The Artificial Intelligence DALL-E is taking a step forward in its creative capacity, because it now has a new function that allows it to expand the edges of any image, including classic paintings, such as Girl with a Pearl Earring or Dogs Playing Poker.
The new tool is called Outpainting and according to OpenAI, the company that is responsible for developing DALL-E, it aims to expand users’ creativity by continuing an image beyond its original edges by adding visual elements to it. artistic style, or constructing a new narrative for the works.
Some users who have access to DALL-E – there is a waiting list to use the tool – have already generated interesting images, such as expanding the horizons of the Mona Lisa to turn it into an apocalyptic painting or even to see what else was in Abbey Road on the iconic cover of The Beatles.
How does DALL-E Outpainting work?
To understand how Outpainting works, it is necessary to understand how DALL-E works. It is an Artificial Intelligence capable of creating realistic images or paintings from a natural language description in text, where concepts, artistic styles or characteristics can be combined.
Although a person provides a single description, the AI has the ability to create different variations of the same image and its operation is based on the fact that it has learned the relationship between the images and the text that describes them through a process called “diffusion”. ”.
The AI has been trained with databases of millions of images and what it does is recognize references in text and extract elements or details from each one to form a completely new image.
For example, a person might write the following description: “Teddy bears working on new underwater AI research using 1990s technology.”
So, from all these tools, the AI evolved into Outpainting, which even takes into account elements of the existing image, such as shadows, reflections or textures, to maintain its context and make the result more convincing.
It is worth mentioning that OpenAI is not the only company that is developing this type of artist technology and a very similar example is , by Microsoft, which describes it as “a multimodal generative model designed to generate high-quality images and videos from a text, image or video input”.
The big difference between NUWA-Infinity and these systems is that it can generate long-duration videos through a description, as well as high-resolution images with arbitrary sizes.
On the other hand, Google has Brain, an AI that beyond creating new images, has the purpose of strengthening artificial intelligence systems from data that makes them even more sophisticated and try to reduce the prejudices that these technologies can reproduce. .
