A new artificial intelligence, developed by OpenAI, makes it possible to create a photorealistic from words. It can even replace an element of an image, or even take those of a photo to build a photo seen from another angle.
Have you always dreamed of being able to materialize a thought immediately with an image? It is the ambitious promise ofOpen AI, which has just released version 2.0 of its artificial intelligence, intended from the start to generate images. This artificial intelligence new generation, which responds to the sweet name of DALL-E 2 may well change the future. She decodes the words of a sentence to turn them into photorealistic. And this is the whole evolution of this AI: the first draft already achieved the feat of drawing your thoughts, but more soberly. This will be about “true” photos…straight out of your mind. Fabulous, right?
Founded by Elon Musk, the company is in the small papers of Microsoft, from which it received investments. This is not the first time that a high-tech company has offered us to convert our thoughts into images (and also the reverse). Nvidia had already invented the artificial intelligence GauGAN, which made a sketch or a sentence, a photorealistic landscape. But the DALL-E 2 AI attacks in another register. Indeed, the image thus created does not represent just a landscape, it is based on other real objects, elements of everyday life. Even if for the moment, the descriptive aspect of the examples given seems a little sketchy (“Un teddy bear skateboarding in Times Square”)this AI is very promising, because it goes further.
A complex neural network
Yes, because the whole system is based on a neural network, such as our brain (on a larger scale though): CLIP. This network is designed and trained to store and recognize a large database and images, along with their description. Originally designed to sift through an image and give an accurate description of it, AI can do the opposite. The image is created after a decoding which allows the coherence of the image (yes, because otherwise it might look like a scrapbooking collage).
This new generation of artificial intelligence would also make it possible to fool the human eye, by replacing an element in an existing image, by another element. Neither seen nor known ! It would even be able to give the visual of a photographed scene, seen from another angle, for example. Artificial intelligence has been restricted to the human face, to prevent its use from being diverted from its initial missions. The company also issued another limit, that of producing pornographic or violent images.
We would love to give thanks to the author of this write-up for this amazing material
