Draw Me a Picture: NVIDIA’s GauGAN2 Raises the Bar with a Feature That Converts Text to Images

They say an image is worth a thousand words, but the truth is that NVIDIA’s GauGAN2 only takes a small sentence to create, out of nowhere, a fantastic high-quality image that will completely fit the description you gave it. Do you think it’s magic? Absolutely, is the power of artificial intelligenceAlthough the truth is that the results achieved by NVIDIA GauGAN2 are so good that it is impossible to deny that they have a magic touch.

NVIDIA GauGAN2 is backed by a deep learning model based on artificial intelligence that, thanks to the training it has undergone, is able to understand different types of requests made in writing, and of convert them into images. Thus, for example, if we write “a sunset on the beach”, this artificial intelligence will identify what we want to say and create a highly realistic image that, in addition, will fit perfectly with what we have written.

If we are not happy with the result, or if we want to achieve a more precise adjustment, we only have to add the adjective, the adjectives, that we consider pertinent. For example, in the above we could add “sunset on the beach with stones”, and the image would show a more specific design to fit that new description. We could also change dusk to dawn, and NVIDIA’s GauGAN2 I would adjust the image to that request again.

NVIDIA GauGAN2: How do you convert words to images?

That this NVIDIA platform is capable of identifying what we write and converting it into images is already, by itself, something truly impressive, but that it also does so with photorealistic quality is simply incredible. To do this, they combine mapping segmentation, painting and text-to-image generation in a single GAN model.

Thanks to this approach, NVIDIA’s GauGAN2 is able to work faster and offers, at the same time, very simple operation, so much so that we can say that it is available to any user. He just has to think about which image he would like to recreate and describe it with a set of words, no more. NVIDIA’s GauGAN2 will take care of the rest.

I know what you are thinking, that NVIDIA’s GauGAN2 works well with things “of this world”, but that it could not help us to recreate other types of images that fit within what we can consider as science fiction. Well, nothing is further from reality, since we can introduce very rich and varied descriptions, and then we can use the base that it gives us to add other elements manually.

To achieve these results, NVIDIA’s GauGAN2 received intensive training based on a whopping 10 million high-quality country images. The green giant has confirmed to us that the NVIDIA Selene supercomputer was used, an NVIDIA DGX SuperPOD system that is proud to be located in the top 10 of the most powerful supercomputers in the world.

Regarding the interpretation of words, and their connection with specific images, the researchers resorted to a neural network that he was able to carry out that complicated process. Thus, when that neural network reads “fog” it knows perfectly well which images are associated with that word, and the same happens with others such as “winter” or “sunny”. Yes, the idea is, in essence, the same one on which human thought rests.

It goes without saying that this represents an important advance for designers, creatives and artists, since can help them save time by effortlessly generating high-quality images, on which they can work later, manually, to give them that personal touch they had in mind, and create unique designs more quickly without having to give up a totally professional quality. I remind you that the NVIDIA Canvas application incorporates the latest advances in GauGAN technology, and is compatible with GeForce RTX 20 series GPUs and above.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *