ReChar: Revitalising Characters with Structure-Preserved and User-Specified Aesthetic Enhancements

Despite recent advances in generative models, artistic character generation remains an open problem. The key challenge is to balance the preservation of character structures to ensure integrity while incorporating aesthetic enhancements, which can be broadly categorized into visual styles and user-specified decorative elements. To address this, we propose ReChar, a plug-and-play framework composed of three complementary modules that preserve structure, extract style, and generate decorative elements. These modules are integrated via a fusion model to enable precise and coherent artistic character generation. To systematically evaluate artistic character generation, we introduce ImageNet-ReChar, the first large-scale benchmark for this task, covering multiple writing systems, diverse visual styles, and over 1,000 semantically grounded decorative prompts. Extensive experiments show that ReChar outperforms state-of-the-art baselines in structural integrity, stylistic fidelity, and prompt adherence, achieving an SSIM of 0.8690 and over 93% human preference across all criteria.