OpenAI is back in business with the new version of DALL-E. And the effect is simply unique; It produces even better images than its predecessors.
OpenAI, one of the world leaders in the field of artificial intelligence, often offers excellent work in this field. Some of his most impressive projects are based on GPT-n; This is a series of linguistic models based on in-depth learning that make it possible to create everything directly or indirectly related to language, from dialogue to computer code.
The system is very powerful, which worries some viewers, but it is not in GPT-3, OpenAI has recently shown itself. Instead, the company released a new version of DALL-E, another AI-based system. It is not necessary to create text; Instead, it creates images from a default language interpretation … and dies as a result.
From text to image, there is only one AI
The first edition, named after the famous painter Salvador Dali and the picture robot Wall-E, is already stunning. This system is already able to visualize objects and people from a simple text description. This first edition was already very active.
It worked with a conceptual approach similar to GPT-3, which gave it better flexibility. It worked with a lot of downline descriptions, but – it’s very exceptional for this type of AI – with a lot of foreign proposals like thisAn example of a radish walking with his dog”.
The first name DALL-E was never given to the general public. But if this idea resonates with you, it may remind you of the functionality of WomboDream, an application based on the first version of the system (see Our post Here)
It allows you to create stunning images from user-provided text and some pre-defined stylistic options. The result was mostly visually very interesting, though not exactly. So that was already surprising. But this second edition simply goes to another dimension.
Second Edition to Die
Since then, it has been greatly strengthened by another AI-based system. The latter is derived from a computer vision system called CLIP; It is used to analyze an image to describe a human being. OpenAI has modified this process to create a CLIP, which guides the composition of an image from words.
Thanks to this procedure, DALL-E 2 can generate images known as the “diffusion” process; It starts with an anarchic dot and then fills the image as it goes, gradually adjusting the size of the details.
Faster, less greedy, generally better enhanced and above all, these DALL-E 2 offers images that will undoubtedly turn some artists upside down. In fact, it now has incredible add-ons that allow you to manage simply many dazzling scenes.
DALL-E 2 is capable of editing existing images with incredible accuracy. So you can add or remove elements while taking into account the color, reflections, shadows … this can be done even while respecting the style of the original image.
This system is capable of generating various variations Release”In different styles like the astronauts above. It can have the same overall style as the bears.
DALL-E 2 can also create fusion of two images with characteristic elements taken from both sources. After all, it can do all this at a resolution of 1024 x 1024 pixels, 4 times more than its predecessor!
Built-in protections to prevent abuse
OpenAI also includes security measures aimed at preventing malicious hijacking of the system. For example, DALL-E 2 refuses to create any images based on a real name. The purpose is to prevent the generation of deepfaqs.
The same thing applies with ezines and newsletters. This applies to all elements closely or remotely associated with “”.Plot “. No question of creating relevant content “For major geopolitical events currently taking place”. Theoretically, such a system could not be used to create ambiguous content; We will not see false images of the Russian-Ukrainian conflict created in this way.
But despite all these obstacles, OpenAI is aware of the potential harm of its best technology. So future users need to show their credentials; You have to go through a lengthy verification process to get partner status. Consumers like WomboDream need to wait a while before this 2nd edition arrives. Too bad for those who already expect it to be fun… but still fair.