Cross-Cultural Visual Traps of Generative AI: A Multimodal Discourse Analysis Based on Comparative Prompt Design

Abstract

Generative AI is playing an increasingly prominent role in the advertising market. While it boosts production efficiency, it also triggers a series of cultural and ethical risks—the data-driven visual generation mechanism is systematically reproducing cultural stereotypes in cross-cultural advertising. This study takes Midjourney, a text-to-image artificial intelligence model, as the research tool and selects the American, Indian and Chinese markets as research objects. Two groups of prompts, stereotyped and de-stereotyped, are designed to generate 72 advertising samples, revealing the visual production logic of AI from the three dimensions of symbols, styles and narratives. The study finds that AI places non-Western modernity in a marked and interpreted other position: at the symbolic level, it presents a "festivalized representation", with the coverage rate of Chinese Spring Festival symbols reaching 75% and Indian ritual symbols being fully covered and stubbornly remaining even under intervention; at the stylistic level, it forms a "strong stylistic binding", with the intervention success rate in India being only 33.3%, far lower than 83.3% in the United States; at the narrative level, it replicates the "hegemonic narrative template", with 17% of Chinese individual images being endowed with additional emotional burdens. The conclusions of this study provide a critical perspective for understanding the cultural politics in the AI era.