Texture filtering: Difference between revisions
No edit summary |
|||
| Line 147: | Line 147: | ||
*Looks pixelated, highlighting the low resolution. | *Looks pixelated, highlighting the low resolution. | ||
*Textures scaled horizontally and vertically at unequal levels can produce awkward results. [https://old.reddit.com/r/finalfantasytactics/comments/10iqeph/i_spent_a_couple_of_hours_testing_ppsspps/ Resolution hacks] may be able to fix this in some instances. | *Textures scaled horizontally and vertically at unequal levels can produce awkward results. [https://old.reddit.com/r/finalfantasytactics/comments/10iqeph/i_spent_a_couple_of_hours_testing_ppsspps/ Resolution hacks] may be able to fix this in some instances. | ||
|- | |||
|'''Spline36''' | |||
|A high-order mathematical interpolation method. Unlike Bilinear, which only looks at the immediate surrounding pixels, Spline36 uses a larger 6x6 grid of pixels to calculate new color values. | |||
| | |||
*Sharper and more accurate than Bilinear/Trilinear. | |||
*Reduces "aliasing" (stair-stepping) on textures compared to simpler filters. | |||
*Very computationally efficient on modern hardware. | |||
| | |||
*Can occasionally produce "ringing" artifacts (faint halos around high-contrast edges). | |||
*While sharper than Bilinear, it still looks like a "blurrier" upscale compared to dedicated pixel scalers like xBRZ. | |||
|- | |||
|'''NNEDI3''' | |||
|A neural network-based "edge-directed" interpolator. It uses a pre-trained model to predict what missing pixels should look like based on existing edges and structures. | |||
| | |||
*Exceptional at preserving sharp edges without the "oil painting" look of xBRZ. | |||
*Virtually eliminates jagged edges (aliasing) on 2D elements and UI. | |||
*Considered one of the highest-quality "clean" upscalers available. | |||
| | |||
*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. | |||
|} | |} | ||