Advances in Generative AI for Visualizing Classical Chinese Poetry: Prompt Design and Multimodal Evaluation

This study examines how generative AI visualizes classical Chinese poetry through systematic prompt engineering, analyzing three representative works across various poetic styles. For Li Bai's 'The Difficult Road to Shu,' we developed baseline prompts using classical translations, mood-enhanced prompts with parameter adjustments, and style-transfer prompts leveraging ChatGPT's text reinterpretation, demonstrating AI's capacity for diverse artistic expression. With Li Qingzhao's 'A Dreamy Chant,' we refined prompt design strategies through verse-by-verse visualization, emphasizing scene construction, cultural element integration, and compositional control to capture the subtle emotional atmosphere of graceful Song-dynasty lyrics. The study's most significant contribution stems from Su Shi's 'Ode to the Red Cliff,' where we established a multimodal evaluation framework that integrates assessments from literature experts, painting specialists, and ChatGPT. This framework reveals profound cognitive divergences: ChatGPT prioritizes technical metrics and stylistic completeness, literary experts emphasize emotional depth and philosophical contemplation, while painting specialists focus on traditional technique authenticity. These findings illuminate the subjective nature of art evaluation and validate an iterative 'creation-evaluationoptimization' model for human-AI collaboration. By demonstrating how large language models can facilitate complex knowledge analysis through strategies like knowledge prompting and chain-of-thought prompting, this research provides a viable framework for integrating AI with literature and art education, transforming abstract literary concepts into visual experiences while advancing multimodal teaching methodologies.