The non-profit organisation OpenAI has unveiled a new version of its text-to-image generator that creates realistic images at higher resolution and with lower latency than the original.
Our newest system DALL·E 2 can create realistic images and art from a description in natural language. See it here: https://t.co/Kmjko82YO5 pic.twitter.com/QEh9kWUE8A
— OpenAI (@OpenAI) April 6, 2022
In DALL-E 2, users can select and edit specific regions of existing images, add or remove elements along with their shadows, create collages and variations of completed drawings.
The neural network generates images at a resolution of 1024 pixels — four times higher than the original model.
The DALL-E service (a portmanteau of the artist Salvador Dalí and the animated character WALL-E) is based on the OpenAI CLIP computer-vision model, announced in 2021.
“The original model simply took a GPT-3-style language approach and applied it to the creation of images: we fed the images into a bag of words and learned to predict what would come next,” said OpenAI researcher Prafulla Dhariwal.
As of today, DALL-E 2 is available to testers who have signed up on the waitlist. Users are not allowed to create obscene or extremist imagery, as well as material related to “current geopolitical events”.
Earlier this year, in January, a machine-learning engineer created a Pokémon generator based on the DALL-E model.
In August 2021, an enthusiast developed a reduced-version of OpenAI’s text-to-image generator.
Follow ForkLog on TikTok!