From Brushstrokes to Texture: Content Analysis of Qi Baishi's Shrimp Paintings and Exploration of AI Reproduction

Abstract

This study explores the regeneration of Qi Baishi’s renowned shrimp paintings using artificial intelligence, guided by a systematic content analysis framework. We first deconstruct the master’s style by identifying and categorizing key visual elements, including line fluidity, ink intensity, dynamic expression, and negative space composition. These artistic features are then translated into descriptive labels to train a specialized Low-Rank Adaptation (LoRA) model based on a diffusion framework. The generated works are evaluated for stylistic fidelity and emotional resonance, with results indicating a high degree of similarity to Qi Baishi's originals in morphology and dynamism. However, the study also notes limitations in the AI's ability to replicate the nuanced subtleties of ink techniques and deeper emotional expressions. This research validates the effectiveness of content analysis in bridging traditional art aesthetics with computational generation, offering a new pathway for the preservation and innovative reinterpretation of cultural heritage.