I am not sure, what might be the best way to deal with Video Enhance AI and interlaced video.
I’ve tried Dione, but I get much better results with Artemis LQ, so I want to use that one instead.
capture my VHS tape with deinterlacing on the fly?
capture my VHS tape interlaced and deinterlace afterwards with other software?
load interlaced video in Video Enhance with Artemis and deinterlace afterwards with other software?
I capture video with lossless UYVY codec and in Video Enhance I save it with Mov ProRes 422, at the end I want to convert to x265, preferable with Handbrake (or any other software you might recommend?).
You should capture interlaced video as doing this on the fly is throwing away information that will provide better results later on (and important for preservation). Capture in ProRes for best results.
You can then de-interlace using Handbreak which can give acceptable results with the bob filter. However, I have found superior results come from using Avisynth. I am on a Mac so use Hybrid (which is interface that allows the Avisynth filters to run). QTGMC is the best of these, which is a Vapoursynth filter that yields superior de-interlaced results (basically nice and sharp with better handling of combing artefacts which is the main problem with every other NLE or plug-in that de-interlaces. It does noise reduction too so can avoid the need for a plug-in like Neat video). It was a little tricky to install and took time learn Hybrid, but worth persevering as the results are great. However, the rendering can be slow.
The Dione model can de-interlace, but in my tests I found it better to do that step prior to importing to VEAI, so you have more control. As I am in Europe I de-interlace to 50fps from 25fps PAL, which makes each field into a full frame and retains the look of video when it’s played back. You can throwaway the second field for a film look, but it you plan any slow-motion effects it’s better to retain the 50fps.
Dione is great and fast if you want to avoid the steps I outlined as they are quite complicated to achieve. I have spent many hours of trial and error to figure all of this out and think it’s best to separate each process to have more control. You can upscale interlaced video using Artemis and de-interlace later, but it would be worth testing with Dione to compare the results.
I am now running VEAI on a MacBook Pro M1 and it’s much faster. I managed to get Hybrid to work on the M1 but it’s not yet supported but found for de-interlacing it does work for QTGMC filter and is much faster to export. in VEAI I have found that the Proteus AI model is great as you have more control to tweak things such as noise reduction. Artemis is good, but the AI may make things look too plastic which is why I prefer to use Proteus.
Thanks for the tips.
QTGMC seems to be the deinterlacer with the best results, while reading in other forums, this one was recommend the most and now it’s the one I use. I capture now interlaced, deinterlace with QTGMC and load the deinterlaced video in VEAI, works very well.
At the moment I try to find out how to reduce haloing/ghosting. With the Dehalo Avisnyth filter I see no real difference and with VEAI and Artemis Strong Dehalo it makes other parts more worse.
I think I should look also more into the Proteus model.
hello! I couldn’t have said it better than you, I also work on 25 fps and 50 fps deinterlaced, and I agree with everything you said, the only thing that makes me think is the fact of to scale before deinterlacing, does it really work? because by scaling an interlaced video with any model, it should make a really weird rendering, right?
Yes, de-interlace first to create a frame based clip of 50fps. Then upscale output with VEAI. The Dione model is designed to do both an upscale and de-interlace but I prefer to do this in separate steps as the result to my eye is better.
Okay, yes it seemed good to me :') On the other hand deinterlacing 100% (720 x 576 or in hd (1280 x 720) give me the same result on the video quality, but for now, it’s over dione for me as long as it is not up to the task against qtgmc…