AMV/HowTo

I'm not gonna presume to tell you how to make your music video. I don't have a clue, I've only ever made one myself. I'll just describe what I did.

Conclusions wot I had come to beforehand

..from the net, from people's comments on AMVs they watched, and from my own observations:

Don't be afraid to cut the audio short. Fade it out, or cut the middle out. People typically get bored after three minutes or so at most, however good your video is.
As well as all the usual gumpf about quality source, no subtitles and so on: don't have people's mouths moving unless you're doing lipsync. It's invariably distracting.
Try and plan to make everything fit together. A number of videos look like someone had an idea for the chorus and then just chucked their anime collection at the rest of the audio and hoped something fit. A number of others don't even bother - the music is good, so are the video clips, and even the cutting and effects may be technically very savvy, but if they don't go together the audience gets bored.

These are reiterated over and over again in the [guides] and are the things newbies get flamed for the most. I figured we'd make an effort to avoid those and at least get flamed for something not immediately obvious instead :)

Overview

1. Grabbed a legal copy of the soundtrack.
2. Listened through it, wrote out the lyrics.
3. Made a powerpoint presentation; started with a slide per line of lyrics. Wrote ideas on each slide,
flicked through them manually with the music playing to get an idea of how it would look. Changed things. By the time we were happy, we had a slide per cut describing rougly what we wanted and a list of places we could remember off the top of our heads to look for it.

At this point we split up. SunKitten started looking for the clips, and I started marking up the audio in Cinelerra. Loading in the audio track, I added a position label in each place we wanted a cut - typically just before a line started. Being able to scroll around and see the waveform made this relatively straightforward, though tedious. Hint: set Cinelerra to work in frames rather than samples or time. Hint: having marked your position, select a second or so just before it and play to make sure it's not too late in the sound; then try selecting an extra frame past the mark and playing to make sure that the mark is as late as it can be. [todo: screenshots]

SunKitten handed the clips over to me. We opened them in Cinelerra without adding them into the playlist. The procedure then went something like this:

Select the range of video track between the start and end labels of the chunk of audio the clip is meant to go alongside of. Look at the selection readouts to see how long it is in frames. Set in and out points.
Use the previewer to find a suitable chunk of clip of the required length. Mark it with in and out points. Double-check that in and out points on the timeline are in the correct place and the entire timeline section to be overwritten is also selected. Hit the overwrite button in the previewer. Check that you got the framecount right by looking at surrounding clips already on the timeline = if the clip you're pasting is longer or shorter than the section of timeline you selected to replace, everything will get shifted along.
Rinse, lather, repeat for the other clips. Change existing clips if you find better ones. Keep checking how the whole thing looks.

That's basically it.

Techie notes

We used Cinelerra to cut about a minute from the audio track (the main couplets get repeated, very slowly; the entire track would have lasted way too long to retain an audience's attention) before marking up the audio.

Cinelerra is good at the mechanics of editing, but currently not very good at converting between video formats. The least troublesome way to use it, IMO, is to use image sequences for video and raw PCM for audio. This also gives you the added advantages of being able to trivially lipsync and otherwise process the video by editing the index files with a text editor (which IMO is a much quicker way of doing frame-by-frame editing than trying to drag frames around in a GUI).

SunKitten used VirtualDub? to convert the clips she ripped from DVD / captured from video to TGA image sequences. I then used a trivial Perl script (read a directory listing, tack on a fixed piece of text and spew the lot to a file) to produce the index files, which have the following format:

 TGALIST
 # Frame rate:

floating point number; eg. 23.976

 # Width:

in pixels; eg. 720

 # Height:

in pixels; eg. 480

 # List of image files

one filename with full path per line

e.g.

TGALIST
# Frame rate:
23.976
# Width:
720
# Height:
480
# List of image files
/home/mag32/AMV clips/CB 1/00000.tga
/home/mag32/AMV clips/CB 1/00001.tga

.
.

I started out getting SunKitten to note down the framerate of each image sequence as it was captured, and faithfully passed it on to Cinelerra. Predictably, when the framerate differed from the project's, Cinelerra would generate tweens for the clips. Normally this is a good thing; however, I decided I didn't actually care that much about a small difference in playback rate and would prefer to have cleaner video, and altered all the index files so they contained the project framerate instead.

A very few clips came from fansubs, and the frames were much smaller than the project size. Rather than learning my way around the Cinelerra resizing tools, I decided it would be quicker to use the ImageMagic? command line tools to resize the image lists directly, since they were handily sitting there on my hard drive in nice ImageMagic?able formats. This worked, to a point: by the time I hit on the idea, I had already indexed the clips and placed them on the timeline. Now, Cinelerra performs no editing as such; it works with pointers to the source. If the source changes, so does the edited video. In theory. In practice, Cinelerra caches the size and framerate in the project file. So resizing the images behind its back had no effect - Cinelerra simply silently resized them on load. Looking at the project files, I discovered they were in fact stored in XML and were eminently editable. I proceeded to adjust the sizes Cinelerra had recorded, and abandoned the ImageMagic? tools since Cinelerra was actually quite happy to do the resizing for me. I imagine someone more at home with GUI""s would have done this the proper way documented in the Cinelerra manual.

