<![CDATA[Background of study: Recent breakthroughs in Artificial Intelligence (AI) have significantly advanced text-to-image generation, enabling machines to convert natural language descriptions into realistic visual outputs. Stable Diffusion has emerged as a promising solution, offering high-fidelity results with improved controllability and accessibility. To leverage these strengths, this study introduces PixelCraft, an AI-powered text-to-image generation system designed to support creative, educational, and industrial applications.Aims: The purpose of this paper is to design, develop, and evaluate PixelCraft, an intuitive AI system that generates coherent images from textual prompts using Stable Diffusion.Methods: PixelCraft integrates a Stable Diffusion pipeline implemented using Hugging Face libraries and wrapped in a Tkinter-based graphical interface for seamless user interaction. The system processes user prompts, executes diffusion-based denoising stages, and outputs generated images that can be viewed and saved. A structured evaluation was conducted using widely accepted performance metrics, including CLIP similarity scores, Fréchet Inception Distance (FID), and Structural Similarity Index Measure (SSIM). Comparative analyses were performed against models such as BigGAN, VQ-VAE-2, and DALL·E-2.Result: Experimental findings show that PixelCraft achieves strong semantic alignment and visual coherence, yielding an average CLIP score of 0.95, an FID score of ~15, and an SSIM of 0.91. These results outperform several benchmark models, demonstrating superior consistency across both simple and moderately complex prompts.Conclusion: PixelCraft effectively demonstrates Stable Diffusion's ability to generate high-quality images from natural-language descriptions. The system provides a practical, accessible platform for artists, educators, and digital content creators, significantly reducing barriers associated with traditional design tools.]]>
