How to achieve a similar result without AI?

[Filipon]

I saw someone on Twitter touting the merits of "AVC Labs Video Enhancer AI" and posting two comparison images to show the difference.

I have read different threads here indicating that AI software cannot surpass Avisynth and in this case I would like to know what filters you recommend to obtain a result which would be as close as possible to what the person obtained.

- https://twitter.com/migzou/status/1764228777612320775

Image links for those who don't have Twitter:
- https://i.imgur.com/hL2NE83.png
- https://i.imgur.com/lfSFm50.jpeg

Not being the owner of the video used, I do not have an extract to provide but given the quality, the person probably worked with VHS.

This thread is intended solely to gather as much information as possible on filters to use to improve poor quality anime, such as VHS.
I'm not necessarily looking to upscale HD, just to resize them to a decent resolution (720x540) and obtain a significant quality gain.
As in the image on Twitter but without the defects on the edges.

Note that I am familiar with Avisynth tools but know nothing about Vapoursynth.

Thanks.

[Selur]

a. That doesn't look good at all, image still has tons of blocking.
b. Note that Avisynth with https://github.com/Asd-g/avs-mlrt also supports some machine learning stuff. And there are tons of models over at https://openmodeldb.info for handling animes&cartoons and avs-mlrt can use quite a few of them. (there are also tons of images)

If you want to upscale an image, use Adobe Photoshop or similar.
-> if you want Avisynth suggestions for a video, share a sample of your source.

Cu Selur

[Filipon]

It is true that the image has a lot of block and the contours are also quite strange but I find that there is an improvement in terms of sharpness, particularly at eye level.
Likewise regarding the leaves in the background, I found it not bad.
In fact what I'm trying to say is that despite the glaring defects in the image, we really have the impression of having low-end DVD in front of us while the basic image clearly reminds us of from VHS.

I didn't know about the Asd-G project. If I understand correctly, it works under AviSynth+? If yes, how to use it?
Because I know how to use Avisynth under Megui but I have absolutely no knowledge of AviSynth+, nor how it works.

I'm a little afraid of being lost quickly. Does Asd-G work standalone or do you need to install things in addition to the requirements to make it work?

Thanks.

[lollo]

I have read different threads here indicating that AI software cannot surpass Avisynth and in this case I would like to know what filters you recommend to obtain a result which would be as close as possible to what the person obtained.

I didn't know about the Asd-G project. If I understand correctly, it works under AviSynth+? If yes, how to use it?
Because I know how to use Avisynth under Megui but I have absolutely no knowledge of AviSynth+, nor how it works.

You asked for a AviSynth processing, and then you use a GUI without knowing it? You need to learn AviSynth to use it at its best

[Filipon]

Yes but it seems to me that AviSynth and AviSynth+ are not the same thing, am I wrong? Hence my question.
Likewise, it seems to me that VapourSynth doesn't work the same way as AviSynth.

The question may seem stupid to you but not everyone is necessarily comfortable making scripts and I prefer to ask rather than find myself in a dead end.

What I'm trying to explain to you is that I know how AviSynth works, I know how to load a plugin, etc. but I haven't experimented with AviSynth+.

[Alwyn]

The operation of AVISynth+ is the same as AVISynth; you'll probably find + uses different-dated or different-versioned filters.

The use is the same; create a script and open it in AVSPMod or VDub. If you know how to use AVISynth, you know how to use +.

https://github.com/AviSynth/AviSynthPlus

[_Al_]

And Vapoursynth does the same as Avisynth,Avisynth+, except code is in python. For example vapoursynth script can just load avisynth script. Basic differences:
Avisynth script:

BlankClip() #8bit rgb black clip 640x480, perhaps 10 second long, or so

vapoursyth script

import vapoursynth as vs
clip = vs.core.std.BlankClip()
clip.set_output(0)

avisynth users would say, it is complicated, but in reality, you can do much more just because of it, download portable vapoursynth and just particular vapoursynth version is loaded, or set_output() could be used to set just more outputs , not one only, set_output(1), or set_output(2) ...
also it is python , so you are set up using latest AI python things. Those latest AI setups could be ported to vaporsynth distributing dll. If you do anime, you do not waste much time and choose vapoursynth right from the beginning. It is anime scene what drives vapoursynth forward and distribute solutions. They jump on it right away.
loading video in avisynth:

v=LSMASHVideoSource("video.mp4")
a=LSMASHAudioSource("video.mp4")
AudioDub(v, a)
Spline36Resize(640,360)

Encoding avisynth script in ffmpeg is easy, ffmpeg can load video and audio, simplified:

ffmpeg -i script.avs -vcodec VIDEO_CODEC -acodec AAUDIO_CODEC output.mp4

vapoursynth:

import vapoursynth as vs
v=vs.core.lsmas.LibavSMASHSource("video.mp4")
a=vs.core.bas.Source("video.mp4")
v=v.resize.Spline36(width=640,height=360)
v.set_output(0)
a.set_output(1)

Encoding vapoursynth script, although having proper ffmpeg build you can load vapoursynth script, but I'd recommend using vspipe (comes with vapoursynth).

VSPipe --outputindex 0 --container y4m script.py - | x264 --demuxer y4m -o output.264 -
VSPipe --outputindex 1 --container wav script.py - | neroAacEnc -ignorelength -lc -cbr 96000 -if - -of output.m4a
ffmpeg -i output.264 -i output.m4a -c copy output.mp4

Also example of loading avisynth script (just video, not audio) to vapoursynth script:

import vapoursynth as vs
from vapoursynth import core
v=core.avisource.AVISource("script.avs")
v.set_output()

[User_Name]

I'm a rank amateur at all of this, so take this post with a grain of salt.

If it helps any, I had posted this a few months ago:

If you're tempted by Topaz VEAI, look at free Hybrid and thank Selur


A clean version of my thoughts in that link
- download Selur's Hybrid software, install it
- drag your video over/select the input file
- go to the tab labeled "filtering" and choose "vaporsynth"
- choose "sharpen" under there
- click the box for "CAS"
- choose the "line" tab (right next to the earlier sharpen)
- click the box for "NNEDI3AA"

Once you've clicked those two boxes look down at the bottom right of the program and you'll see a "compare view" box above what looks like a filmstrip. Check that box, then press the filmstrip. It will show you the original and the potential conversion side by side.

See if that's close to what you want. I'm betting it is. If it's not, there are lots of numbers and adjustments to tinker with, although things will get sideways quickly so approach it deliberately and only change one setting at a time. I would recommend increasing CAS from 0.5 to 0.9 just to see what happens and then adjust down from there before you change anything else, and to see if that gets you close to where you want to be.

I'm far from an expert but working with things is the best way to learn. One change at a time is crucial for me so that I know which change created the effect I was after.

Good luck.