A Survey Report On Text to Image Generator Using Stable Diffusion

Abstract

In recent years, the advancement of artifical intelligence has led to remarkable progress in generating realistic images from textual descriptions. This project introduces “Stable Diffusion”, an innovative text-to-image synthesis model that achieves photorealistic image generation through a unique iterative refinement process. Trained on a diverse dataset of images, the model employs a fixed CLIP ViT-L/14 text encoder to condition image synthesis on textual cues. Stable diffusion employs a stepwise approach, gradually enhancing a random noise image while aligning it with the given text prompt. This iterative process continues until convergence, yielding high-quality images that faithfully represent the text description. The model demonstrates its capabilities aceoss a spectrum of a humans, animals, landscapes, and abstract art. The potency of stable diffusion materializes across diverse domains. From evocative portraits of people and enchanting depictions of animals to sprawling landscapes and abstract artistic expressions, the model encapsulates the intricate essence of textual descriptions, yielding images that extend beyond mere representation.