In January 2021, OpenAI introduced DALL·E. Now, the AI research company has unveiled DALL.E2, the latest iteration of its text to image project. DALE2 can create realistic images and art from a description in natural language. The AI system has the capability to make realistic edits to existing images from a natural language caption as well as add and remove elements while taking shadows, reflections, and textures into account. It can also take an image and create different variations of it inspired by the original.
DALL·E 2 learns the relationship between images and the text used to describe them. It uses a process called “diffusion,” which starts with a pattern of random dots and gradually alters that pattern towards an image when it recognises specific aspects of that image.
DALL·E 2 is preferred over DALL·E 1 for its caption matching and photorealism when evaluators were asked to compare 1,000 image generations from each model.
However, DALL·E 2 is a research project and is not available in their API. As part of OpenAI’s effort to develop and deploy AI responsibly, DALL·E’s limitations and capabilities are being tested with a select group of users.
The safety mitigations developed include:
- Preventing Harmful Generations: DALL.E2’s capabilities are limited to avoid generating violent, hate, or adult images and removing such concepts during training itself.
- Curbing Misuse: Filters are placed to curb text prompts and image uploads that may violate our policies.
- Phased Deployment Based on Learning: Access to DALL.E2 is limited to trusted users who will help the team learn about the technology’s capabilities and limitations.
To get on the preview waitlist, click here.