Google has introduced avatars for its Gemini Live AI system, giving enterprise customer service bots a more humanlike visual presence. The avatars, powered by real-time speech and low-latency video generation, lip-sync to text output from the AI.

What You Need to Know

Gemini Live Avatars use real-time video generation to sync lip movements with speech. They are designed for enterprise use cases such as customer support and training. Google says the avatars include SynthID watermarking to verify they are AI-generated. The new model also supports automatic language switching across 97 languages.

How the Avatars Create a Visual Presence

The avatars come from a preset library and can be customized with company branding such as logos on uniforms. Research scientist Shuo-yiin Chang and software engineer CJ Zheng described the effect as giving AI a more visual presence. Businesses deploy these avatars for their own end users, making interactions with chatbots feel less robotic.

The technology combines Gemini's real-time speech capabilities with low-latency video generation. The result is a digital face that appears to mouth the words the AI is speaking. Google positions the feature primarily for enterprise use cases like receptionists, training modules and customer service portals.

Safety and Cost Efficiency

Google embeds SynthID watermarking into both audio and video output of the avatars. The company describes the watermark as imperceptible but detectable, helping to minimize misinformation and misattribution. The avatars, however, are clearly labeled as AI-generated through this technical measure.

  • Live Extended Thinking: A more powerful version of Gemini 3.8 Live that outperformed GPT-Live-1 Astra and Grok Voice Think Fast 2.0 across multiple benchmarks
  • Cost reduction: The Live Extended Thinking model costs less per hour of input audio than competing systems
  • Language support: The base Gemini 3.8 Live model now automatically switches between 97 languages in real time

Why This Matters

The introduction of avatars pushes AI interaction closer to human conversation but also raises new trust concerns. Anthropomorphism can make customers more comfortable with automated support, yet it may also blur the line between human and machine. For enterprises, the lower cost of the Live Extended Thinking model could accelerate adoption of AI-driven customer service. Regulators will need to consider whether visual avatars require additional transparency rules beyond current watermarking. The balance between engagement and deception will shape how these tools are received in the market.

Google's move signals that the company sees visual presence as a competitive advantage for enterprise AI. As more businesses deploy Gemini Live Avatars, the success of the feature will depend on user acceptance and clear labeling of AI-generated content.