In April 2022, OpenAI announced DALL-E 2, a system that creates realistic images and art from a written description. The accompanying research paper, “Hierarchical Text-Conditional Image Generation with CLIP Latents” by Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen, was submitted on April 13, 2022. The paper describes a two-stage approach that links OpenAI’s CLIP image-text understanding model with a diffusion-based image generator.
DALL-E 2 could produce higher-resolution and more coherent images than its predecessor, and it could also edit existing images and create variations of them, all driven by plain-language instructions. The OpenAI announcement page presents the product and its capabilities to a general audience.
DALL-E 2 was a turning point for public awareness of generative AI for images. It demonstrated that anyone could describe a scene in words and receive a usable picture, which reshaped expectations for creative tools and helped trigger the wave of text-to-image products that followed in 2022.