Google introduced Live Avatar for Gemini 3.8 Live on September 24, adding generated video to spoken AI conversations. Launch coverage identifies Gemini Enterprise as the product’s availability channel.
A visual layer for voice agents
Google’s developer documentation describes talking avatars whose video is synchronized with synthesized speech. Developers can pair built-in avatars with supported voices through the Cloud console or Gemini Live API.
Custom avatars can be generated from a reference face image, but Google limits that option to selected customers. Its documentation makes customers responsible for the necessary consent and rights. This is distinct from general access to the supplied avatar options.
Limits beyond the demonstration
The Gemini 3.8 Audio model card says the avatar variants support a few minutes of continuous interaction rather than hours. It also notes possible hallucinations, slow responses and timeouts.
The model card describes audio, video and text output for Live Avatar. A visually fluent response therefore remains generated model output, with the same need to check factual claims.
For businesses evaluating a customer-facing assistant, session length, response accuracy and handling of customer information are practical considerations alongside appearance. The release establishes a new interaction format; it does not independently demonstrate better customer-service outcomes.
Featured image: AI-generated conceptual illustration.
