Source
- URL: https://x.com/NOTimothyLottes/status/1871630300222259212
- Author: NOTimothyLottes (@NOTimothyLottes)
- Posted: 2024-12-24 18:53:31
Thread
1/ @NOTimothyLottes
Been playing with the spatial energy redistribution for half a decade for CRT-stylized scaling. Solving 2x area scaling was hard, had to use a 3-pixel vertical pattern. Example shader here: https://www.shadertoy.com/view/lfcBDs (left=bilinear, middle=CRT-scaling, right=nearest)
2/ @NOTimothyLottes
We can engineer CRT-inspired spatial scaling techniques that make it more acceptable to render at lower resolutions without leaning on either {scaling-TAAs like DLSS2+, or scalars like FSR1}, instead just try to make the panel actually look like a lower resolution panel!
3/ @NOTimothyLottes
Nearest 1:2x2 scaling tends to fail in that it amplifies the step function, which sets off all sorts of alarms in the mind with false edges. Where just inserting near black lines and correcting for the energy loss, tends to invoke the brain to reconstruct the image instead
4/ @NOTimothyLottes
Interesting to see positive reviews for a scaling TAA in 2024 that wouldn’t do well at 4x area, but hits a better balance at 1440p-2-4K, in less temporal flicker and less ghosting. Seems like people actually favor stability more than they realize consciously or want to admit
5/ @NOTimothyLottes
So many games are now designing around high scaling-TAAs, and I think it’s time people seriously consider the other options, actually making no-ghosting no-thin-flicker a priority and finding other ways to scale output, that’s why I built that DBM scalar (as an example option)