DLSS renders the game at a lower resolution then upscales it to a higher resolution using AI to modify the final image. The models are trained from super-resolu

DLSS renders the game at a lower resolution then upscales it to a higher resolution using AI to modify the final image. The models are trained from super-resolution renders and used to be trained per-game but are now generalized. They're designed to upscale textures, denoise, fix aliasing and overall increase quality based on that high-end source data so you get the same output without having to actually run at those settings.