What Google announced on September 24

Google announced Gemini 3.8 Live with Live Avatar as a way for enterprise agents to pair live dialogue with near-real-time video. The system can listen, process visual and audio input, and respond with generated speech and an animated persona. Google says the feature is available in Gemini Enterprise; custom avatar creation from a reference image is currently limited to enterprise allowlisting.

That boundary matters if you are reading the headline as a new AI companion app. The announcement describes an enterprise conversational capability, not a consumer roleplay service, a character catalog, or a promise that the feature is available in every Gemini surface. Availability can differ between the Live Avatar announcement and the separate Gemini Live model rollout.

Sources

The notable shift is visual presence during live dialogue

A video avatar changes how a conversation is presented: instead of receiving only text or audio, a user can see an expressive on-screen persona while speaking with an AI. Google describes speech, visual input, turn-taking, lip-sync, and background tool calls that can continue while dialogue is active. Those are interaction capabilities; they are not, by themselves, evidence of a more consistent personality or a better long-running story.

Google also says its system can transition across 97 languages and that generated output is watermarked with SynthID. These are provider statements about this release, not an independent comparison with other voice agents. The practical questions for a builder are therefore not only whether the avatar looks convincing, but whether the conversation stays clear when interrupted, what the system can see, and how generated media is identified.

Sources

An avatar is one layer—not the whole companion

For a roleplay or companion experience, visual presence could make a spoken scene feel more embodied. That is an inference about how the feature might be used, not a use case Google says it has launched for consumers. A companion still needs decisions about character behavior, user control, memory across sessions, privacy, safety boundaries, and what happens when a model misunderstands the scene.

It helps to separate three layers: the conversation model handles responses; the interface presents them through text, voice, or an avatar; and the product determines identity, continuity, account controls, and data practices. A more expressive interface can improve one part of the experience without resolving the others. The related guide to roleplay memory and retrieval explores that separate continuity problem.

What users and builders should check next

If you are evaluating a live-avatar demo, start with access and scope: is it a public product, a developer API, an enterprise feature, or a limited preview? Then test the conversation itself rather than judging a polished clip alone. Use the same short scene, interrupt it, refer to a visual detail, correct a mistaken assumption, and see whether the system makes its limits clear.

For any avatar that can be customized from a real person’s image, consent and likeness controls deserve equal attention to animation quality. Also check what video and audio are processed, how long they are retained, what the watermark does and does not tell viewers, and which regions or accounts can use the feature. Google’s announcement does not answer every deployment or privacy question for every customer, so consult the current product documentation before building around it.

A practical checklist

  • Availability: which product, account, region, and access tier?
  • Interaction: can it handle interruptions and corrections naturally?
  • Continuity: does the experience preserve character facts between sessions?
  • Privacy: what happens to camera input, voice, and reference images?
  • Transparency: can people tell when audio or video is AI-generated?
Sources

The short version: more presence, with important scope limits

Gemini Live Avatar is a meaningful interface development: Google is bringing real-time visual presence together with live conversation for enterprise agents. Its relevance to AI companions is a useful design signal, but it should not be mistaken for proof that a consumer roleplay companion, persistent character memory, or broad access has arrived. As of September 30, 2026, check the official product surface and account eligibility before treating the announcement as something you can try.

Jess takes a different route: conversations happen in private Discord DMs, and the Jess World homepage explains the current product and plans. The related links below provide the product overview and a guide to Discord-based companions.

Sources and further reading

Research and product details were reviewed September 30, 2026 where applicable. Vendor policies, plans, and features can change; verify current details with the linked first-party source.