In January 2021, OPEN AI introduced DALL-E and then launched its second version last April. On July 20, it announced the availability of this AI system that creates realistic images and art from natural language description in beta. In the coming weeks, one million people on the waiting list will have access to DALL-E 2 and will be able to use it for commercial purposes.

Last year, OPEN AI announced DALL-E, a 12 billion parameter version of GPT-3, trained to generate images from text descriptions, using a dataset of text-image pairs.

The new version, DALL-E 2, would generate more realistic and accurate images with 4 times the resolution of its predecessor. In April, it was introduced to a limited number of users, allowing the company to better understand the system’s capabilities and limitations and to improve its security systems. Reviewers were asked to compare 1,000 generations of images of the 2 models, 71.7% preferred it for caption matching, 88.8% for photorealism.

DALL-E 2 can create original, realistic images and illustrations from a text description and combine concepts, attributes and styles. It also allows you to make realistic modifications to existing images from a natural language caption, add and remove elements while taking into account shadows, reflections and textures.

It uses a diffusion process, which starts with a pattern of random dots and gradually changes that pattern to an image when it recognizes specific aspects of that image.

The move to the beta version

Each user will receive 50 free credits in the first month of using DALL-E 2 and 15 free credits in subsequent months. Each credit can be used for an original DALL-E prompt generation, returning four images, or a modification or variation prompt, returning three images.

In this first phase of the beta, users will also be able to purchase 115 additional credits (460 images) for $15 on top of their free monthly credits. One credit is applied each time a prompt is entered and a user clicks “generate” or “variants.”

Commercial Use

Users have full usage rights to commercialize the images they create with DALL-E 2, including the right to print, sell or market. Examples of commercial projects include illustrations for children’s books or newsletters, concepts and characters for games, moodboards (montages) for design consulting, and storyboards for movies.

Preventing harmful generations

Before making DALL-E 2 available in beta, Open AI worked with researchers, artists, developers and other users to assess the risks and took steps to improve its security systems.

For example, Open AI has limited DALL-E 2’s ability to generate violent, hateful or adult images and used advanced techniques to prevent photorealistic generations of real people’s faces, including those of public figures and politicians. To reduce bias, a new technique has been implemented to generate images of people that are more representative of the diversity of the world’s population, a technique that should be improved with this first beta release.

Translated from DALL·E 2, l’IA de génération d’images d’OPENAI, prochainement accessible en version bêta à un million d’utilisateurs