AI video: interactive avatars learn to see the person they're talking to

Video neural networks are moving to a new level: instead of racing for image quality, AI avatars are now learning to see and hear the person they are talking to. According to TNW analysts (July 2026), the video generation industry is approaching a tipping point toward interactivity: the ability of avatars to perceive and respond to a conversation partner in real time.

Why the race for image quality is no longer the main event

For a long time, the only yardstick for an AI avatar was its appearance: convincing skin, believable lighting, smooth lip sync with speech. These metrics still matter, but on their own they no longer determine the leader.

An avatar that looks flawless but does not notice the other person's emotion and does not adapt its intonation to the context remains a video clip: convincing, but lifeless. This is where the next barrier appears: it is not enough to generate convincing visuals; the perception loop has to be closed.

TNW notes that the race is starting to move toward an avatar's ability to perceive the real world and respond to it meaningfully: to see, hear and interpret context.

What are the three levels of interactivity?

The authors break down avatar interactivity step by step, from basic response to commands to full multimodal perception.

At the entry level, the avatar answers a predefined script or text input: it reacts to a command, but not to live context. This is the typical scenario for most current corporate products: video presentations, onboarding clips, synthesized news reports.

The next level adds speech perception: the avatar hears the other person, distinguishes intonation and adapts its answers to what was said. This is closer to real dialogue, but the avatar is still "blind."

The highest level is full multimodal perception: the avatar simultaneously sees, hears and interprets what is happening in the frame. It notices facial expressions, gestures and shifts in the context of the conversation. Its behavior changes in real time in response to what happens in front of the camera.

What scenarios does full interactivity open up?

The move to the third level is not an evolutionary step but a change in the class of problem. It opens up fundamentally new uses:

• A virtual coach that sees a student's facial expressions and adapts the pace of its explanations
• A character in a game or metaverse that recognizes the user and changes its behavior from session to session
• A customer support agent that notices a customer's confusion before the customer has put the problem into words
• A language tutor that responds to the learner's pronunciation and emotional state

None of these scenarios works with a static avatar, however realistic it may be. Here, interactivity is not an option but an architectural requirement.

What this means

Competition in AI video is shifting from the question of "how does the avatar look" to "what does the avatar perceive." Companies that are first to close the real-time multimodal perception loop will gain a durable position in applications where visuals and dialogue are inseparable.

What is the essence of the new stage in AI video generation?

Avatars are shifting their focus from appearance to interactivity. Previously, the main criteria were skin quality, lighting and lip sync with speech. Now avatars must perceive the person they are talking to and respond to them. Those metrics remain important, but they are no longer the deciding advantage.