Tips, tricks, and troubleshooting


This is taking forever! Isn't there any way to speed things up?

     The easy answer is to choose a faster QTGMC preset. "Fast", "Very Fast", something like that. The more interesting answer is to try out a multithreaded version of Avisynth. Multithreaded operation currently requires a bit more user intervention than I feel comfortable recommending to most people, but if you like experimenting with that sort of thing, you can try Avisynth 2.6.0 MT.
     If you opt to go the MT route, don't forget the special package of modified QTGMC plugins, found in the Doom9 thread for that filter: http://forum.doom9.org/showthread.php?t=156028


Error message: "AVISource: couldn't locate a decompressor for fourcc dvsd"

     This error is an old standard, and is the result of not having a system-wide DV codec installed. Most NLE packages, large and small, have internal decoders for the DV25 format, and can play the footage without issue. Avisynth's AVISource, however, uses the Video for Windows architecture, and in order for it to decode DV, a VfW codec for the format must be present on the system.
     For DV, a very common recommendation is Cedocida, by Andreas Dittrich. I'm a user myself, and I'm quite happy with the results of both decoding and encoding. Donald Graft has compiled a Windows version, which can be found at http://rationalqm.us/hosted.html. At the bottom you'll be able to download version 0.2.2, which is the most recent.
     Extract the files to a convenient location, right click the .inf, and choose Install. That's it! Just don't forget that when using Cedocida to compress widescreen DV, you must go to the codec configuration dialog and set "Aspect Ratio" to 16:9.


Green, rainbow-blocked, or otherwise glitchy frames in VirtualDub

     Frankly, my education in the field of video compression is insufficient to properly identify the cause of this one, but one SimpleSlug user has reported to me that setting VirtualDub to Video->Full processing mode and turning on "Preserve empty frames" solves this. I can't say why that works, but if you find yourself facing the problem, I'd suggest trying it.


The video looks okay, but how can I give it that last little push?

     It's never a good idea to offer one-size-fits-all solutions, and I hate to suggest there's only one answer to this, but I find GrainFactory3, by Didée (the author of TempGaussMC, my previous default deinterlacer and the basis for QTGMC; I've lost track of all the great scripts he's produced), and AddGrainC, on which it depends (download, discussion), really help sell the upscale. A little noise in the right places works wonders, and GrainFactory, even at its default settings, makes everything seem sharper. It's psychological, to be sure, but it's exceptionally effective.
     Please use this with caution! Beside the reduction in render speed, the addition of grain dramatically increases the bitrate necessary to compress your footage. You'd need something like x264's --tune grain to really do it justice, and even then you'll be turning the bitrate up rather high. And if your goal is Youtube, no matter how high a bitrate you use, any grain you add likely won't survive their recompression, so I'd say forget it in that case.


Isn't upscaling a waste of time for Youtube uploads?

     Yes, probably. These days Youtube produces fair-to-middling SD encodes if you give them a good source file to work with. The upscaling I walk through in this tutorial is aimed at video producers that need to scale SD clips for inclusion in HD projects (inside their NLEs), and isn't intended to suggest your videos will look any better on Youtube than a well-deinterlaced SD upload.
     It's still worth considering, however, that their HD encodes get higher bitrates; by upscaling, you have a higher pixel count without adding any actual detail. SD detail, HD bitrate (relatively speaking), I wouldn't discount the potential value of an upscaled file without testing it first. I have, personally, and I think an upscaled upload looks more like the SD original than the Youtube-recompressed SD version does. But you'd really need to do your own tests to judge the value of this procedure for yourself.