I use a lot of Virtualdub. It annoys me how some filters are 32bit and can't be used in my 64bit install that I need for some quality filters that are only in 64bit. Is there a way to turn the not working 32bit filters into 64bit?
Specifically, I want to use some antialiasing filters along with NeatVideo 64bit, since NeatVideo in 64bit is MUCH faster than in 32bit. I would also not like to do multiple encodes, since it'd possibly reduce quality and I was also advised against it by some video professionals.
+ Reply to Thread
Results 1 to 11 of 11
Besides the lossless intermediate route, another option might be to use vdub x86 frameserver to vdub x64 . But this will be slower than native x64 workflow
Which filters specifically ? There might be alternative filter options in x64 available, or other supplemental workflows like avisynth or vapoursynth that you can combine with vdub2, or use neat video virtualdub plugin in an avs script
I need debanding, anti-aliasing and removal of hairs&scratches. Would also be nice to have upscaling, since I was told that you can get better upscaling in virtualdub/avisynth than with Video2x or even Topaz's Video Enhance AI.
vdub filters specifically ?
You have more selection of filters in avisynth. Dozens of more choices for each category. You can still use neat in portions of the workflow if you wanted to. You can load virtualdub only plugins in avisynth with LoadVirtualDubPlugin (in case there was an order of operations you needed to perform in a certain sequence). You can run most x86 filters in a x64 host with mp_pipeline in avisynth
Would also be nice to have upscaling, since I was told that you can get better upscaling in virtualdub/avisynth than with Video2x or even Topaz's Video Enhance AI.
A general observation is NNEDI3_rpow2 reduces aliasing and jaggies, but the lines are "coarse" compared to modern methods. It doesn't do a good as a job with high frequency details (but on a bad source, there will be no high frequency details) - but it's stable and artifacts are rare or minimal compared to other methods . You can call it more "conservative"
But "AI" scaling (more appropriately called GAN scaling) requires an appropriate trained model. The problem with Video Enhance right now is there is only a handful of models. Use a generic model and you'll get generic results. Using a screwdriver in place of a hammer or saw won't produce ideal results in every situation. Another issue is artifacts - some big issues with temporal artifacts and with text/numbers . Also it doesn't give you the ability to use interpolation between models yet (e.g. 60% GAIA, 40% Artemis etc... such as other "AI" frameworks. )
For GAN upscalers, the end quality you get is largely based on preprocessing and appropriate model choice. A big issue with hundreds of research models is they use clean sources, and a clean downscale to train. But in reality, the sources most people want to upscale have other issues such as noise, compression artifacts, dirt etc... So most publicly distributed models are not necessarily appropriate for most scenarios
Note AVISynth also has access to a few "AI" upscalers now too including FSRCNN and VDSR; and vapoursynth has access to many more, hundreds of models and frameworks on github such as ESRGAN, because it's python based and most "AI" research projects on github and similar repositories use python
I know that I would have better results with Avisynth, everyone keeps praising Avisynth to high heavens. Problem is that it does nothing. I do not know how to make it function. Every line of script gives me an error and I can't even import a video with it. Worst video tool I have used, or actually only tried to use so far.
I wish there was a "for morons" guide or actually a decent GUI. Being able to simply select video would be much better.
I frequently use ESR GAN. The reason why I was interested in Avisynth upscaling was that I was told it was fast and as good if not better. If it indeed works better, I would be a fool to continue using subpar methods of video enhancement.
I'll check out that Github link, it looks interesting. Hopefully it works.
Edit: It's a .avsi. What am I supposed to do with this?
Video Enhance AI is not necessarily "subpar"; it can produce good results if used correctly. You often have to preprocess for ideal results. And there are cases where it does a better job than other tools as well, or at least on a few scenes. There is not one tool or workflow that is the "best" at everything, every situation
An .avsi is an auto importing script. Compare that to .avs (no "i") that does not auto import. When placed in the plugins directory , an .avsi will auto-import, instead of specifying Import("PATH\script.avs") in your script
Plugins such as .dll's are autoloaded when placed on the plugins directory too. Instead of LoadPlugin("PATH\yourplugin.dll")
Each script or plugin usually comes with instructions and documentation about prerequsites
You need to start with avisynth basics if you want to get more complicated scripts to work . A bit of a learning curve, but not too bad once you get started. It very versatile and has plugins/functions that can do many things. Very useful tool
An .avs script is just like a "virtual" video and audio. All the commands in the script are run in sequential order, and the output node, is what the receiving application "sees" . So if you open a valid avs script in vdub2 or ffmpeg or anything that accepts avs scripts, it "looks" like a video that had all those filters applied.
I don't even want to do anything complicated with Avisynth. I want to select a video and run a filter on it. That just doesn't seem to work. It realistically shouldn't be this hard to do. The learning curve is more like a learning wall.
Do you have a very simple step by step guide to achieve this?
Write a script in notepad, save it , change extension to .avs . (There are better utilities like avspmod that allow for previewing, and are a sort of minimalistic GUI, but start with notepad for now, because it's simple)
Open that .avs in vdub2 by drag and drop or file=>open video (not file=> run script). Also, it has to match the avisynth version in terms of x86 or x64 . e.g. Vdub2 x64 would initialize avisynth x64, but vdub2 x86 would initialize avisynth x86
A good general purpose source filter is LSmash, HolyWu fork . It can open almost any type of file.
A script that applies the filter "blur" at a strenght of 1 might look like this. You just have to change the path and names to your specific video
Hmm, it appears that I got some of this to work. I'll still have to do further tests but my file opens in Virtualdub via a .avs file now, which I think is the first and most important step. Thank you for recommending LSmash. Is Blur included in the default install? What others are?
Plain "blur" is an internal (bundled) filter. But there are different types of blur algorithms, different types of blurs. It was just an example to get you started
I don't know there is a comprehensive list of internal filters somewhere
Those lists are probably not comprehensive, especially the external list. Often there are some random developers with their own website somewhere making plugins