In partnership with

LUXE PROMPTING

LIVE VIDEO NEWS | SEPTEMBER 24

Gemini 3.8 Live Avatar Arrives for Enterprise Teams

Google announced Gemini 3.8 Live with Live Avatar on September 24: a conversational agent with a generated visual presence. The launch is for Gemini Enterprise, and custom avatars require enterprise allowlisting. Here is what creative teams can prepare before proposing a pilot.

The agentic era needs a different CRM. That’s Attio.

Teams like Parallel, Turbopuffer, and Wordsmith are already setting the pace on Attio. Get an always-on revenue engine, with agents and workflows that build pipeline, chase every buying signal, and move deals forward with your team. Whether you're working in your browser, inbox, or favorite agent, connect to your customer data in real-time through Attio's web app, MCP, API, and SDK.

The Short Version

  • This is a responsive conversation, not simply a rendered talking-head clip.
  • Confirm enterprise access and custom-avatar approval separately.
  • Start with one task, an authorized character, and a short evaluation plan.

What changed, and what did not

Google describes speech paired with near-real-time generated video, preset characters, and custom characters created from reference images. A finished promotional video follows a prepared sequence; an interactive avatar must respond to the person speaking. That changes the creative brief from a script alone to a conversation with boundaries.

This is not a consumer Gemini-app walkthrough. We have not tested the feature, measured latency, or verified access for your organization. Google's demonstrations are vendor examples, not our results.

Read Google's September 24 announcement

1. Give the avatar one useful job

Choose a small, reversible task: explain three product options from an approved fact sheet, for example. Write the question it should answer, the information it may use, and the point where a human takes over. Do not start with a character expected to handle every customer problem.

Draft five realistic questions and two awkward ones. Include a request outside the task and a question the fact sheet cannot answer. Decide what a useful admission of uncertainty sounds like before judging the avatar's appearance.

2. Prepare a character and permission brief

If pursuing a custom character, gather only reference material your organization is authorized to use. Record the intended audience, character identity, voice direction, and who can approve changes. Keep identity approval separate from whether a picture looks technically suitable.

For an early concept review, a clearly fictional character avoids presenting a colleague as the system's speaker. Decide how the experience will identify itself as AI, offer a text alternative, and let people leave the interaction. These are planning choices, not promises about built-in controls.

3. Test the conversation, not just the face

Google's model card says Live Avatar supports a few minutes of continuous interaction, rather than extended hours. It also notes possible hallucinations, slowness, and timeouts. Plan a short pilot with an obvious recovery path instead of promising an always-on attendant.

During an approved trial, record whether answers match the source sheet, interruptions are handled acceptably, and failed requests reach a person. Ask participants to complete the same task with a text-only alternative. A more expressive interface is useful only if it helps them finish.

Check the model card's known limitations

A copyable pilot brief

Help me draft a pilot plan for an AI avatar that explains [one product or process] using [approved source sheet]. The audience is [audience]. Produce five realistic questions, two out-of-scope requests, a clear AI introduction, and a human-handoff rule. List which claims need verification and which permissions or access approvals remain open. Treat this as a planning document, not a tested deployment or a promise that the platform supports every proposed control.

Before anyone commits a budget

Ask your enterprise contact to confirm availability, custom-avatar approval, current pricing, supported deployment arrangements, and data handling for the proposed use. Keep unanswered items in the brief. Your useful output today is a testable concept with clear limits, not a launch promise based on a demo.

Source Desk

Official announcement, developer overview, and model card checked September 26, 2026. Planning steps are original editorial guidance. No hands-on test or access approval is claimed.

Google: Introducing Gemini 3.8 Live with Live Avatar (September 24)

Google Cloud: Gemini Live API overview

Google DeepMind: Gemini 3.8 Audio model card

A clearer brief. A more useful result.

Privacy policy