Beiträge von iAvoe

    Hi Vouk! Thank you for replying.

    Use cases for me for small files are emailing review versions.

    I am an animator and motion designer, working on short ads etc. Also the internet is slow around here compared to Europe and the US.

    Is there any where you might point me to to learn further about these terms? ABR, CRF or CQ(P).

    Cheers!

    Emailing video is not really a good idea (I remember there are 25MB file size limit), but you can try shrink resolutions; use OneDrive/GoogleDrive to share larger videos; or build a public torrent transmission and email just the torrent file to the client

    HDR Support (x264 & x265):

    Tagging DCI-P3: --master-display G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(?,1)

    Tagging BT.709: --master-display G(15000,30000)B(7500,3000)R(32000,16500)WP(15635,16450)L(?,1)

    TaggingBT2020: --master-display G(8500,39850)B(6550,2300)R(35400,14600)WP(15635,16450)L(?,1)

    The L(?,1) indicates lux ratio (e.g., 1000:1). This value has no standard and needs to be checked and written manually on each HDR video.

    If tagging for the source vid is unknown, find 1 of the following format in source video metadata:

    DCI-P3: G(x0.265, y0.690), B(x0.150, y0.060), R(x0.680, y0.320), WP(x0.3127, y0.329)

    bt.709: G(x0.30, y0.60), B(x0.150, y0.060), R(x0.640, y0.330), WP(x0.3127,y0.329)

    bt.2020: G(x0.170, y0.797), B(x0.131, y0.046), R(x0.708, y0.292), WP(x0.3127,y0.329)

    Content lumiance x265: --max-cll <max content light level cd/m2, max frame-avg light level cd/m2> e.g., 1000,640

    Content lumiance x264: --cll <max content light level cd/m2, max frame-avg light level cd/m2> e.g., 1000,640

    CLL has no standard and needs to be checked and written manually on each HDR video

    Indicate HDR10 content in supplemential enhance info (SEI),  x264 --hdr10

    Optimize HDR10 content (increase video size) per block (optional, x264 ): --hdr10-opt

    Indicate color range and transfer properties: --colormatrix <as source> --transfer <as source>

    Color range can vary thanks for both compatibility and multple HDR implementations, (e.g., gbr bt709 fcc bt470bg smpte170m YCgCo bt2020nc bt2020c smpte2085 ictcp). Check the source video metadata for them

    x265 is just too complex to figure out for past myself, so after watched too much lectures and papers, testing myself and checking forums, here I'm giving out parameters years late. I'll try deleting the old posts regarding x265 as well. In this post I'll give out 6 combinations, and explain key features of each. Since x265 has hundreds of parameters to tackle with, I'll remove all the customizations and just give the fixed amounts. :*

    Note: --strong-intra-smoothing filter is enabled, because it's a parallel feature with 3-tap filter in intra frame search. It only applies in specific conditions according to the formula, and makes overall encoding faster with minor quality loss. Data smoothing is also not blurring which preserves patterned details.

    Note: I'm more of an enthusiast instead the expert, the borderline inbetween expert and enthusiast is probably being able to read and explain rate distorsion function where everything becomes university master/doctor level, so please don't exprect these parameters to be prefect

    General Purpose for video platforms

    Simple and quite effective, less capable at compressing some scenes, qpmax is still configured for perserving output quality, YUV4:2:0 8bit, 10bit

    Block Splitting - Transformation: --min-cu-size 16 --limit-tu 1 --tu-intra-depth 2 --tu-inter-depth 2

    ME - Compensation: --me umh --subme 5 --merange 48 --rskip 1 --weightb
    Referencing - Rate Control: --ref 3 --early-skip --max-merge 2 --no-open-gop --min-keyint 5 --fades --bframes 11 --b-adapt 2 --radl 2 --fast-intra

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Intraframe Search: --fast-intra

    Quantization: --crf 18 --crqpoffs -3 --cbqpoffs --qpmax 26

    Adaptive Quant: --aq-mode 3 --aq-motion --qg-size 16

    RDO - Mode Decision: --rd 3 --rdpenalty 1 --splitrd-skip --rdoq-level 1 --limit-modes --rect --tskip-fast
    Sample Adaptive Offset: --limit-sao --sao-non-deblock
    Output: --hash crc --allow-non-conformance

    Multiple CPU nodes: --pools +,-

    Standard for video platforms

    Takes more compute resources, no advantage against General Purpose in simpler scenes, YUV4:2:0, 4:2:2, 4:4:4 8bit, 10bit

