TECH NEWS – NVIDIA has detailed the three models that DLSS 5 will use, allowing developers to alternate between different visual presentations even during gameplay. The technology can process characters, objects, and entire scenes separately, while studios retain control over the intensity of the AI-based modifications.

Although the controversy was relatively short-lived, it is worth remembering that the issue surrounding DLSS 5’s “hyper-realistic and misplaced faces” is still ongoing while NVIDIA attempts to properly focus its new upscaling technology. Four months after its official unveiling, the Californian company used its appearance at SIGGRAPH to explain how, following the considerable criticism it received, it intends to preserve developers’ artistic vision through a tool that can apply different models and levels of processing to characters, objects, and even individual scenes. In other words, developers will be able to decide how the visuals of their games change.

DLSS 5 Will Be Defined by Customization

According to the company, DLSS 5 will initially feature three different models for managing AI-generated imagery, identified as Model A, Model B, and Model C. Each has been trained to provide a different interpretation of what appears on screen. Studios will not be required to use one model throughout the entire game, as they will be able to switch between them for individual levels, environments, or cinematics according to the visual style they want at a particular moment, with no apparent limitations.

One of the technology’s most important new features will be the level of customization it provides. Developers will be able to modify the intensity of the models and the final finish of the image separately. Certain settings will affect shadows, reflections, ambient occlusion, and the scattering of light across materials, while others will modify broader elements such as lighting, color, and the overall appearance of a scene. DLSS 5 will also be able to identify characters automatically through AI, allowing the effect to be applied only to them or, alternatively, leaving the environment completely untouched while modifying only foreground elements.

NVIDIA has also confirmed that manual adjustments can be created directly within the graphics engine to control DLSS 5’s behavior on individual objects or groups of elements. During the event, the company demonstrated how different levels of processing could be applied to bottles, grapes, jugs, and other objects within the same scene. The technology could even be disabled completely on certain elements while being intensified on others. According to NVIDIA, the goal is for AI to complement traditional rendering rather than replace it.

However, NVIDIA did not address the criticism from players, although it did explain the three major challenges it had to overcome while developing DLSS 5: preserving the creators’ original artistic intent, ensuring temporal stability between frames without introducing artifacts, and maintaining high performance. To achieve this, the technology will continue to begin with the original frame rendered by the game engine, which will still define the geometry, lighting, and composition of the scene, before applying any modifications. The objective is to avoid altering the project’s artistic direction.

Source: 3DJuegos