Vivix Unveils A1 — the First Real-Time Model for Interactive Conversations with Virtual Characters
Vivix
Chinese company Vivix announced the A1 model, which enables real-time video calls with virtual characters. The 30 billion activated parameter model runs on consumer GPUs thanks to new compression methods and infrastructure. Also unveiled is W1, a tool for story management.
Vivix has unveiled A1, the world's first interactive multimodal real-time model based on a unified streaming architecture. A1 enables users to communicate with virtual characters via video calls: the characters can move, change poses, interact with the environment, and respond to the user's voice with an average latency of 0.6 seconds. The model uses approximately 30 billion activated parameters, but thanks to MJD (Multidimensional Joint Distillation) technology, which compresses sampling steps from 8 to 2, as well as the use of NVFP4 and its proprietary VMI infrastructure, it can generate over 10,000 video tokens per second on a single consumer-grade graphics card. Additionally, Vivix introduced W1, a tool that allows users to influence the storyline and character actions in real time. Beta testing is available through the official website.
Source: QbitAI 量子位 —
original