摘要
Abstract
With the expanding use of generative artificial intelligence(GAI)technology in film and television production,vid-eos generated by large models still present structural limitations in color science,including non-standardized color spaces,dynamic range compression,temporal color shifts,and banding artifacts caused by insufficient bit depth.These issues hin-der the direct integration of AI-generated footage into modern film and television industry wide color gamut(WCG)and high dynamic range(HDR)workflows.This paper aims to analyze the intrinsic deficiencies of diffusion models in color en-coding from the perspective of underlying technical principles,and reveals how the implicit sRGB/Rec.709 bias in large-scale training datasets exerts a fundamental constraint on generated image quality.Based on these findings,this paper proposes proactive control strategies,including physically based lighting prompt engineering,LUT-embedded guidance,and reference-image color transfer mechanism using IP-Adapter,and further explores the use of LoRA fine-tuning for lock-ing specific color styles.Experiments validate the feasibility of the proposed strategies,providing theoretical basis and tech-nical references for editable GAI video production in the film and television industry.关键词
生成式人工智能(GAI)/视频色彩空间/扩散模型/前置化控制Key words
Generative Artificial Intelligence(GAI)/Video Color Space/Diffusion Model/Proactive Control分类
社会科学
