Bringing a Visual Presence to AI

Google has officially introduced its latest innovation for the Gemini 3.8 Live model: customizable AI avatars. These avatars are designed to make digital interactions feel significantly more personal by simulating a human presence during conversations.

This functionality is achieved by merging Gemini's real-time speech processing with low-latency video generation. As the AI speaks, the chosen avatar performs lip-syncing in real-time, matching the output text to create a cohesive visual experience. According to research scientist Shuo-yiin Chang and software engineer CJ Zheng, this development provides the AI with a distinct “visual presence.”


Enterprise Applications and Branding

While the technology may seem futuristic, Google is positioning these avatars primarily for enterprise-level use cases. Rather than internal corporate utility, the company envisions these tools being deployed to engage directly with end users. Potential applications include:

  • Automated customer service support
  • Virtual receptionists
  • Interactive corporate training modules

At launch, businesses will have access to a library of pre-set avatars. Furthermore, companies have the flexibility to design custom avatars that align with their specific branding, such as incorporating company logos into the avatar's attire.


Safety and Multilingual Capabilities

As AI-generated content becomes more realistic, security remains a top priority. Google has integrated SynthID watermarking into these avatar videos to guarantee authenticity. The company noted: «This imperceptible watermark is woven directly into the audio and video output, helping to ensure AI-generated content remains detectable to help minimise misinformation and misattribution.»

In addition to the visual updates, Google announced significant performance improvements for Gemini 3.8 Live:

  • Broad Language Support: The model now supports 97 languages with automatic switching, facilitating seamless multilingual communication.
  • Enhanced Efficiency: The new “Extended Thinking” version of the model has demonstrated superior results in various benchmarks compared to competitors like GPT-Live-1 Astra and Grok Voice Think Fast 2.0, while maintaining lower operational costs per hour of audio processing.