100 mb for a 23 minute episode at CRF 17!?

[mahsah]

Strange results from an anime encode I just finished... I encoded it at CRF 17 in megui with the AE-Maxquality profile, and it only turned out to be 100 megabytes for a 23 minute episode!

I did process it with mvdegrain2 in avisynth, but I didn't expect the compressability gains to be THIS much. It looks OK, except for some banding (which I understand is more of a limitation of yv12 then x264, yes?), but 100 megabytes seems very small for something that people usually use as much as 250 megabytes for...

Would using the maxquality profile really cause that much of a compressability gain?

[cogman]

Anime (and cartoons in general) are VERY compressible. The thing is, most illegitimate encoders that distribute copies not authorized by the maker use a 2 pass encode rather then a constant quality. that means that even though the cartoon requires only 100 mb to look good, they overshoot it (or undershoot it sometimes) because they want a consistent file size.
So if you have a high quality cartoon with low noise it will compress very well compared to live footage, and that is basically because of large chunks that are all the same color for a long period of time (IE seconds)

if you really want to get it smaller, make sure you b-frames are at 16 as well as your reference frames. BTW what is the resolution that you are using?

[*.mp4 guy]

It can also take a lot of bitrate to have no visible banding, and no visible ringing at the same time, since what you do to remove one, will generally cause the other (as far as encoding is concerned, processing is a different matter). Yv12 does inherently have some banding issues, but they aren't insurmountable, its perfectly possible to have lossily encoded yv12 video with no visible banding, on any type of video, its just a matter of proper conversion to yv12, then its up to the encoder.

[Sagekilla]

Indeed, using fairly insane settings I managed to get an anime @ 720p using less than 1 mbps average bitrate.

Try cranking up B-frames to 16, using more refs (try 6 for example, but even more helps a lot) and using trellis 1 (2, if you feel like it). Trelllis 2 may be worth it depending on the source, since it usually decimates small details which may not be present in anime to begin with.

[mahsah]

I think the AE-Maxquality profile uses trellis 2 and 16 b frames.

I think most of the banding/blocking was in the source, and was just covered up by grain.

Here is a screenshot:
Original DVD:
http://img404.imageshack.us/img404/5301/dvdny2.png
Encoded:
http://img125.imageshack.us/img125/5964/encodedni5.png

Are there any filters to clear up the blocking/banding in dark areas?

[Dark Shikari]

gradfun2db can clean up the banding/blocking in dark areas, but it only works on playback; very little of the dither is kept when encoding.

You can also try Haali's old AQ; it might be useful in cleaning up the flat areas on your anime source.

[mahsah]

But if the blocking is already present, how would using AQ clean them up...?

Also does Haali's AQ work better on anime then the new one?

[Dark Shikari]

But if the blocking is already present, how would using AQ clean them up...?If its already there, your only real choice is to do something like gradfun2db. HQDN3D also seems to turn blocking into "smooth" lines of banding.
Also does Haali's AQ work better on anime then the new one?This isn't really known for sure at the moment--at low bitrates and resolutions the newer AQ doesn't seem as good. Haali's I recommended because it has an extremely strong effect in extremely flat areas, so it might work.

[mahsah]

Do you think using the blockbuster filter could help?