ArtisanFlow: Style Customization System for Illustration Generation
Zeyu Liu, Taiheng Ye, Yiming Li, Song Chen, Yi Xu · 2025
We propose ArtisanFlow, a framework designed to enable illustrators to seamlessly incorporate their personal artistic styles into AI-assisted image generation workflows. ArtisanFlow leverages Low-Rank Adaptation (LoRA) fine-tuning, achieving high stylistic fidelity while maintaining compatibility with popular generation pipelines such as ControlNet and Layer Diffusion. Our approach introduces multi-dimensional evaluation metrics, assessing style, semantic similarity, as well as generation stability and flexibility. Furthermore, we integrate tailored prompt engineering strategies to enhance layered composition and pose guidance. Experimental results demonstrate that ArtisanFlow can effectively capture artistic details, delivering outputs that closely resemble the artist’s original works. Moreover, the modular and node-based interface offers easy customization, with reasonable computational demands that do not require enterprise-grade hardware, while maintaining a user-friendly experience for both individual artists and creative teams.