In the era of generative artificial intelligence, the boundaries of literary reading and interpretation are being fundamentally reconfigured. In this project, I investigate how state-of-the-art AI image generators—specifically MidJourney, DALL·E 3, and Stable Diffusion—“perceive” and visualize canonical Chinese classical poetry, offering both a distant and close reading of poetic imagery across languages and modalities. My research addresses two core questions: How does machine vision transform poetic metaphor into visual representation? And what happens to meaning as it travels through translation, prompt engineering, and algorithmic interpretation?

Building on recent developments in digital humanities and computational literary studies, I select ten widely-anthologized poems, each paired with authoritative English translations, to serve as the corpus. Both Chinese originals and English translations are systematically used as prompts for multiple AI platforms, generating a diverse gallery of visual outputs. Each image set is analyzed through qualitative close reading and semiotic frameworks, revealing how machine-generated imagery reflects or departs from the poems’ symbolic universe. I extend this analysis with quantitative tools: using the CLIP interrogator, I measure text-image alignment and employ embedding models (BERT, CLIP) to transform text and image into a shared semantic space. Through dimensionality reduction techniques (t-SNE, UMAP), I produce embedding maps that make visible the semantic “distances” and clusterings between poetic concepts, translations, and their algorithmic visualizations.

A distinctive feature of my project is its commitment to public engagement and cross-cultural comparison. I design and deploy a bilingual interactive website, which hosts the AI-generated poetry-image gallery, embedding visualization, and an online audience study. In this study, participants from varied linguistic and cultural backgrounds are invited to match images to lines of poetry, rate the resonance and visual accuracy of each AI output, and share qualitative reflections. This large-scale, participatory “distant viewing” experiment foregrounds not only the machine’s perception, but also the social imagination of AI-mediated aesthetics.

Preliminary results point to several significant findings. First, the translation and prompt language decisively influence the AI’s visual “reading,” sometimes preserving, sometimes transforming, and at times distorting traditional metaphors—particularly those grounded in Chinese-specific cultural and sensory contexts. Second, embedding visualization uncovers both expected convergences and surprising semantic gaps between human and machine perception, charting new territory for comparative and multimodal poetics. Third, audience responses reveal diverse pathways of interpretation, suggesting that AI-generated visualizations can both reinforce and productively disrupt conventional readings of classical poetry.

By bridging distant reading/viewing with close, qualitative analysis, my research demonstrates the potential of multimodal digital humanities to expand the horizons of literary study. My methodology, combining generative AI, embedding visualization, and audience-centered analysis, provides a model for rethinking cross-cultural literary interpretation and human-machine collaboration in the age of artificial intelligence.