Application of AI Technology in Art Concept Map Based on Intelligent Information Technology

Abstract

This study presents a dual-path collaborative generation framework for art concept maps based on multimodal inputs, including textual descriptions, hand-drawn sketches, and style reference images. The system integrates a precise control flow, leveraging ControlNet for structural and semantic alignment, with a creative inspiration flow guided by a style memory module to enhance artistic expression. Both streams are adaptively fused via a gated network, enabling dynamic weighting across spatial regions and generation stages. Experimental validation on a curated high-resolution concept art dataset demonstrates that the proposed framework achieves a Fréchet Inception Distance (FID) of 18.73 and a CLIP score of 0.321, outperforming baseline models including Stable Diffusion + ControlNet (FID 24.80, CLIP 0.293), Midjourney (FID 20.10), and LoRA-based variants (FID 23.50). Subjective evaluation by 150 professional concept artists confirms superior semantic accuracy, structural controllability, and stylistic consistency. By interpreting the dual-path framework as a multi-node signal acquisition and propagation system with dynamic fusion control, the study provides an engineering-oriented methodology for multimodal information fusion, precision guidance, and high-fidelity generation of structured visual content.