Google has introduced a new visual representation for Gemini through Gemini 3.8 Live, which incorporates the Live Avatar feature. This functionality allows an animated persona to appear on screen to interact with the user in near real-time.
This avatar has the ability to synchronize mouth movements with speech, display various facial expressions, and conduct conversations in a more natural way. The announcement of this novelty occurred on Thursday, the 24th, and is already available in Gemini Enterprise.
The technology employed combines low-latency video generation with Gemini's live dialogue features. Consequently, the artificial intelligence is not limited to responding only by voice but also offers a visual presence during the interaction.
According to Google, the Live Avatar was primarily designed for corporate uses, such as customer support and interactive experiences. Furthermore, the system is capable of processing visual and audio data simultaneously during conversations.
Partner companies have the option to choose from a pre-existing collection of avatars or develop custom characters. In the case of customization, developers can use a reference image to generate an animated avatar, maintaining the visual characteristics, character identity, or elements of a brand. However, the creation of these custom avatars is currently restricted to a specific list of authorized companies.
Additionally, Google ensures that all content produced by its AI products is marked with the invisible watermark SynthID, which is integrated into both audio and video, aiming to aid in identifying AI-generated material.
Next advances in AI models
While Google implements this new mode of interaction with Gemini, the company is also preparing for the next major advance in its AI models. Gemini 4 is currently in refinement phase and is expected to launch 'well before' the end of 2026, according to Koray Kavukcuoglu, head of Google DeepMind.
In an interview given to The Information, Kavukcuoglu expressed the company's intention to release an initial version of the model shortly after training completion, as the results achieved so far are very promising. He told The Information: 'Our intention is, like—as soon as possible—to release an initial post-training version because we see the results and we are excited.'
The executive also mentioned that Google plans to maintain a fast pace in releasing new iterations. This information was confirmed by reports released on Thursday, the 24th, indicating that Gemini 4 has entered the post-training stage, a period dedicated to adjusting the model's behavior before its launch. This process includes internal testing and the implementation of safety safeguards.
Kavukcuoglu assumed leadership of Google DeepMind after an internal reorganization of the division. He had previously publicly addressed the development of Gemini 4 in September, detailing the progress of the new model's training.
According to Kavukcuoglu, Google's goal is to remain at the forefront of AI development. When asked about the possibility of the company falling behind competitors, he replied to The Information that, in his view, it is a 'certainty' that Google will always be on the frontier.
The progress of Gemini 4 happens in parallel with the expansion of the Gemini 3.8 family by Google. Gemini 3.8 Live was presented in September as a model focused on near real-time dialogues, capable of processing visual information, operating in 97 distinct languages, and performing background tasks during conversations. With the Live Avatar, Google adds a visual dimension to the experience: besides listening and responding, Gemini now communicates through an animated persona, displaying facial expressions and real-time lip synchronization.