现代电影技术Issue(10)：14-20,7.DOI:10.3969/j.issn.1673-3215.2025.10.002
影视虚拟场景智能优化方法研究
Research on intelligent optimization method for virtual scenes in film and televi-sion
摘要
Abstract
Virtual scenes have become the essential vehicle for conveying visual intent in contemporary film and television production.However,when faced with high-complexity semantic expression and stringent style consistency,manual and experience-driven parameter tuning is inefficient and lacks of effective feedback pathways,making it difficult to support high-quality content generation.This paper proposes an intelligent optimization workflow for virtual scenes that leverages the perceptual capabilities of large language models.By constructing an expression unit,generating images,evaluating consistency,and feeding back adjustments,the method forms an adaptive closed loop between semantic goals and ren-dered scene outputs.Language-vision multimodal large language models are used for semantic evaluation and parameter refinement,enabling high-consistency mapping from natural language descriptions to visual output.The workflow offers a generalizable,semantics-driven mechanism that enhances the automation and intelligence in film content creation.Experi-mental results show that the Unreal Engine-based prototype achieves stable closed-loop operation in controlled scenarios,exhibiting good semantic consistency and stylistic uniformity.关键词
虚拟场景/大模型/智能优化/感知反馈Key words
Virtual Scenes/Large Language Model/Intelligent Optimization/Perceptual Feedback分类
艺术学引用本文复制引用
刘梦雅,闫大鹏..影视虚拟场景智能优化方法研究[J].现代电影技术,2025,(10):14-20,7.基金项目
2024年北京市超高清视听产业发展支持项目"国产化超高清中国元素影像生成大模型技术的创新应用"(JLZJ2024040100062). （JLZJ2024040100062）