    Block Splitting - Transformation: --min-cu-size 16 --tu-intra-depth 3 --tu-inter-depth 4 --limit-tu 1

    ME - Compensation: --me star --subme 5 --merange 48 --rskip 1 --weightbReferencing - Rate Control: --ref 3 --max-merge 2 --early-skip --no-open-gop --min-keyint 5 --keyint 390 --fades --bframes 13 --b-adapt 2 --radl 3

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Intraframe Search: --fast-intra

    Quantization: --crf 20.5 --crqpoffs -4 --cbqpoffs -2 --qpmax 26

    RDOQ: --rdoq-level 2

    Adaptive Quant: --aq-mode 3 --aq-strength 0.9 --aq-motion --qg-size 16

    Mode Decision: --rd 3 --rdpenalty 1 --splitrd-skip --rdoq-level 1 --limit-modes --rect

    Rate Distorsion Optimization: --psy-rd 1.6 --rdpenalty 2 --splitrd-skip

    Deblocking: --deblock 0:0Sample Adaptive Offset: --limit-sao --sao-non-deblock
    Output: --hash crc --allow-non-conformance

    Multiple CPU nodes: --pools +,-

    NAS/Server playback recovery: --idr-recovery-sei

    High Compression - Film

    Burst high bitrate scenes will be compressed. Not recommended for video platforms, recommend shifting chroma plane resolution up for this one: YUV4:2:2, 4:4:4 8bit, 10bit

    Block Splitting - Transformation: --min-cu-size 16 --tu-intra-depth 4 --tu-inter-depth 4

    ME - Compensation: --me star --subme 5 --merange 48 --rskip 1 --weightb

    Referencing - Rate Control: --ref 3 --max-merge 4 --early-skip --no-open-gop --min-keyint 2 --keyint 400 --fades --bframes 15 --b-adapt 2 --radl 3

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Intraframe Search: --constrained-intra --b-intra

    Quantization: --crf 21 --qpmin 10 --crqpoffs -3 --cbqpoffs -1 --hrd --vbv-bufsize 30000 --vbv-maxrate 50000

    RDOQ: --rdoq-level 2

    Adaptive Quant: --aq-mode 2 --aq-strength 0.9 --qg-size 8

    Mode Decision: --rd 5 --rdpenalty 1 --splitrd-skip --rskip 0 --limit-modes --rect --amp

    Rate Distorsion Optimization: --psy-rd 1.9 --rdpenalty 1 --rd-refine

    Deblocking: --deblock 0:0

    Sample Adaptive Offset: --sao-non-deblock --selective-sao 3Output: --hash crc --allow-non-conformance --nr-inter 5 --nr-intra 5

    Multiple CPU nodes: --pools +,-

    NAS/Server playback recovery: --idr-recovery-sei

    Render to Reuse (video editor --> video editor)

    Some filters costs too much computing resources, therefore render a finished video and re-import is cost saving. Not sure when will premiere support hevc, but voukoder is not just for premiere, so this should still be useful: YUV4:2:2 8bit, 10bit

    Block Splitting - Transformation: --ctu 32

    ME - Compensation - Intrasearch: --me star --subme 6 --merange 48 --analyze-src-pics --weightb --max-merge 4 --early-skip --b-intra

    Referencing - Rate Control: --ref 3 --no-open-gop --min-keyint 1 --keyint 350 --fades --bframes 7 --b-adapt 2

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Quantization: --crf 17 --qpmax 24

    Mode Decision: --rd 5 --limit-modes --limit-refs 1 --rskip 1

    Deblocking: --deblock 0:0

    Sample Adaptive Offset: --no-sao

    Output - Mastering: --hash crc --allow-non-conformance --tune grain

    Multiple CPU nodes: --pools +,-

    NAS/Server playback recovery: --idr-recovery-sei

    High Compression - Anime

    For subtitle/translation groups and uploading to video platform. This parameter considers both high and low budget anime, include opening, ending and main content. The secret is to max out all the possible block splittings possible, because high-freq info here are mostly hand drawings and should be perserved: YUV4:2:0 8bit, 10bit

    Block Splitting - Transformation: --ctu 64 --min-cu-size 8 --tu-intra-depth 4 --tu-inter-depth 4 --max-tu-size 4 --limit-tu 1

    ME - Compensation: --me star --merange 48 --analyze-src-pics --subme 3 --weightb --max-merge 4 --early-skip

