To address the insufficient application of traditional Chinese-style fur elements in modern new Chinese-style design and issues of poor detail retention, semantic deviation, and overfitting in existing style transfer methods based on Stable Diffusion, this study proposes a FLUX-based reconstruction method for Chinese-style fur clothing to realize modern reconstruction and high-quality style transfer of traditional elements.
With FLUX as the core, integrate FLUX-Low-Rank Adaptation (LoRA) and FLUX-Redux; analyze Chinese-style fur design elements, build a fashion fur dataset, train the FLUX-LoRA model, perform conditional encoding of fur elements via Redux, and verify feasibility through qualitative experiments, Structural Similarity Index, and Peak Signal-to-Noise Ratio.
The FLUX-LoRA model generates text-matching fashion fur designs with good generalization; Redux enables natural fusion of traditional elements. This method outperforms IP-Adapter and StableDiffusion_v1.5+LoRA in detail, consistency, and realism.
Combining FLUX with LoRA and Redux for Chinese-style fur reconstruction solves issues of traditional Stable Diffusion, provides a feasible path for applying Chinese-style fur in modern new Chinese-style design, and contributes to protecting and inheriting Chinese traditional fur clothing culture.
