面向电影前期视觉开发的图像智能生成框架研究OA

In film development,storyboards,concept art,and art assets often require multiple rounds of iteration around character design,prop style,scene atmosphere,and shot composition.Existing global semantic editing and mask-based repainting methods still struggle to balance structural stability and semantic independence in multi-region collaborative edit-ing.To address this issue,this paper constructs ReMAP,an intelligent image generation framework for film pre-production visual development.The proposed method decomposes complex storyboards into independent semantic asset regions and establishes explicit constraint mappings between spatial masks and corresponding prompts,enabling precise local control of film art assets and stable preservation of non-edited regions,while effectively supporting multi-region collaborative re-painting and multi-variant creative comparison.To reduce the cost of prompt engineering in complex scenes,this paper fur-ther constructs a multimodal standard operating procedure prompt generation mechanism,which adaptively extracts visual semantic features from reference assets and original storyboards and automatically generates reusable scene description templates containing global style and regional constraints.Experimental results show that,under the unified Flux2-Klein backbone,ReMAP outperforms baselines such as Prompt-to-Prompt global semantic editing and DiffEdit local mask edit-ing in multi-region semantic isolation,asset structural consistency,and non-edited region stability,providing low-cost and highly interactive intelligent technical support for film art design and storyboard production.