Contrastive Learning Approach for Text-to Image Synthesis

Natural language processing models have shown an extensive neural network to perform various text generation tasks. It is also possible to generate visual images based on text prompts using transformers by establishing the relationship between the image caption pairs to create novel and plausible photos for a vast variety of prompts which can produce a high level of abstraction in the generated images. The motive of this model is to establish the generation of ideas based on text prompts by harbouring the strength of transformers by learning the text embeddings of already available images to create a model capable of zero-shot age which means that the model knows the context of the sentences to portray visual representation based on the input. This model can be used for exploring the user’s imagination.