Generative AI for Text to Image

Abstract

Text-to-image (TTI) synthesis models represent a creative approach in the realm of artificial intelligence, specifically designed to transform textual input into visually realistic images. The essence of TTI generation lies in its ability to harness the power of language and convert it seamlessly into visually compelling content, showcasing creative image synthesis. Initially using GANs and transformers, text-to-image generation evolved with diffusion models introducing noise. Integration with large models, TTI models now produce results near-real images. Breakthroughs like ControlNet and 3D object synthesis redefine possibilities. Editing text or images adds versatile dimensions, showcasing generative technologies' transformative capabilities. The survey explores scaling TTI models, focusing on various descriptions and ControlNet's role. The authors categorize literature, offer nuanced comparisons, and discuss applications. Looking ahead, they foresee TTI's potential for productivity enhancements, especially in the Metaverse era, and expansion into intricate tasks like video and 3D generation.