Once again, technology firm OpenAI demonstrated a new innovative program inspired by a previous model of DALL-E, a text-to-image generation architecture.
DALL-E 2 by OpenAI
The new system, also known as DALL-E 2, functions at a lower-latency rate and higher resolution than its predecessor. The improvement is evident, as the program could materialize an image based on a user's descriptive input.
In addition to its text-to-image capacity, the DALL-E 2 offers more functions, such as editing a finalized image. However, the new system is not yet available for public use instead of OpenAI's first release.
According to The Verge, curious people and researchers may sign up to join a preview of the DALL-E 2. Experts from OpenAI expect that the program will be utilized in third-party apps on any device.
OpenAI introduced DALL-E back in 2021. The system is inspired by the popular artist Salvador Dali and the fictional robotic character known as WALL-E. During the time, the system was still under development but had already demonstrated what might be the best text-to-image processing algorithm of modern times.
DALL-E can display almost anything that users require it to draw, including bizarre descriptions such as 'giraffe made from turtle' or 'radish walking a dog.' During its presentation, experts guaranteed a lot of improvements in the technology and assured that problems such as misinformation and biased image generation would be attended to.
Since its launch, the DALL-E concept has undergone several adjustments to computation load and resolves technical safeguards.
DALL-E 2's new architecture allows a better inpainting capability that is accurate up to the last detail of the image. The feature can help people customize an existing image, select a specific area, and command the system to curate something out of that space.
Surge of Text-to-Image Technologies
Replacements of any part of the image are possible. In addition, the fill and deletion of complex objects can also be processed by DALL-E 2. Users can search for an image or an idea that does not exist to include it in the image.
DALL-E 2 also allows a combination of two distinct pictures to harvest elements from each subject. The size of a generated image could range up to 1,024 x 1,024 pixels, which is significantly higher than the original system's 256 x 256 pixels.
DALL-E 2 is embedded with a separate CLIP system, which OpenAI also launched in 2021. OpenAI research specialist Prafulla Dhariwal explained that DALL-E 1 was combined with the GPT-3 language and customized to carry out images tasks.
The GPT model is a model that is being utilized by many applications specializing in AI texts. With the help of CLIP, the program can perceive objects similar to how humans do in the real world, producing a more concrete and precise interpretation of images.
The DALL-E 2 works relatively the same way the free application called Wombo Dream does. In this separate program, users can type in words that will be processed by the application to form associated images through abstract shapes.
RELATED ARTICLE : Artificial Intelligence System by French Startup NukkAI Defeats Eight World Champions of Bridge Card Game
Check out more news and information on Artificial Intelligence in Science Times.
