Artistic Intelligence: A Diffusion-Based Framework for High-Fidelity Landscape Painting Synthesis

Abstract

Generating high-fidelity landscape paintings presents significant challenges, necessitating precise control over both structure and style. This paper introduces LPGen, a novel diffusion-based model specifically designed for landscape painting generation. LPGen incorporates a decoupled cross-attention mechanism that serves as a style controller, independently processing stylistic features to manage the style of the generated images. Furthermore, LPGen incorporates a structural controller—a multi-scale encoder that governs the layout of landscape paintings, achieving a balance between aesthetics and composition. The model is pre-trained on a curated dataset of high-resolution landscape images categorized by distinct artistic styles, followed by fine-tuning to ensure detailed and consistent outputs. Extensive evaluations demonstrate that LPGen outperforms current state-of-the-art models in producing structurally accurate and stylistically coherent paintings. This work advances AI-generated art and opens new avenues for exploring the intersection of technology and traditional artistic practices. Our code, dataset, and model weights will be made publicly available.

Keywords

• Image generation
• decoupled cross-attention
• stable diffusion
• controllability