpoints. At its stated conversion rate of 6,192 video tokens per second, that works out to approximately $0.37 per minute while the avatar speaks. The video charge does not apply while it listens. Audio, input processing and other applicable charges are additional.

What this means (and what it does not)

Businesses can now give customers a visible avatar to interact with while an automated assistant answers questions or performs tasks. Google presents customer service and interactive guidance as intended uses.

The enterprise release means people may encounter these avatars through a business’s services. It does not amount to a rollout of avatar conversations in the personal Gemini app: the announced access is through Gemini Enterprise.

What we still do not know

Google’s launch materials do not provide independent evaluations of lip-sync quality across the supported languages, response delays in everyday use, or whether adding an avatar helps people understand answers and complete tasks compared with a voice-only assistant.

The customer examples also leave adoption unclear. They do not establish how many businesses have deployed Live Avatar, how many people are using it, or how it performs over sustained use.