OpenAI’s New “DALL-E 2” AI Model Can Now Edit Select Parts of an Image, Generate HD Images

Earlier last year, we saw researchers from the OpenAI consortiumdevelop an advanced AI modelthat can generate colorful and artistic images based on certain keywords and texts. Dubbed DALL-E, the AI-based bot was a digital version of a human sketch artist that could generate accurate images based on descriptive text sentences. Now, the research team from OpenAI has developed the next iteration of DALL-E with improved features and capabilities. Let’s take a look at the details below.

OpenAI Debuts DALL-E 2 AI Model

TheDALL-E 2 AI botis an improved version of the previous DALL-E model that lets users do more with the tool. The newer model offers higher-resolution images and lower latency than its predecessor, which could only generate images at 256 x 256 pixels. The DALL-E 2, on the other hand, can generate 1024 x 1024 pixels images, which is a significant leap over the previous version.

The DALL-E 2 modelleverages OpenAI’s CLIP image recognition system, which is a computer vision system that was announced by OpenAI last year, along with its image generation capabilities to generate an image. The new modelalso allows users to select specific parts of an existing image and instruct the bot to make edits to those partsinstead of making changes to the whole image. This feature was not available in the initial model and lets users create some pretty unique images.

Now, although the DALL-E model was available on OpenAI’s website, it was never released publicly to developers and researchers. But, a lot ofcompanies have taken the concept of the DALL-E bot and implemented it into their models. One of the most popular ones was theWombo Dream app, which generated artistic images based on vague descriptions.

As for the availability of the DALL-E 2 model, unlike its predecessor,the DALL-E 2 bot is currently only available to vetted partners and researchers for testing purposes. However, the tool comes with some limitations that block users from generating images involving nudity, hate symbols, and obscene gestures. Researchers and interested testers can currently join the waitlist to try out DALL-E 2 fromOpenAI’s official website.

Dwaipayan Sengupta

A geek at heart, you’ll find me fidgeting with an iOS device or sitting in front of the laptop, scouring the internet to find interesting technological innovations. In my free time, you’d find me playing Valorant or listening to Pink Floyd.

Add new comment

Name

Email ID

Δ

01

02

03

04

05