FLUX Model Based Image Generation of Ancient Chinese Artifacts via Low Rank Adaptation

With the rapid progress of AI image generation, output quality has greatly improved. However, traditional models still struggle to capture the unique styles of ancient Chinese artifacts. Most current approaches use the Stable Diffusion (SD) model with Low-Rank Adaptation (LoRA) for targeted generation, but results often appear distorted or unstable when handling culturally complex objects. To address this, we propose a new method combining the FLUX model with LoRA fine-tuning, enabling high-quality generation of Chinese artifacts. We built a rich dataset of artifact images and detailed textual cues, and designed preprocessing and labeling strategies to help the model better learn cultural features. Experiments show that our method significantly improves detail accuracy, color fidelity, and style consistency. This work supports the digital preservation of Chinese heritage and offers new directions for style-specific generative models.