View Full Version : Opinions for avchd x264 anime settings
gizzin
25th January 2010, 05:39
Looking for peoples opinions on x264 settings when encoding anime to avchd. 2pass is a must, time doesn't matter. Thanks for your opinion.
J_Darnley
25th January 2010, 13:09
--tune animation
aBra
25th January 2010, 13:15
...time doesn't matter...
--preset placebo ^^
MatLz
25th January 2010, 20:41
More b frames and refs (7 is good....more is useless IMNO)
If you have enough bitrate:--no-mbtree --b-pyramid normal --aq-mode 0 --psyrd 0:0
If not: same commands but with AQ enabled and --aq-strength 0.4to0.7
Well you need to test to get a good compromise ringing/banding...
julius666
26th January 2010, 09:03
If you have enough bitrate:--no-mbtree --b-pyramid normal --aq-mode 0 --psyrd 0:0
--no-mbtree ? Why? It's pretty usable at anime. :confused:
Anyway, I would recommend this settings (if time doesn't matter):
--preset veryslow --tune animation --aq-mode 4 --qcomp 0.72 --b-pyramid normal
--ref-frames: the more the merrier, but watch out for DXVA restrictions (it depends on your resolution).
keep in mind, that you need a patched build for --aq-mode 4 like this http://forum.doom9.org/showpost.php?p=1367766&postcount=2923
MatLz
26th January 2010, 11:11
--no-mbtree ? Why? It's pretty usable at anime. :confused:It's just my half blind's opinion, don't be:confused:!:D
I find some edges are broken and some details completely killed....theoricly it's maybe the right allocation of bits....but I don't like the result!
shon3i
26th January 2010, 12:46
His target is AVCHD, so this exotic setings such as bframes > 3 are not allowed :) Resoultion then must be 720x480, 720x576, 1280x720, 1920x1080, 1440x1080 and no exceptions, btw here you have allowed combinations http://www.avchd-info.org/format/index.html. Based on this other settings must be applyed such as max ref frames, VBV, level. To be more safe i recommend you to use Level 4.0, VBV-Maxrate 15000, VBV-BufferSize 15000, and maximum 4 reference frames for any resolution, and 3 bframes. Also you need nal-hrd patched x264 buld and you need to use --nal-hrd and --aud switches aslo.
Asmodian
26th January 2010, 22:39
It's just my half blind's opinion, don't be:confused:!:D
I find some edges are broken and some details completely killed....theoricly it's maybe the right allocation of bits....but I don't like the result!
Is this comparing the same sized files or are you using no-mbtree with the same crf value as a way of increasing the size? I haven't noticed this is why I am asking (but maybe I am encoding to to low of a crf value for this effect to be seen). :)
MatLz
27th January 2010, 03:56
I usualy use 2 passes encoding, even for testing.
Well I did a test: I compared my old reference encode (without mbtree and with regular settings on r1332) of my favorite dvd sample test with complete insane overkill placebo settings+mbtree of the last 1400...
At my eyes, the old regular encode is highly better.
But I repeat: it's just my opinion.
julius666
27th January 2010, 10:21
I usualy use 2 passes encoding, even for testing.
Well I did a test: I compared my old reference encode (without mbtree and with regular settings on r1332) of my favorite dvd sample test with complete insane overkill placebo settings+mbtree of the last 1400...
At my eyes, the old regular encode is highly better.
But I repeat: it's just my opinion.
Did you try it with higher qcomp too?
Asmodian
27th January 2010, 21:55
I usualy use 2 passes encoding, even for testing.
Well I did a test: I compared my old reference encode (without mbtree and with regular settings on r1332) of my favorite dvd sample test with complete insane overkill placebo settings+mbtree of the last 1400...
At my eyes, the old regular encode is highly better.
But I repeat: it's just my opinion.
What resolution and bit rate are you encoding to? I would have just used mb-tree for pretty much everything (after my own limited testing on one source lol) so I ask to know in which situations I should do my own tests. :)
Caroliano
27th January 2010, 22:13
I usually use qcomp around 0.7 or a bit higher, so if AQ or MB-tree decide to screw up my image, they can't do too much damage. The higher the bitrate/quality that I'm aiming for, the higher I set the qcomp. I'm aware that I'm trading compression efficiency for more safety and a more constant quality acros the video. Kinda like when you chose to do more or less denoising before feeding the video to x264. IMHO, the default value of 0.6 is too low as a default...
MatLz
28th January 2010, 01:53
I never try to change qcomp...
Will do tests by increase it a little.
My test source is really difficult to compress, lot of motion, fades, grain....so good for testing!
I output it at 768x416 at 800kbps (same than my reference old encode)
julius666
28th January 2010, 11:59
I never try to change qcomp...
Will do tests by increase it a little.
My test source is really difficult to compress, lot of motion, fades, grain....so good for testing!
I output it at 768x416 at 800kbps (same than my reference old encode)
As Dark Shikari said in the mbtree thread:
there's some possibility that the default qcomp 0.6 is not the psy-optimal value--it might be higher.
http://forum.doom9.org/showpost.php?p=1310980&postcount=1
I've experienced a bit, and found 0.72 the best for film. For anime it's even higher, the optimal value for my eyes is between 0.75 and 0.82.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.