[HELP] Hybrid Developer questions & Dark Mode

[Selur]

Sometimes the Vapoursynth tab is disappearing (and the selector Vapoursynth/avisynth is greyed out non accessible )(when I have select Convert output to PAR SP 1:1 crop and auto asjust) a bug?

If you while enabling CropView Avisynth&Vapoursynth-tabs are disabled.

Can I have a link to download the Hybrid Developer version? (it has more experimental stuff right?)

Dev version usuallly just includes the latest changes (which might break stuff) I made and unless a bug has been reported since the last release which you also ran into, there is no good reason to use the dev version.
I share dev downloads with those reporting bugs for them to test.

Can I have it in parallel with normal version?

Yes, you can have multiple Hybrid versions if you convert them to portable versions. (read: [INFO] *hidden* Hybrid options,...)

Is there any way to turn hybrid in "dark mode" or a plan to implement it? it's a saver for the eyes.... (especially when waiting for the preview to render).

You can adjust the coloring however you want using style sheets. (read: [INFO] *hidden* Hybrid options,... and the linked threads)

For VSGAN where can I find model parameters?

Those parameters are added to the 'vsgan.load(model ...)' invocation so, looking into the source code of VSGAN is your best option. (iirc. not all architectures have unused options)
That option is probably mainly interessting for developers which modified the source code of vsgan.

Is there any way I can fine tune BasicVSR? model parameters?

BasicVSR++ only supports:

def basicvsrpp(
    clip: vs.VideoNode,
    device_index: int | None = None,
    model: int = 1,
    length: int = 15,
    cpu_cache: bool = False,
    tile_w: int = 0,
    tile_h: int = 0,
    tile_pad: int = 16,
) -> vs.VideoNode:

(see: https://github.com/HolyWu/vs-basicvsrpp/...4C1-L33C19)
as options.

(we are missing a good GUI for model training...)

I'm not planing to write one, since:
a. I have no experience with it
b. I haven't seen a good description how to do it
c. I don't have the time and motivation. (Especially for just image based models without a temporal component.)
So, let me know when you write one.

Cu Selur