Modders have attracted AI to improve textures in games.

    Artificial intelligence (of course, its weak form) allows you to automate the solution of a huge number of tasks and, in general, simplify human life. Most different branches of science and technology are already using technology based on artificial intelligence. The game industry also works with AI, in particular, to improve the “intelligence” of the NPC and the laws of the game universe.

    But AI can also be used to optimize graphics. Cases are known when technologies helped to restore an old, almost spoiled photo or to update some old film recorded on a damaged film. But there is another possibility - the improvement of graphics in games.

    Many modders are engaged in creating fashions for old popular games or updating the games themselves with improved graphics. As an example, Resident Evil 4 with HD textures. The game with the updated graphics is much closer to modern gamers than its original version.

    But to update the graphics in the whole game you need a huge amount of time, so modders often do their work for years. Now, for this task, you can involve artificial intelligence. Recently, the network has information that Doom 2 and Morrowind have already been released with exactly this updated graphics.

    In the case of Doom 2, the author of the novelty, hidfan, used the specialized tools of Nvidia Gameworks, as well as the AI ​​Gigapixel tool from Topaz Lab. Both services are needed in order to improve the quality of textures, then return to the previous resolution with the removal of AI artifacts. True, some artifacts and transparency need to be configured manually. By the way, according to the same hidfan, AI cannot deal with Black & White.

    As for the second game, Morrowind, a whole team of modders were working on it, using ESRGAN (Enhanced Super Resolution Generative Adversarial Networks). This tool allows you to improve the image with not too high-quality resolution.



    According to the representatives of the team, after a couple of tricks the service starts working, making the images more realistic. In particular, the texture of Morrowind was improved thanks to the ESRGAN. An example of the image before and after processing - below. Texture quality was improved fourfold.



    Currently, this kind of technology is only in its infancy, but its potential is clear. If you “train” artificial intelligence to improve the graphics of old games, gamers around the world will get a large number of classic games in the “new wrapper”. If the AI ​​learns to improve the quality of the video, then Star Trek: Deep Space 9, Babylon 5 lovers will be able to watch their favorite episodes with double pleasure.

    Above mentioned tools from NVIDIA, and not in vain. This company has long been working on the integration of AI capabilities in work with graphics and video. Earlier this month, the company introduced a software platform that allows you to combine the effects generated by artificial intelligence with the graphics engine of games. The basis of the platform is a neural network of the GAN type (generative contention network). Using the same neural network, NVIDIA has learned to generate people's faces in unprecedented quality.



    This technology can be used not only for games, but also to create movies or VR-content. As for the images placed above, they are taken from the “game” that AI created. Initially, the system was trained to drive on robotic algorithms. Then, when the learning process was completed, the AI ​​trained to identify various objects, including the sky, cars, trees, roads, marking, etc. As a result, the platform created its own universe based on the Unreal Engine 4. The graphics in this world are generated in real time.



    However, even this modern technology is only at the beginning of its journey - it can take a long time to commercialize it. Nevertheless, NVIDIA has achieved a lot, if this company will improve its product, in a few years we can talk about the possibility of using the platform for solving practical problems.

    Also popular now: