ThumbGen – Thumbnail Generation using Text-Image Synthesis

Abstract

The major issue faced in text-image synthesis is the difficulty in the conversion of ideas into detailed images. Most of the content creators and digital marketers face difficulty in converting their ideas into visual aids. In addition, generating thumbnail images manually is time consuming process and increases the workload of the end users. Furthermore, the existing research faces significant challenges, such as lack of relevancy, and low-quality image generation. Therefore, there is a need for a system to generate high-quality thumbnail images automatically. To overcome the above-mentioned problem, in this research, the author implements a thumbnail image generation system which takes text descriptions as user input and generates high-quality thumbnail images which are relevant to the input text description. The developed system is able to generate thumbnail images automatically, which saves the time of end users and reduces the effort. The implemented system was able to achieve a promising result by achieving a CLIP loss rate of 0.6. This system is able to generate high-quality images, ensuring the relevancy between the generated image and the input text description. In addition, the system allows the users to download the generated images. However, the system takes a certain time to generate an image, which affects the efficiency of the system.