    Referencing - Rate Control: --ref 3 --no-open-gop --min-keyint 5 --keyint 360 --fades --bframes 15 --b-adapt 2 --radl 2

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Intraframe Search: --b-intra

    Quantization: --crf 20 --crqpoffs -3 --cbqpoffs -1 --cu-lossless

    RDOQ: --psy-rdoq 2.3 --rdoq-level 2

    Adaptive Quant: --aq-mode 3 --aq-strength 0.7 --qg-size 8

    Mode Decision: --rd 5 --limit-modes --limit-refs 1 --rskip 1 --rect --amp

    Rate Distorsion Optimization: --psy-rd 1.5 --rd-refine --splitrd-skip

    Deblocking: --deblock 0:0

    Sample Adaptive Offset: --limit-sao --sao-non-deblock

    Output: --hash crc --allow-non-conformance --single-sei

    Multiple CPU nodes: --pools +,-

    NAS/Server playback recovery: --idr-recovery-sei

    High Quality - Anime BDRip Coldwar Duels

    For Blue-ray DVD Ripper social network's common agreement on: high quality encoding means "paused dark scene zoomed-in 4~5 times size" should look just as good as the source footage, and I bet the x265 devs will freakout after reading that line, but that's reality on why some BDRip providers/groups are decent but gets less downloads:S . Compare to HC-Anime, x265 needs to give up some compression features to realize this requirement: YUV4:2:0 8bit, 10bit

    Block Splitting - Transformation: --ctu 64 --min-cu-size 8 --tu-intra-depth 4 --tu-inter-depth 4 --max-tu-size 4 --limit-tu 1

    ME - Compensation: --me star --merange 52 --analyze-src-pics --subme 3 --weightb --max-merge 4 --early-skip

    Referencing - Rate Control: --ref 3 --no-open-gop --min-keyint 1 --keyint 360 --fades --bframes 11 --b-adapt 2 --radl 2 --pbratio 1.2

    Various Parameter Sets: --opt-qp-pps --opt-ref-list-length-pps

    Intraframe Search: --b-intra

    Quantization: --crf 18.5 --crqpoffs -5 --cbqpoffs -3 --cu-lossless

    RDOQ: --psy-rdoq 2.3 --rdoq-level 2

    Adaptive Quant: --aq-mode 3 --aq-strength 0.7 --qg-size 8 --aq-motion

    Mode Decision:  --rd 5 --limit-modes --limit-refs 1 --rskip 2 --rskip-edge-threshold 2 --rect --amp --no-cutree

    Rate Distorsion Optimization: --psy-rd 1.5 --rd-refine --splitrd-skip

    Deblocking: --deblock 0:0

    Sample Adaptive Offset: --no-sao

    Output: --hash md5 --allow-non-conformance --single-sei

    Multiple CPU nodes: --pools +,-

    NAS/Server playback recovery: --idr-recovery-sei

    Thats it. <3

    I think AV1 is still too early, shoukd wait for it become mature and get more format supports (I'm not sure current state, at least wait for support in yuv4:4:4 4:2:2 and 4:2:0, and a bunch of instruction sets)

    I use it for basic editing, CD ripping, extraction & converting. compare the quality of Audition with Pro Tools & other pro DAW here:

    https://src.infinitewave.ca/

    Sorry for previous probably offensive comment, just had a bad sleep

    I didn't realize you were not talking about exporting, but just saying after lots of attempts, export a lossless version and use QAAC -a 320 does give the best quality in lossy audio among all other options

    For importing, the best way for now is to convert your lossless source file to another lossless file that Audition supports, it depends on Adobe to consider adding any codec support so I recommend you to leave a feedback on Adobe forum and Audition subreddit:sleeping:

    So you're saying that MainConcept products are horrible? So the MainConcept MP3 encoder is horrible too!

    Guys, if Audition users donate to Voukoder, does he develops FFmpeg import plug-in?

