Here’s how AI is used to improve the level of detail in games

The term Neural rendering At the moment it is unknown to the masses, but over time it will become more prominent. And it is that behind this name hides the concept of using the advantages of artificial intelligence for the generation of graphics. Either by computer vision or by Deep Learning. In recent years we have seen reconstructions of images at a higher resolution or with a greater color range, but other utilities are beginning to be seen, such as the possibility of predicting the lighting in a 3D environment to be able to apply ray tracing or in the case it concerns us to be able to create elements of smallest detail in an automated way.

Level of detail in 3D graphics and AI

You are playing a racing game and unfortunately the rival car has just passed yours in the distance and it is showing less and less detail. It appears to us to be the same model, but there are actually several that are loaded into memory that must have been modeled by an artist during the design phase. Because visual fidelity is lost, those representing the car at the furthest point end up using fewer vertices and, therefore, fewer polygons.

Well, one of the things that games are going to benefit from will be the automatic creation of models with a lower level of detail than the originalsimply through the use of computer vision of the most detailed model. That is, through a simple observation of the model, the AI ​​will be able to create a chord model at a shorter distance that can be used as modeling in the game. In addition, it can include how to generate it procedurally, which means that instead of containing the graphical data of the model, what it will have is the information to generate it automatically with less detail and using different styles for it.

Like current generators, but with vertices instead of pixels

The concept is the same as that used in the AI ​​image generators that have become fashionable and is based on the idea that an algorithm is developed that allows you to learn to observe it and then reproduce it. It has been tested with two-dimensional images, but keep in mind that we will also see it in 3D models made with vertices.

Obviously, the AI ​​is not completely foolproof and has limited reproducibility and is often buggy. Professionals will continue to be needed to polish and give the correct form and artistic personality to the work. However, it will serve to get rid of tedious work. Not to mention that it will speed up design times and not just in games. Who tells you that the industrial design of your future car will not be generated in part by an AI?

Generation of 3D worlds by photography

Another of the applications of all of this, although it’s more of a combination of everything we’ve seen that’s going to be revolutionary in the mass market, is the fact that through a photograph, a three-dimensional world can be generated. That is to say that the Artificial Intelligence after seeing a photo reconstructs it as a 3D model with textures and everything.

This, which in principle may not be useful and is no longer a mere curiosity, has very wide applications in the face of simulation and so-called virtual twins. Which are computerized versions of real environments used in simulations. What is the ideal flow of people to avoid falls and misfortunes during a concert? Which route is the most efficient for building a new highway? All this will save you from having to build the scene in 3D, just one or more photos will suffice.

Even after the destruction of a district or a building, it will be possible to rebuild it as it was originally or to perform historical reconstructions. How about a virtual trip to your favorite city at a specific time? Well, all of this is possible through the use of artificial intelligence such as GANvers 3D.