Restore/Improve old anime quality

[doomleox999]

I'm trying to restore/improve the quality of the movie End of Evangelion. I'm a total noob with Avisynth because I use MeGUI to encode but I don't write any script.

I'd like to achieve the same result this person got when restoring/improving the original series, the images on the left are the original, the ones on the right are after the encode, the lines are thinner and it has less noise and blockiness (I think, not sure)

https://i.imgur.com/fylqM5P.jpg

Now, these are some screenshots from the End of Evangelion movie:

https://i.postimg.cc/3rS3PYDF/22.jpg

Is it posible to write a script to get similar results?

Here's one last comparisson, first image is the original series, second image is the improved one, and the third image is from the movie I want to "improve"

https://i.postimg.cc/DyS7hHrk/33.jpg

EDIT: why the hell the images are gigantic in the post?!

[StainlessS]

If you use PostImage.org, you can select ThumbNail for Forums, for small graphic which when clicked on, pulls up the original image.
(dont need an account but better if you do).

[camjac251]

For the lines to become thinner and sharper, aWarpSharp could help with it. It also looks like the levels were adjusted from an SD source to HD and are brighter as well.

For film sources which this may have been scanned from, I'd recommend using TemporalDegrain2 found on this forum. It can easily clean up grain while keeping the detail.

Try using autolevels and TemporalDegrain2 first and see if it's able to remove the grain enough. If not, I'd recommend checking out waifu2x which is a cartoon/anime upscaler and can be used to upscale or denoise. Waifu2x Caffe offers more features and models than the avisynth plugin though so it won't be possible to do it all in Avisynth.

[lansing]

You shouldn't do anything beside degrain, do not thin the line.

As for the color, if you have an identical working and reference image, you should be able to get the curve used by the reference image with 3dlut creator: first color match to the reference and then extract the rgb curve. But since the images are not identical, this probably is not going to work.

[Yanak]

Also you have both identical frames, one before, one after color changes (In you last picture the 2 first pics are not exactly the same) it is also possible using Gimp and G'mic plugin "CLUT from After - Before Layers" to get a Clut file that will be the difference between the two pictures, then possible to directly use the clut file with ffmpeg or use "IWLTBAP LUT Generator" to convert the clut into a .cube file and use it with the 3dlut plugin for avisynth.

It needs good quality identical frames, imgur host compress them way too much and there is a big loss of quality. The host StainlessS recommended i think do not compress pics so it should be good one if you identical full size reference frames.

Still in Gimp it is possible to use the elsamuko's "get curves" (https://sites.google.com/site/elsamuko/gimp/get-curves) plugin to get a curve that will be the difference of the 2 images, then apply this curve into a reference haldclut image, then again use this haldclut file directly in ffmpeg or convert it to a .cube file to use in avisynth.

[LemMotlow]

The images on then right are over saturated and look like bad phony attempts at Disney cartoons.
Maybe you've been watching too many bad YouTube videos.