TRANSLATION OF SKETCH INTO IMAGES BASED ON TEXTUAL DESCRIPTIONS

In creative industries such as fashion, architecture, gaming and art, professionals often begin with rough sketches.A reliable system for automatically generating realistic images from sketches could greatly enhance efficiency, reduce iteration time, and unlock new creative possibilities.Sketch to Image focuses on developing an image generation system that converts textual descriptions and hand-drawn sketches into realistic images using advanced deep learning techniques.Existing methods include Style GANs, Cycle GAN, Variational Autoencoders (VAEs) and other recent advancements such as DALL-E and CLIP, which have demonstrated the potential to generate photorealistic images, but often face challenges in preserving fine details, accurately interpreting complex descriptions, and ensuring consistency across generated images.This work aims to address these limitations by combining the strengths to enhance both image quality and fidelity.Employing Generative Adversarial Networks (GANs) in conjunction with Diffusion based models, the system learns to generate high-quality images from text inputs by training on diverse datasets of paired text and images.The proposed system aims to bridges the gap between textual and visual content, empowering artists and designers with a powerful tool to enhance creative processes across various industries.By quickly generating compelling visuals from simple descriptions or sketches, it accelerates workflows in fields like advertising, game design, and virtual reality that demand rapid production of high-quality images.