A relatively short time ago, in March 2019, the American multinational NVIDIA published a video in which it showed the new advances they are achieving in one of their most innovative projects: GauGAN. This innovative software is capable of transforming simple drawings and color sketches into “photographs” of truly incredible natural landscapes.
Nowadays, Artificial Intelligence takes more and more weight in the technological advances, in the news and in the interests of the big companies, be of the sector that they are. Artificial intelligence promises to revolutionize our daily lives since it has, and will have, thousands of different applications that we are not even capable of imagining today.
In this sense Nvidia has developed a new technology that, although still in the development phase, allows to transform simple drawings that look like made by a 3 year old child into images with great realism.
In this way, as we have seen, the operation and interface are tremendously simple: we make a quick sketch with a brush in the same way as we would in Paint, we assign a material to each area of that drawing, and instantly the software generates the image we are looking for.
GauGAN transforms simple children’s drawings into images of super realistic natural landscapes.
There are many articles that speculate that Artificial Intelligence and all its related variants will put an end to many of the works we know today. Normally, the most marked jobs are those of a more physical or repetitive nature, in which human thought and skills have less weight. But without wishing to discuss the veracity of these opinions, it has always been thought that artistic jobs would be difficult to replace. Nevertheless, we are faced with the proof that even designers, photographers or the enlightened can come to face, or at least coexist, with these new technologies.
But without wanting to create controversy or generate frustration, let’s look at the many applications and uses that this marvelous technology can offer us in a few years’ time.
GauGAN can be a very interesting tool for creating virtual worlds quickly and without having great knowledge of graphic design. In this way, any person could draw and present ideas in a simple and fast way, because with a simple sketch you could create perfectly realistic scenarios in real time, without the need for renderings, plans or sketches. This could greatly facilitate the task of architects and designers when transmitting their initial ideas about a project, since this IA can come to understand how surfaces work in reality and the lights reflected on them: if you paint a lake, it will try to reflect the objects close to it.
But there are other sectors in which this technology can offer great advances, such as the development of video games and 3D experiences. In these sectors, virtual spaces could be generated almost instantaneously to get a first idea of whether what we have in mind is right and needs to be developed further.
NVIDIA thinks that it is much easier and easier to transmit ideas with simple sketches, but also that this technology can turn them into very realistic images.
A trained AI with a million images
Nvidia has trained this artificial intelligence with a million images, and the learning model understands what these photographs look like. It’s actually quite a similar process to human learning, since we are little we have to go seeing, touching and testing everything to know what each thing is.
Something similar happens with this type of learning models, because after “teaching” thousands of different images of grass, snow, water or any other material, the artificial intelligence itself ends up having the ability to apply them in our sketch when we indicate it.
In this way, with the simple fact of selecting that a material is grass or snow, we will achieve that all the image changes, acquires new textures, shines and reflections, and is represented as we were imagining. And if this is not the case, we can change this representation by modifying the sketch.
For the moment, the only publications made by Nvidia show that this intelligence is capable of working only on landscape aspects. However, according to its developers, this technology also allows the introduction of animals, objects, food, means of transport and even people.
The result is, as long as the user does not resort to drawing whimsical shapes away from the real model, surprisingly photorealistic images in most cases. Just look at a few examples: