图像生成式人工智能Midjourney生成的故事绘本的视觉叙事研究

Research on the Visual Narrative of Story Picture Books Generated Using Image-generated Artificial Intelligence Midjourney

中文关键词: 图像生成式人工智能  Midjourney  故事绘本  视觉叙事

英文关键词: background with black clouds cover the sky, snowing, Warm colors, Low brightness, full body, long shot, front view, face to the front --ar 128:61”,生成的图像见图2。

中文摘要:

      目的 伴随图像生成式人工智能技术的成熟，绘本中视觉图像制作的既定形式与过程正发生着改变，为故事绘本中视觉叙事的创作提供了新的方法。然而利用人工智能生成的绘本图像在表象事物，与读者的互动和构成文本等方面的视觉叙事能力还不明确。因此，通过对人工智能生成的绘本图像进行考察，确定其视觉叙事的优势与欠缺。方法 利用图像生成式人工智能平台Midjourney制作《小年兽》绘本的部分页面。从概念功能、人际关系功能、构成功能三方面对比分析Midjourney生成的图像和绘本原图像在视觉叙事上的差异。结果 Midjourney生成的绘本图像可以呈现相似的登场人物视线，表现风格，环境色彩等，较好实现与原文一致的人际关系意义。概念意义和构成意义在登场人物为两名或以下，画面构图简单、连续性好及叙事信息量小的情况下可以被一致呈现。但不能准确生成登场人物数量多，画面构图复杂，场景多变的信息量大的视觉叙事。结论 Midjourney生成的绘本图像可以极好地呈现登场人物的感情和环境的氛围，在传递感染力方面能力强;Midjourney生成的图像可以清晰的演绎简单画面，反过来说只能传递相对单一的视觉叙事，因此Midjourney在算法和大数据方面还有待进一步的优化;用Midjourney生成的‘绘本’只能算是图画书，无法表现绘本中多样的图文关系。目前情况下，一本优质的绘本是不能只通过Midjourney来表现视觉叙事的。

英文摘要:

      With the maturity of image-generating artificial intelligence technology, the established forms and processes of visual image production in picture books are changing, providing new methods for the creation of visual narratives in story picture books. However, the performance of visual narratives in picture book images generated by artificial intelligence is still uncertain in terms of the ability to represent things, the ability to interact with readers, and the composition of the text. Therefore, by examining the images of picture books generated by artificial intelligence, their visual narrative capabilities and deficiencies were determined. The image-generating artificial intelligence platform Midjourney was used to produce some pages of the "Nian" picture book. The differences in visual narrative between the picture book images generated by Midjourney and the original images were comparatively analyzed from three aspects:conceptual function, interpersonal function and textual function. The picture book images generated by Midjourney could generate similar lines of sight of the characters, expression styles, environment colors, etc. to better realize the interpersonal meaning consistent with the original text. Conceptual meaning and compositional meaning could be presented consistently when there were two or fewer characters on the scene, the picture composition was simple, the continuity was good, and the amount of narrative information was small. However, it could not accurately generate pages with a large number of characters and complex or changeable picture compositions. The following conclusions are drawn from this. First, the story picture book images generated by Midjourney can excellently present the emotions of the characters and the atmosphere of the environment, and have strong ability to convey appeal. Second, the images generated by Midjourney can clearly interpret simple scenes, but in turn can only convey relatively single visual information. Therefore, Midjourney still needs further optimization in terms of algorithms and big data. Third, the 'picture book' generated by Midjourney can only be regarded as an illustration book and cannot express the various relationships between pictures and texts in the picture book. Under the current circumstances, it is impossible for a high-quality picture book to express visual narrative only through Midjourney.