Texture filtering: Difference between revisions

@@ Line 177: / Line 177: @@
 *Extremely hardware intensive compared to every other filter on this list; requires a capable GPU.
 *Can sometimes introduce small "wobbles" or artifacts in very complex, noisy textures where the AI misidentifies an edge.
+|-
+|'''ArtCNN'''
+|A lightweight Convolutional Neural Network (CNN) designed for real-time upscaling and artifact removal. It uses machine learning to reconstruct sharp edges and smooth gradients.
+|
+*Produces some of the cleanest results for 2D art and UI elements.
+*Excellent at removing "ringing" and compression artifacts that other scalers might amplify.
+*Provides a much more natural, "hand-drawn" look compared to the "oil painting" effect of xBRZ.
+*Highly optimized; much faster than NNEDI3 while providing similar or better visual clarity.
+|
+*Requires a GPU that supports modern shader languages (Vulkan/GLSL/HLSL).
+*Can occasionally "over-clean" textures, removing intentional film grain or fine grit in 3D textures.
+*May be too demanding for very low-end integrated graphics or older mobile devices.
 |}