CyberFSR2 VS FSRCNN-TensorFlow

Besides, since DLSS is in a DLL, you can actually replace the DLSS DLL with an FSR DLL such as this and make most DLSS games work with FSR. The reason it's not talked about much is because FSR is inferior.

Let me break it down for you. With games that use the standard (NGX) implementation of DLSS, you can use DLSS2FSR to replace the DLSS .dll with FSR, enabling it to run on non-Nvidia cards.

How well does FSR2 work here? If it doesn't work well, are there ways of improving it's quality via CyberFSR?

While I couldn't test it on my GPU, DLSS apparently works pretty well on GPUs that support it. The FSR in-game at the moment is basic FSR 1.0, which works well enough at high resolutions with some of the post FX enabled (Panini Projection, Chromatic Aberration ect). I haven't been able to get the FSR2 mod to work in this game yet, any tips would be appreciated!

It's similar with FSR: https://github.com/PotatoOfDoom/CyberFSR2

By the way thin will be a little random but you can try this on windows to get DLSS on cyberpunk with amd card (convert DLSS on FSR 2.0) https://github.com/PotatoOfDoom/CyberFSR2 I don’t know the type of performance you have with AMD.

And a about fsr if the game don’t supported and run on proton you can force fsr 1.0 (wine only support fsr 1.0). And the hack will be comming to native Linux game in case of Linux native but Linux will be receiving fsr2.1 (https://github.com/GPUOpen-Effects/FidelityFX-FSR2/pull/60#issuecomment-1452817440), thanks amd to make open source. I know a hack exist that covert DLSS into FSR you can try on game only work with DLSS (https://github.com/PotatoOfDoom/CyberFSR2) your problems can be solve with a bit of tweaking and searching.

There's already mods that convert the DLSS to FSR to use on other GPUs: https://github.com/PotatoOfDoom/CyberFSR2

Fsrcnnx

For VLC I don't know, but with MPC-HC you can use madVR [0]. My preferred setup is mpv [1] with a few shaders [2] such as FSRCNNX [3] and Krig Bilateral.

[0]: https://madvr.com/

[1]: https://mpv.io/

[2]: https://github.com/mpv-player/mpv/wiki/User-Scripts#user-sha...

[3]: https://github.com/igv/FSRCNN-TensorFlow/releases

https://github.com/igv/FSRCNN-TensorFlow/releases (in checkpoints_params.7z)

Both are IMO sharper and more accurate than ACNet.

FSRCNN-TensorFlow

Not sure if there is anything newer, but FSRCNNX is probably the highest quality, upscaling only, shader.

You can of course use other shaders like Nvidia's Image Scaling or the popular FSRCNN (not to be confused with AMD's FSR).

As for shaders...well, it depends on what ye are searching for. In general using shaders can be a tad overkill, but in any case I would recommend using FSRCNNX for upscaling, SSimDownscaler for downscaling, and KrigBilateral for chroma. DO NOTE, you will need a relatively powerful CPU (or GPU, depending on your config) to run “heavy” shaders.

These features are not unique to madVR, as mpv also has debanding and "AI" based upscaling, along with their compression removing variants.