DALL-E 2 is a powerful deep learning tool created by OpenAI. It is based on the paper released by the organization in 2021 titled “DALL-E: Generative Text Understanding and Imagery Synthesis by Natural Language Conditioning.” DALL-E 2 takes Creative Commons images and transforms them into unique shapes and designs based on written descriptions given by the user.
The tool is an extension of OpenAI’s Generative Pre-trained Transformer-3 (GPT-3) model, which was first released in 2020 and is a highly refined version of Google’s Transformer-2 (BERT) architecture. GPT-3 is a language model based on a generative approach to Artificial Intelligence (AI) that can generate new content based on the input it is given. In contrast, DALL-E is a generative autoregressive model, which is composed of multiple layers of a rectified linear unit (ReLU) allowing for the generation of complex shapes and images based on text-based descriptions given by the user.
To use DALL-E, one must enter a written query into the command line interface along with a URL link to an image for reference. For example, a user could input “a tiger with a top-hat” to generate the picture of a tiger wearing a top-hat.
Though the full potential of the tool is still unknown, it has already proven to be a powerful tool for creating unique images and shapes from written descriptions. As such, DALL-E could potentially be used for more creative pursuits, such as aiding the creation of artwork, product designs, and even animations.
DALL-E 2 is an impressive tool created by OpenAI and has the potential to revolutionize the world of computer-generated visuals and positively influence a multitude of industries. It has yet to reach its full potential but already been shown to have an incredibly expansive impact.
