一种基于知识增强的迭代式自优化文生视频方法及系统
By constructing a static physical knowledge base and a dynamic constraint memory, and combining static appearance verification with dynamic physical verification, the problem of implicit understanding of physical rules in the T2V model is solved, and explicit injection of physical rules and reuse of historical experience are realized, thereby improving the physical fidelity and generalization ability of video generation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2026-05-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing T2V models lack an explicit understanding of physical rules, resulting in physical illusions in generated videos. Furthermore, they lack dynamic memory and experience reuse mechanisms across sessions, failing to meet the physical fidelity requirements of professional scenarios.
We construct an iterative self-optimizing text-based video system based on knowledge enhancement. Through a static physical knowledge base and a dynamic constraint memory, we achieve explicit injection of physical rules and cross-use reuse of historical experience. By combining static appearance verification and dynamic physical verification, we form a closed loop of knowledge retrieval, prompt generation, verification scoring, and memory accumulation.
It significantly improves the physical rule compliance and temporal coherence of video generation, enhances the generalization ability of out-of-distribution scenes, and achieves continuous iterative improvement in the quality of system generation.
Smart Images
Figure CN122138025B_ABST