Google introduces talking avatar feature for Gemini 3.8 Live

Google has released Live Avatar for Gemini 3.8 Live, a feature that gives its voice AI assistant a visible form. During a conversation, an animated character speaks, synchronizes its lip movements with the audio and changes its facial expressions in real time. This means users not only hear the voice but also see a digital character responding to them on the screen. For now, the new feature is available only to Gemini Enterprise customers.

Live Avatar combines instant speech processing with low-latency video-generation technology. Gemini 3.8 Live processes visual input almost immediately and can display additional information on the screen during a conversation. The model can automatically switch between 97 languages and maintains visual quality when moving from one language to another. During a demonstration, the avatar spoke in English and Japanese, with its mouth movements matching both languages.

Google developed the new avatars primarily for customer service, virtual reception, training and other business uses. Companies will be able to choose from ready-made characters or create an avatar that matches their own style. For example, a company's logo can be added to the character's clothing.

An invisible SynthID marker is added to the audio and video to prevent AI-generated visuals from misleading people. This digital marker makes it possible to determine later that the content was created by AI. Google also says it has implemented additional safeguards to prevent the identities of real people from being misused as avatars.

Live Avatar became available approximately 1 week after the Gemini 3.8 Live model was announced. Gemini 3.8 Live Extended Thinking, which has a more powerful reasoning mode, outperformed the GPT-Live-1 Astra and Grok Voice Think Fast 2.0 models in several tests. It was also announced that the cost of processing 1 hour of audio is lower.