Text-to-Image Generator with Consistory

Artificial Intelligence (AI) and computer vision technologies have evolved rapidly, enabling new creative possibilities. One of the most exciting developments is text-to-image generation, where a system creates realistic images based on a natural language description. However, traditional models often struggle with subject consistency, making it hard to generate images that represent the same character across different scenes. Our project, “Text-to-Image Generator with Consistory,” addresses this problem using NVIDIA’s Consistency Model, known as Consistory. This model enables consistent subject generation across multiple scenes without needing retraining. We developed a web application using Streamlit for the frontend and integrated NVIDIA’s Consistency Model API as the backend. The system allows users to input a subject description and two different scene prompts. It generates high-quality, visually coherent images, maintaining subject identity across scenes. The application is useful for storytelling, branding, education, and entertainment.