Interrupting an AI character mid-sentence, and seeing it adapt

The Vivix-A1 model makes it possible to interrupt a virtual character mid-gesture—an advancement for integrating responsive avatars into branding projects.

Avatars remain for the most part talking busts, capable of moving their lips and performing a few conventional gestures, without understanding anything about the space around them. Vivix, a company that develops interactive video models, offers with Vivix-A1 characters that move around a scene, manipulate objects, and react to what they see and hear.

The claimed breakthrough lies in the rhythm of the exchange. Avatars usually function through turn-taking: you speak, you wait, you watch the response play out to the end. Vivix-A1 receives new instructions while speaking, moving, and acting, which allows for the possibility of interrupting it or redirecting the conversation at any time. Asking it to cross a room, turn toward an object, pick it up, and bring it over is part of the same continuous movement.

Instructions are delivered via voice, text, or image, even after the interaction has begun. A photo sent along the way is not limited to freezing the first frame: showing a product leads the character to bring it into the scene, grab it, and present it. On the rendering side, the model produces audio and video in small, continuous segments rather than whole clips, with the publisher presenting this streaming generation as the prerequisite for a seamless exchange, maintaining continuity across the character, the setting, the voice, and the movement.

The rendering covers several styles, from photorealistic to 3D characters and Japanese animation, with expression engaging the face, gaze, voice, and entire body. Vivix targets entertainment, commerce, training, customer support, and companionship, with the stated ambition of transforming video from content you watch into an experience that unfolds with you.