生成式人工智能（Generative AI）正深刻重塑艺术与人文领域。本文以《诗经》中的色彩意象为文本,对比Midjourney和即梦AI在跨文化视觉转译中的表现。研究发现,AIGC输出结果深受算法架构、训练数据及文化语境的影响：Midjourney受西方中心数据库限制,转译结果流于形式,强调通用美学且易产生AI幻觉;即梦AI依托于中国本土化训练数据与国风美学优化算法,实现文化内涵准确、语义一致的视觉转移。研究还指出提示词工程与人脑智能在缓解AIGC文化偏见、提升创作可控性方面的关键作用,为AI艺术和文化遗产数字化保护提供了理论支持与实践路径。
Abstract
Generative AI is reshaping art and humanities. This study compares Midjourney and ImagenAI in translating color imagery from the Classic of Poetry. Results show outputs are influenced by training data and cultural context: Midjourney, trained on Western data, produces visually appealing but culturally shallow results with AI illusions; ImagenAI, using localized Chinese data, achieves greater cultural accuracy. The study highlights prompt engineering and human oversight in mitigating bias and improving control, supporting AI art and heritage preservation.
关键词
色彩意象 /
AIGC /
视觉转译
Key words
Color Imagery /
AIGC /
Visual Translation
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 刘晓杨,夏铭晨.图像生成式人工智能Midjourney生成的故事绘本的视觉叙事研究[J].包装工程,2024,45(20):346-353.
[2] 向继友. 文生视频类人工智能的意识形态风险及其防控路径[J].江汉论坛,2025(8):67-74.
[3] 赵莹. 基于意象意境论的生成式人工智能审美营造[J].现代传播(中国传媒大学学报),2025,47(3):123-130.
[4] 顾春芳.“意象生成”对艺术创造和阐释的意义[J].中国文学批评,2020(3):47-58+158.
[5] 李发,向仲怀.《诗经》中的意象“桑”及其文化意蕴[J].蚕业科学,2012,38(6):1093-1098.
[6] 杜仲赢. 《诗经》色彩词与情感的联动机制[J].文艺评论,2025(4):86-93.
[7] 许哲娜.从五色文化的视角看传统“天人合一”政治思维[J].天津社会科学,2015(5):139-143+152.
[8] 谢亚兰,陈敏,张振华.基于Midjourney的湘西苗绣文创生成化设计研究与实践[J].家具与室内装饰,2025,32(8):18-22.
[9] Midjourney. Prompt basics[EB/OL]. (2024-05-22) [2024-06-12]. https://docs.midjourney.com/hc/en-us/articles/32023408776205-Prompt-Basics.
[10] Zheng L, Zhang H, Sun Z, et al. A Comprehensive Survey on Efficient Large Language Models: From Architectures to Training Strategies[EB/OL]. arXiv, 2025. arXiv:2504.04865v1.(2025-04-17)[2025-06-12]. https://arxiv.org/html/2504.04865v1.
[11] O‘NEILL R. When AI Doesn’t See Us: A Visual Test of Bias Across Three Leading Tools[EB/OL]. Towards AI, 2024. (2024)[2024-06-12]. https://pub.towardsai.net/when-ai-doesnt-see-us-a-visual-test-of-bias-across-three-leading-tools-cb2699c6dabf.
[12] The University of Hong Kong, Business School. HKU Business School Releases a Comprehensive Evaluation Report on the Image Generation Capabilities of AI Models[EB/OL]. (2024-06-12)[2024-06-12]. https://www.hkubs.hku.hk/sc/media/school-news/hku-business-school-releases-a-comprehensive-evaluation-report-on-the-image-generation-capabilities-of-ai-models/.
[13] 姚梦悦,孙亮.从文化涌现到心灵延伸:AIGC赋能新国风动画的算法文化接受进路[J].视听,2025(16):81-85.
[14] 王金鹏,曼纽尔·卡斯特.从搜索引擎到生成式AI:互联网接口权力的演变与社会重构——访谈曼纽尔·卡斯特教授[J].全球传媒学刊,2025(4)213-224.
PDF(2731 KB)
