档案信息化2026年第40卷第5期《档案学研究》
构造拟态历史记忆：大模型驱动的档案文本语境影像生成机理研究
Constructing Mimetic Historical Memory: A Study on Mechanism of Contextual Image Generation from Archival Texts Driven by Large Model
苏州城市学院城市治理与公共事务学院 苏州 215104
摘要
生成式人工智能技术的发展使得以较低成本、较高效率实现档案场景记忆的可视化回溯成为可能。在回顾档案记忆回溯技术发展历程的基础上，提出“档案文本语境影像生成”的学术概念，定义评价档案文生语境影像质量的“拟态真实性”原则，解析“真实性缺口”的形成机理，设计体系化的生成方法，通过实例验证其可行性。“档案文本语境影像生成”是指档案管理机构以向用户提供多模态、可视化、可体验档案服务为目标，依托特定类型人工智能语言大模型和视觉大模型，从档案自然文本中抽取并构造提示词，将曾经存在于档案形成者记忆当中，通过文本符号记录和描述的历史语境信息转化为符合拟态真实性原则的可视化数字图像或数字视频的技术过程。“档案文生语境影像生成”在档案文化展览、档案文献遗产活态化开发、档案领域记录片或影视剧作品创作、历史元宇宙工程和辅助历史领域学术研究等领域具有广阔的应用前景。
关键词：AIGC文生图文生视频档案文本语境拟态真实性提示词工程
Abstract
The development of generative AI technology has made it possible to achieve the visual retrospection of archival scene memories at lower cost and higher efficiency. Based on a review of development history of archival memory retrospective technology, the academic concept of contextual image generation from archival text is proposed in this paper. Then, the mimetic authenticity principle for evaluating the quality of archival text-to-contextual images is defined, the formation mechanism of the authenticity gap is analyzed, a systematic generation approach is designed, and its feasibility is verified through a practical example. Contextual image generation from archival text refers to a technical process in which archival management institutions, aiming to provide users with multimodal, visualized, and experiential archival services, rely on specific types of large language models and large visual models, to extract and construct prompts from natural archival texts, and convert the historical contextual information that once existed in the memory of archive creators and was recorded and described through text symbols into visualized digital images or videos that conform to the principle of mimetic authenticity. Archival text-to-contextual image generation has broad application prospects in domains such as archival cultural exhibitions, dynamic development of archival document heritage, creation of documentaries or film and television works in the archival field, historical metaverse projects, and assisting academic research in the historical field.
Key words: AIGC; text-to-image; text-to-video; archival text context; mimetic authenticity; prompt engineering
引用格式
参考文献