    That aac encoder is horrible because it destroys your high pitch audio profile, and there are lots of distorsions in mid range, QAAC is way better

    I don't know who made this aac encoder but if that is the quality it offers then that MP3 encoder would be sub par

    Can you tell what's the reason Audition requires a output plugin support? I would just output .wav and then compress it

    On obs studio u have to choose a filter, even if i'm not downscale/upscale. Then maybe i selected it but not using it at all? I heard lanczos not much better

    that's upscaling/downscaling filter, make sure you record resolution is same as output resolution, it's in the settings-video tab, in your case both are supposed to be 1920x1080 (the filter process will be ignored then)

    You,can also change recording fps on there, that's why I asked you does Twitch allow you to record in 57fps lol

    I've studied video encoding as a hobbit for 4 years, and unfortunately I've eventually found you get what you paid

    the best way to get small file size, decent quality and speed all together, you need bare minimum a ryzen 3600x or i7 8700k running x264 with low latency ram, and that's the baseline acceptible you are looking for

    but that may offer a good quality, it's not going to be the case on file size, if you dare want extra 40% size reduction, get 10 cups of coffee or upgrade your CPU

    how about x265? well, you can get 3fps on a 3800x / 9900ks for a 65% size reduction, but if you use is's potential for the best file size and quality, x265 will beat down a 3970x into a 6fps snail

    Disney had a house sized expensive video rendering supercomputer, and it take weeks to sqeeze out a clip movie, it's very difficult to compress and render video, you can't judt cheat through

    ??

    There are millions of people who want to upload videos created with Premiere to Youtube but have problems either because their file is too big or because final quality on Youtube is not enough good.

    What do you not understand in this simple concept of wanting to upload a video of a reasonable size to have a satisfactory rendering on Youtube?

    To say that there is no solution is not an acceptable answer; So I offer mine to the community,

    because there is no such thing,

    if you want speed and quality, you get huge file size;

    if you want file size and quality, you get very low speed

    if you want high speed and file size, you get bad quality

    especially if you are using a budget computer

    uh... yeah... but... there's no such thing to get good quality and concerning file size with a push of button, unless you are going lossless (and buy an enterprise SSD to spped things up:whistling:)

    (or watch video on a smaller mobile phone screen XD)

    try to remove bicubic filter (not sure why you are having a filter) and make color depth to full instead of partial, that will improve your quality

    my avatar in a touhou project character in a IOSYS flash music video:whistling:

    Later when you upgraded encoding CPU to 2700 in 4Ghz or 3800x default with RAM under CL16 latency in 3200mhz, I guess you can boost subme to 7~10 range, and use other parameters like no-fast-pskip=1, after all these you can bump bitrate from 7777 to 7800, this should give a much better quality at hard scenes

    Youtube compresses your video, so the bestquality you can get is to upload original project output or give enough size of the file in a lossy compression, so the video would only be compressed once

    also qp27 is way too much quality drop, consider qp17

    for lossless, use cineform (not sure how to do that)

    for the smallest file size possible, use hevc yuv4:4:4, its in voukoder but for now you have to manually input a hundred settings in this post

    for the minimal lossy quality, use apple prores 422

    for speed/lossless, buy a RTX 2000 or higher series hard, use built in hardware encoder on the Turing/Amplere or higher chip

    for no visible lossy quality and convinience, use x264 yuv4:4:4, which is in voukoder

    for speed without concerning visible loss, buy a GTX 1000 card, use built in Pascal hardware encoder

    Yes and I love the fact that on non-movement scenes the bitrate is lower, without losing quality. But on hard scenes don't know if the quality beat nvenc.

    edit: I'll test 900p with your preset today, maybe I can go higher then veryfast.

    edit2: why not profile high? just asking

    profile is just a limiter of how much value can be used for most parameters, default not set

    CPU profile is actually called 'preset' in encoder include x264, it's just a set of parameter that can be override by manual input parameters, default is medium

    I went full custom, so they basically are not set, but you definitely can try if it benefits

    Oh yea, amd rocks and Intel money grab start end finally.

    BTW, actually I have no money to spend on a new cpu ? so is there a way to beat new nvenc chip with my 8 core 16 thread or do I have to wait to upgrade the gpu on my stream rig with a nvenc card? The good thing is that new 3000 series doesn't have a better nvenc quality but still the same, then I could put a 1650 super (soon cheaper) on my stream rig and be fine.

    Anyway I'm still mad, I can run 8k medium high profile cbr and probably really close to nvenc quality, but I was dreaming about a secret command that can beat nvenc, abuse vbr to win this war ? but u telling me that for beat that fu**ing nvidia need a 24+ core cpu and for me it's not worth anymore, nvidia won this time Feelsbadman

    For now, at least until I'll have money to spare, I'll try to go in 900p, where as I know, nvenc shine less or atlest 8k bitrate will look good and smooth, at least on smartphones ?

    Nope, Nvidia didn't win this, the quality roughly lies down here:

    Lossless > Apple Prores 422 > CRF > CRF+VBR > RTX Nvidia GPU > ABR > GTX Nvidia GPU > CBR > ...

    Right now you are using CRF+VBR right?