Dynamic Visual Creation: Implementing Text to Image Generation

Abstract—Generative Adversarial Network (GANs) has become one of the most interesting ideas in the last years in Machine Learning. Generative Adversarial Network is a very exciting area and that’s why researchers are so excited about building generative models as they are set to vary what machines can do for humans. This paper proposes the generation of realistic images according to their semantics based on text description using a Knowledge Graph alongside Knowledge Guided Generative Adversarial Network (KG-GAN) that comes with the embeddings generated from the Knowledge Graph (KG) into GAN. This project focuses on the development of a dynamic text-to-image generation system that translates textual descriptions into corresponding visual representations. The system leverages Open- AI's language models for processing and understanding user-input text, generating descriptive prompts which are then passed to a Replipt-based model for image synthesis. The back-end of the system is powered by Flask, which serves as a lightweight framework to handle user requests, manage communication between the language and image generation models, and deliver the resulting images. The integration of these technologies enables real-time creation of unique images based on textual input, facilitating a seamless user experience. This work demonstrates the potential of combining advanced AI techniques— natural language processing and image generation—into a cohesive platform capable of generating high-quality, contextually relevant visuals from textual descriptions. The project emphasizes the need for efficient communication between diverse AI models and frameworks, ensuring scalability and adaptability. Additionally, the system allows for further exploration of model training, enhancing image fidelity and accuracy with iterative improvements. Keywords— Knowledge Graph, GCN, GAN, NLP.