Scholarlink

<p>Artificial Intelligence-Generated Content (AIGC) has revolutionized the world of digital creation with the help of automated image colorization, style transfer, and animation. However, several issues still exist, such as poor colorization accuracy, unbalanced content-style fusion, temporal inconsistency, and high computational complexity. All these challenges hinder the integration of AIGC into creative workflows. To address these issues, an effective Creative Network-Generative Adversarial Network (CreaNet-GAN) framework is proposed. The Dynamic ShuffleNet model improves feature extraction with dynamic convolutions, GeLU activation, and a dual-branch design to improve spatial context and computational efficiency. In the colorization part, the generator of Context-aware Conditional GAN (CAC-GAN) uses multilevel context-aware attention and spatially adaptive modulation to produce accurate and realistic outputs, while the discriminator performs fidelity enforcement through local and global feature-based evaluation. A dual-path Transformer-based stylizer refines the content-style fusion using attention mechanisms and skip connections to avoid identity loss and distortions. Moreover, an Adaptive Temporal DenseNet with Adaptive Residual Temporal Convolutional Network (ARTCN) and Latent Condition Block (LCB) guarantees temporal consistency and identity preservation in generated animation. Experimental results demonstrate the framework's efficiency, delivering superior visual quality with a high Structural Similarity Index (SSIM) of 0.986, a high Peak Signal-to-Noise Ratio (PSNR) of 37.04&#xa0;dB, and a low Mean Squared Error (MSE) of 0.139. These results highlight the framework's potential to set new standards for AIGC-driven digital art creation.</p>

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

Enhancing AIGC-driven creativity: a CreaNet-GAN approach for digital art colorization and animation

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Xinyu Deng,
  • Xianghai Hui

摘要

Artificial Intelligence-Generated Content (AIGC) has revolutionized the world of digital creation with the help of automated image colorization, style transfer, and animation. However, several issues still exist, such as poor colorization accuracy, unbalanced content-style fusion, temporal inconsistency, and high computational complexity. All these challenges hinder the integration of AIGC into creative workflows. To address these issues, an effective Creative Network-Generative Adversarial Network (CreaNet-GAN) framework is proposed. The Dynamic ShuffleNet model improves feature extraction with dynamic convolutions, GeLU activation, and a dual-branch design to improve spatial context and computational efficiency. In the colorization part, the generator of Context-aware Conditional GAN (CAC-GAN) uses multilevel context-aware attention and spatially adaptive modulation to produce accurate and realistic outputs, while the discriminator performs fidelity enforcement through local and global feature-based evaluation. A dual-path Transformer-based stylizer refines the content-style fusion using attention mechanisms and skip connections to avoid identity loss and distortions. Moreover, an Adaptive Temporal DenseNet with Adaptive Residual Temporal Convolutional Network (ARTCN) and Latent Condition Block (LCB) guarantees temporal consistency and identity preservation in generated animation. Experimental results demonstrate the framework's efficiency, delivering superior visual quality with a high Structural Similarity Index (SSIM) of 0.986, a high Peak Signal-to-Noise Ratio (PSNR) of 37.04 dB, and a low Mean Squared Error (MSE) of 0.139. These results highlight the framework's potential to set new standards for AIGC-driven digital art creation.