A method for implementing real-time video generation and interaction system of virtual human on terminal

By employing a lightweight virtual human video generation network and multi-process collaborative technology on terminal devices, the high computational resource requirements and network dependency issues caused by the high complexity of existing models are resolved. This enables low-cost, low-latency real-time virtual human video generation and interaction, improving the stability of the generated videos and the user experience.

CN122437979APending Publication Date: 2026-07-21杨润青
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
杨润青
Filing Date
2026-06-09
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing virtual human video generation solutions rely on highly complex generation models, resulting in long inference times and high computational resource requirements. This makes it difficult to achieve low-cost, low-latency real-time interaction on ordinary terminal devices, and the generated results are easily affected by network conditions, leading to an unstable user experience.

Method used

A lightweight virtual human video generation network with few parameters and low complexity is adopted. By combining static feature encoding and dynamic driving feature extraction, and through multi-process collaboration and nested smoothing processing, real-time video generation and interaction on the edge are achieved, reducing the demand for computing resources. A real-time throttling mechanism with a shared clock is introduced to ensure the stability of video timing.

Benefits of technology

Low-cost, low-latency real-time virtual human video generation and interaction were achieved on ordinary terminal devices, improving the synchronization accuracy of virtual lip movements and the naturalness of facial expressions, reducing video jitter, and enhancing the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122437979A_ABST
    Figure CN122437979A_ABST
Patent Text Reader

Abstract

The application discloses a kind of to realize end on virtual person real-time video generation and interactive system's method, belong to virtual person video generation technical field, including following content: loading virtual person model and related module;Obtain driving data, while loading target virtual person virtual person basic data and idle latent variable;Virtual person basic data is encoded to obtain static feature, and dynamic driving feature is obtained by analyzing driving data;Static feature and dynamic driving feature are uniformly embedded and represented;Uniform embedding representation is input into light-weighted figure video generation network, generates at least one of timing feature, implicit parameter, video frame parameter, further generates virtual person video frame sequence;The generated timing feature, implicit parameter or video frame parameter is nested and smoothly handled, and the virtual person video frame sequence is optimized;The application uses the above method, improves virtual person shape synchronization accuracy, expression naturalness and video timing stability, reduces deployment operation cost and network bandwidth dependence.
Need to check novelty before this filing date? Find Prior Art