SLAB 2: the AI ​​that generates pictures (and not just landscapes) from text

0
7
slab-2:-the-ai-​​that-generates-pictures-(and-not-just-landscapes)-from-text

You may also like this partner content (after ad)


The company Open up AI has brought into the world a new kind of artist. DALLE 2, nicknamed boy, is an artificial cleverness (AI) capable of converting almost any requested circumstance into a picture, starting from a basic textual explanation.

DALLE 2 is a new AI system capable of creating realistic images and sober art From a sober natural language explanation

, can be read on the website of the sober project sober manner sober summary. It is indeed a system able to put, literally, the image words in . It is therefore enough to describe there the image that one wishes to obtain so that the AI ​​composes its spices some select which corresponds.

Recently, GauGAN2, another quite similar AI , had also attracted attention. An editorial staff at Confidence My Technology then checked this tool, which makes it possible to generate landscapes from words. Cuando DALLE 2 ze stands out from this aunty, it is in particular because he is not limited to landscapes. National insurance even, moreover, pictures close to what can be found in reality.

Additional times I actually just need to trip inside the wasteland with my panda motorbike crew. If a lot is observed by you of pandas putting on reddish colored bandanas, that’s us! State hello

Generated with DALL-E 2. #dalle #openai picture.tweets.com/UdW0U8WSzt

Aris Konstantinidis (@ariskonstant) April 6, 2022

This is how Aris Konstantinidis, engineer for Open up AI, was able to generate these Amazing pictures of pandas speeding through the desert, their eyes covered in adorable vintage pilot goggles. Among the pictures put forward to promote the AI ​​in, we also find the koala playing with the container, or even the horse astronaut. creators, combining principles, attributes and designs once again, to create pictures closer to the proposed text. Mira Murati, another employee of the company, was able to obtain the following image for the request: macro photography 550 mm of a large family of mice wearing hats, comfortably installed by the fireside

:

2022DALL-E 2 creates pictures from textual content like macro 35mm movie pictures of a big loved ones of rats wearing caps comfy by the fire place #dalle photo.tweets.com/knLQstkbVz

Mira Murati (@miramurati) Apr 6,

It is also possible for the AI ​​to edit existing pictures. All you have to do is select the area to be changed and indicate what should be removed or added. It can change structure, shadows, textures on the fly This function is added by connection to the first edition of SLAB, released in January 2021. This big brother who started talking about him in January 2022 generates more realistic and accurate images with 4x higher resolution , according to its designers.

Integrate pictures and texts

To deploy all this creativity, DALLE 2 has learned a relationship between the pictures and the text used to describe them, explains Open up AI. As is often the case, what is referred to here under the rather broad term of artificial intelligence could in fact be more precisely called device learning. To learn, DALLE uses what is called the neural network.

A neural network is so named because it is based on a system inspired by the functioning of biological neurons, then moved closer to statistical methods which. Concretely, the learned artificial intelligence fed with a fantastic amount of data could extract logical connections in, and process them for a sober result. A food of this AI was therefore composed of a huge amount of images, associated with text labels. The company’s researchers detail this process in their research work.

This AI also uses a process called diffusion. The idea is to start from a theme of random factors and gradually change this theme to a picture when specific factors are recognized. Obviously, as the company points out, all of that fine creativity can be easily undermined if badly tagged pictures are injected into the system. Like a child learning the wrong word, put an object. Without sufficient safeguards, models like SLAB 2 could be used to generate a huge array of sober content that is misleading and otherwise harmful, and could affect how people perceive the authenticity of sober content generally as well as manner. DALLE 2 additionally inherits various technical scuba biases from its training data, and its outings sometimes reinforce societal stereotypes . For the second, access to the tool is therefore restricted, and you have to register on a waiting list to hopefully test it.