Scholarlink

This study aims at technical bottlenecks such as style transfer distortion and weak model generalization caused by cross-platform data heterogeneity in AIGC image generation, and proposes an image art generation system based on dynamic acquisition control optimization. Its core architecture includes (1) multi-modal adaptive acquisition module, which realizes 10 categories of art through the DANet network Semantic alignment of platform data to build a million-level data set covering 37 art genres; (2) Hierarchical generation control engine, using conditional diffusion model and parameter hyperplane mapping technology to achieve 256-dimensional fine-grained control; (3) Coupled adversarial training framework, through Style-Consistent The Loss function improves the structural similarity of the generated image. Verified by the COCO-Stuff and WikiArt datasets, the system has a maximum generation speed of 29.4 FPS at a resolution of 512 × 512, an average style transfer accuracy of 90.55% in the MIT-Adobe FiveK benchmark, and a parameter adjustment error rate of 1.1%, significantly advancing the engineering application of AIGC technology in the field of artistic creation.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Design of Image Art Creation and Collection System Based on AIGC

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lin Shan

摘要

This study aims at technical bottlenecks such as style transfer distortion and weak model generalization caused by cross-platform data heterogeneity in AIGC image generation, and proposes an image art generation system based on dynamic acquisition control optimization. Its core architecture includes (1) multi-modal adaptive acquisition module, which realizes 10 categories of art through the DANet network Semantic alignment of platform data to build a million-level data set covering 37 art genres; (2) Hierarchical generation control engine, using conditional diffusion model and parameter hyperplane mapping technology to achieve 256-dimensional fine-grained control; (3) Coupled adversarial training framework, through Style-Consistent The Loss function improves the structural similarity of the generated image. Verified by the COCO-Stuff and WikiArt datasets, the system has a maximum generation speed of 29.4 FPS at a resolution of 512 × 512, an average style transfer accuracy of 90.55% in the MIT-Adobe FiveK benchmark, and a parameter adjustment error rate of 1.1%, significantly advancing the engineering application of AIGC technology in the field of artistic creation.