So I have my initial encodes for Tiger & Bunny the Movie: The Rising done. I haven’t started doing any subtitles for it yet, and I don’t have a timetable, though I know it’ll be quite some time before it’s done. I’d joke about saying that I hope to get it done this year, but I probably shouldn’t joke about that considering how much time (very little) I’ve been putting into Samurai Champloo recently. This particular preview is of the 720p encode, since that seems to be the most popular. I’ll get more into the video details below. If you don’t care about any of that stuff, the download link is at the bottom.
Anyway, I’m also experimenting with a few different audio encoding options. I’ve mentioned using Opus in the past, as I find that it just has a bit better quality-to-bitrate ratio than AAC. As such, I have included both AAC-LC (Nero Q0.5) and Opus (VBR target bitrate 360Kbps) encodes of both the English and Japanese 5.1 audios. (Blu-ray PGS subtitles included.) The AAC-LC encodes averaged around 460Kbps, while the Opus encodes ended up around 311Kbps (surround/LFE channels make the target bitrate inaccurate). I doubt most people will be able to hear the difference, but I would like to hear feedback on any issues or preferences between the two codecs.
In terms of frequency response, since that’s easy enough to measure, the Opus tends to win on the high end. AAC-LC at Q0.5 has a significant rolloff starting around 18KHz. Opus at this bitrate (and even lower really) doesn’t rolloff until 20KHz, which is right at the top of the human hearing range (hence why it was designed that way…). Below these rolloffs though, Opus does have a tendency to compress the dynamic range somewhat. This happens especially on the lower end of the human hearing range (20Hz to a bit above that), but it still gets quiet enough so that it gets lost behind all of the other audio. That said, I still have to give it to the AAC-LC on the low end. It really comes down to low bitrate tests I did in the past, where I was really comparing Opus to HE-AAC, so this isn’t exactly applicable, but I liked the Opus tracks more (and most studies I’ve seen tend to agree with this). At the bitrates here, it’d take listening very closely to small samples repeatedly to really tell the difference, so the decision mostly rests on the low bitrate tests and the fact that the Opus encoding gives me more bitrate flexibility. AAC-LC at lower bitrates than Q0.5 starts being noticeably bad on the high end, as that rolloff just starts earlier and earlier. HE-AAC manages to make a decent looking frequency response graph, but it has a noticeable thinness to the sound at times if listening through headphones in my experience, so I defintely wouldn’t want to use it as my 720p/1080p audio option (and I haven’t been). As the bitrate goes down, Opus just seems to compress the dynamic range a bit more and more, which is harder to notice than an early rolloff on the high end (again, in my experience).
As far as compatibility goes, most players seem to support it, but Opus in Matroska container support still seems to be spotty. MPC-HC (well, the internal LAV filters) has supported it for a while, and I haven’t had any issues with it. VLC apparently just got support for Opus in MKV support. I actually had to manually update to 2.2.0 to have it supported, and I’ve noticed an issue where it takes a while for VLC to start decoding the audio correctly after jumping in the video. Hopefully that gets fixed. If it doesn’t, then I’ll probably just stick with AAC tracks, but they likely have a while to get that done. As far as other players go, I really don’t know to be honest.
Now, as far as the video goes, I kept a bit of a lighter touch than previously. I haven’t noticed any weird noisy frames like in the first movie, but I haven’t watched through it all yet on the Blu-ray. I hope someone noticed that and made sure it didn’t happen with this movie though. Anyway, I started dealing with some line aliasing by essentially applying NNEDI3 four different times and merging the results. Two of them were on the video in its normal orientation, and two of them were on the video rotated. For each orientation, one pass kept the even lines and interpolated the odd lines, and the other pass did the opposite. Then I merged the results, weighting them all evenly. As such, each pixel was 50% the original pixel, 25% a horizontal interpolation, and 25% a vertical interpolation. Then I did a very slight sharpening of this and then applying a denoised edge mask to merge it with the original video. At this point, the flat areas were the original video and the edges were the smoothed and slightly sharpened version. Then I debanded with GradFun3, though not much was needed, and from what I can tell, it actually affected detected a surprisingly small amount of gradients. Anyway, at this point, I just output the 16-bit depth result to x264 for the 1080p version. I used the Dither package-based resizing method, as with the current Samurai Champloo rips, to resize in 16-bit depth and then output that to x264. Also, I’m pretty sure that there will be no color issues this time. A) I explicitly set the color matrices in the scripts. B) The Dither package functions change the defaults based on resolution anyway.
Talking about x264, I’ve also considered testing out x265, or rather, I’ve tested in some already, and I still don’t think it’s much better than 10-bit h.264 at similar encoding speeds to what I currently use with x264. It seems to be potentially better if I’m willing to let it go at really slow speeds (already taking 2-3 days for the 1080p version at my settings with x264, not including the Lagarith filtering passes). I recognize that there’s still a lot of room for improvement though, so I’m still open for the final release being in h.265.
As far as the x264 settings go, I think they’re the same as the first movie. I didn’t look particularly hard, but I didn’t modify them much off the very slow preset in the first place. Also, I’m planning the same CRF 16, 15, and 14 values for the 1080p, 720p, and 480p respectively.
Well, I guess I’ll throw in a bitrate table. I included all of the tracks for this preview release. Just as some additional notes, the 720p video is about 46.8% the bitrate of the 1080p video. Then the 480p video is about 57.3% of the 720p bitrate. It looks like the resizing method isn’t causing any weird bitrate drops here. I’m not sure if it’s the material or the different encoding values. That 480p value is actually higher than I’d like compared to the 720p, but it’s right around 26.2% of the 1080p bitrate, so it’s close to the 25% or so that I’d find ideal. I like to target whatever quality/bitrate balance I like for the highest quality version and then reduce the bitrate by about half each quality step down, and this is approximately the case with the above percentages. If I do end up using Opus audio, it probably won’t be quite like this. With the greater encoding flexibility, I’d probably actually different versions for each resolution. I used the 360Kbps bitrate target for 720p. I currently have encodes with target bitrates of 480Kbps and 240Kbps for the 1080p and 480p versions respectively. This may change, but this targets a total 50% reduction with the 720p version just being right in the middle.
Version |
Filesize (MiB) |
Bitrate (Kb/s) |
720p 10-bit h.264 (CRF15) |
2607 |
3628 |
5.1 E AAC-LC (Q0.5) |
332.3 |
460.9 |
5.1 J AAC-LC (Q0.5) |
330.9 |
458.9 |
5.1 E Opus (360) |
219.9 |
306.1 |
5.1 J Opus (360) |
226.2 |
314.9 |
Total |
3722 |
5180 |
Anyway, here’s the link. I don’t blame you if you skipped all of the stuff above…
Download Folder