OpenAI launched DALL-E in January 2021, a neural network that, according to OpenAI’s principal scientist and co-founder, Ilya Sutskever, can “take any text and build an image out of it.” This includes ideas it might not have come across during the training.
Now, the company has returned with DALL-E, an even more amazing version of the prior program. “DALL-E 2 is a new AI system that can create realistic visuals and art from a description in natural language,” according to OpenAI’s website.
Astronauts on horses, among other things
The website provides numerous examples of the AI system’s work, all of which are amazing. For example, the image above was created using the terms “an astronaut,” “riding a horse,” and “in a photorealistic style.”
What’s remarkable is that the AI system recognizes where the astronaut should be mounted and what a photorealistic image is. OpenAI goes on to say that “From a natural language caption, DALLE 2 can perform realistic modifications to existing photos. It has the ability to add and remove components while taking into account shadows, reflections, and textures.”
Examples demonstrate how the application may add characteristics or remove them totally from a photograph while maintaining its quality. Finally, “DALLE 2 may take an image and build numerous variations of it influenced by the original,” according to OpenAI. This is demonstrated with a painting that is somewhat changed in each rendition while maintaining the overall qualities of the picture.
How does DALL-E manage to achieve all of this?
According to OpenAI, “DALLE 2 has figured out the connection between images and the text that describes them. It employs a technique known as “diffusion,” which begins with a pattern of random dots and gradually transforms it towards a picture as it recognizes certain characteristics of that image.”