OpenAI’s new DALL-E 2 system creates realistic images from descriptions

The AI wants to create new memes, images, and art straight from text.

Artificial intelligence research lab OpenAI has announced its brand new system focused on the creation of images and art using AI. DALL-E 2 is said to be able to create original pieces through the use of text descriptions.

OpenAI published a video detailing how the DALL-E 2 system functions. By reading and recognizing text, DALL-E 2 can combine “concepts, attributes, and styles” to create new works of art.

On the company’s website, OpenAI shared an example of an astronaut riding a horse in a photorealistic style. DALL-E 2 then rendered a photo of just that. Though, when toggling through the variations of styles and settings, you can get a sense of how different the images can be.

Another fascinating example shows the AI altering existing images. Once again, using text and captions, the system is able to add or remove elements from a photo. It can even go as far as altering shadows, reflections and textures.

OpenAI claims that the system has closed the gap between images and text. Using a process dubbed diffusion, DALLE-2 “starts with a pattern of random dots and gradually alters that pattern towards an image when it recognizes specific aspects of that image.”

As one may tell, DALL-E 2 is the new iteration of the DALL-E system OpenAI created in early 2021. While working off the same fundamentals, DALL-E 2 can generate and render images with four times the resolution. OpenAI has monitored DALL-E’s capabilities to prevent it from creating hateful images and other misuses.

Through the use of DALL-E 2, OpenAI hopes creators find new ways to express themselves.

For professional artists, creators, researchers, and developers, OpenAI have a waitlist in order to gain access to DALL-E 2. Those interested can sign up here.

Image credit: OpenAI

Source: OpenAI