A study on the visual rhetorical differences in national image representation of China and the United States by generative artificial intelligence:An empirical analysis based on large multimodal models

In today’s era of globalization and digitization, generative artificial intelligence (GAI), particularly Large Multimodal Models (LMMs), presents new opportunities and challenges for shaping national image and has become an influential tool in international communication. However, due to data bias, cultural schemas, and algorithmic design, GAI often generates visual representations with symbolic distortions. Drawing on visual rhetoric theory and visual grammar, this study selects two representative LMMs—Wenxin Yige (China) and Midjourney (U.S.) – to analyze AI-generated national identity images across four dimensions: visual theme, color usage, visual features, and symbolic rhetoric. Results show that Wenxin Yige produces relatively uniform, harmonious images using neutral tones and high brightness to convey a stable, dignified identity, while Midjourney favors more emotionally expressive and stylistically diverse imagery, often characterized by high saturation and visual complexity. The study argues that LMMs’ construction of national image is not a direct reflection of reality but a symbolic re-encoding of existing media narratives. By simulating rhetorical functions – selection, emphasis, and framing – LMMs increasingly occupy a “quasi-subject position” in global discourse, acting as technological rhetors that co-construct ideologically saturated, hyperreal images. These findings extend visual rhetoric theory into AI-generated content and offer a critical lens on GAI’s role in shaping international narratives.