The final rendering was also done to an image list, and VirtualDub? was used to compress the result to Xvid. Cinelerra was used to render the edited audio track straight to MP3, and VirtualDub? was used to attach the audio to the video. The uncompressed clips took 17.5Gb in total at the peak of rendering, which isn't much by modern standards; we ended up using about a tenth of what we captured.

All mentions of VirtualDub? on this page actually refer to the version modified to handle MPEG4 streams rather than the original (that's VirtualDubMod?).

Doing the capturing/editing (SunKitten):

Clips were captured from DVD, from video and from fansub .avi files. I downloaded VirtualDub?, then VirtualDubMod?, then every codec I could lay my hands on (actually, DivX 4 and XviD? 4 did well enough, and both are free). Then I went through the various files. VirtualDub? is fairly easy to use - load file, find clip, set start and end points and save. I used an image stream, as MoonShadow said.

For fansubs, the main thing is to make sure subs aren't in the way. If they are and the background is static, then it's not hard to replace the sub with a section from the unsubbed bit. In 'It's Raining Bishounen,' the clip at the end where Yuki from Gravitation is saying 'Hallelujah' was subbed. I edited each frame in PaintShopPro. If you want lipsynch from something that's only available in fansub, you're going to have to do something like that, since if lips are moving, there will be subs.

It is far, far better to use DVDs if you can. I have my laptop DVD drive set to region 1. I flicked through the DVD on about 8x forward speed and selected the chapters I wanted. I used [SmartRipper] to rip the chapters to my hard drive and loaded the .vob files straight into VirtualDub?. If you were going to actually rip the DVD properly, you would need to do additional steps to render the .vob files playable with sound/subs and so on - since I only needed video, I just used the files as they were. Ignore the rest of what it says on the SmartRipper? site :)

For video - we only had two videos, both of which were good quality, one of which is not and has never been available on DVD (KokoWaGreenwood? update - this situation has now happily changed). We actually went and bought a piece of hardware (and associated software) called PCTVDeluxe, which plugs into a USB slot and captures any kind of video to your hard drive as mpeg2, sized at 720x480. Then I opened these in VirtualDub? and carried on as normal :) There are nasty little flickery lines at the bottom of the two video clips - KokoWaGreenwood? and MahouTsukaiTai? - if you look, but they're of minimal importance.

The only other thing I did was edit Hitomi's pager. The actual time is 5:17 - it's been edited to read '10:27' to synch with 'just about half-past ten'. I was very pleased with that - done as with Yuki in PaintShopPro, but any image editing program, preferably with layer support, will do.

We started with a number of ideas for where to put clips. As I went through capturing them, I found other ones good for places in the song (like the bath scene at the start) or discarded ideas (we were going to have KareKano in it too). It was quite tedious finding some clips - getting Yurika in an appropriate position and speaking slowly enough for the 'Amen' was a nightmare. Others just came straight off (but then, it helps that I know Slayers well :)

As for the idea, it just happened, after the song'd been played twice at a friend's wedding. We kicked it around for quite a long time, and I thought of using ExcelSaga's Great Will of the Macrocosm as the Mother Nature, along with some of the sequences - the Tutu one, and the Utena one of men falling were there from the beginning. We only got round to making it because I pestered MoonShadow enough that he actually went and found the song. I don't know if we'll make another (although I want to) - it has to be something that clicks right because too many AMVs end up with random footage and one or two bits that work, and I don't want to be responsible for another one :)

This seems a sensible page to ask and rant. Damn. Are there any video editing apps for win32 other than virtualdub? Virtualdub can do recompression and filters wel enough, but it dies horribly trying to do pretty much anything else. Any suggestions? --Vitenka

Not anything free AFAIK, but there's stuff around for Linux. If Cinelerra ain't good enough, I've heard good things about [LiVES] too ^^; Tsunami swears by Premiere, which is ****ing expensive. - MoonShadow

No-one has got cinerella to cross compile? I couldn't find it, but thought it might be under a different name. Its system specs scare me, since my linux box is a P2 and runs x at about 3fps. --Vitenka

: My Linux box was a 1Ghz K6-II with 512Mb and a large amount of hard disk space. I have no hardware video acceleration. This was slightly sluggish; adding 512Mb of RAM (~30 quid), making sure DMA was enabled for the hard drives storing source clips and raising the limit on shared memory made a huge difference - real-time previewing drops one frame in three or so, which is perfectly usable. Processor speed doesn't matter as much as how much RAM you have and how fast you can pull data from disk - yes, your renders will be much slower, but you don't care so long as you can edit and preview. - MoonShadow

Um - this should be able to run X, a 486 used to be able to do it acceptably. But it doesn't, so I'm not throwing a video editor at it. --Vitenka

AMV/HowTo

Conclusions wot I had come to beforehand

Overview

Techie notes

Recommended reading