ffmpegでvideotoolbox(macOS)を使う

macOSのhardware encoder

品質指定方法

-crf や -cq は使えない

-q:v (0〜100) と -b:v (ビットレート指定) がサポートされている

-q:v について

ffmpeg 4.4+ かつ Apple Sillicon 限定

0〜100 の整数のみ (小数点以下切り捨て)

h264 における crf と比べて、1加えたときの差がかなり大きい

ソース:

速度と画質

1080p30の実写動画15秒で雑に試したところ、下記がほぼ同じ画質っぽかった (vmafで比較, 速度はM2 Macbook Proで確認)

h264 の-preset veryfast -b:v 6000k -g 300 (2.2x)

h264_toolbox の -c:v h264_videotoolbox -b:v 7000k -g 300 (6.0x)

-q:v 換算だと、54と55の間くらい

hevc の -c:v hevc -preset veryfast -b:v 4300k -g 300 (0.48x)

hevc_toolbox の -c:v h264_videotoolbox -b:v 6100k -g 300 (5.4x)

詳細

0〜100指定の場合

code:txt

-c:v h264_videotoolbox -q:v 50 -g 300

これでlibx264 -crf 20と同じくらいのファイルサイズ

ビットレート指定の場合

code:txt

-c:v h264_videotoolbox -b:v 9000k -g 300

品質指定 (-crfや-cqの代わり) は -q:v

0〜100指定

-b:vと併用はできない。エラーとかは出ないけど -q:v が優先される

-q:v 指定の際は、実際のビットレートがどうなっているか mediainfo とかで確認したほうがよい

-b:v (ビットレート) は、-q:vを指定してない場合はちゃんと指定したほうが画質よさげ

ライブエンコーダの設定、ビットレート、解像度を選択する - YouTube ヘルプを参考に

-gはキーフレーム間隔の設定 (これ自体は videotoolbox に関係ない)

減らすとキーフレーム分ビットレートが上がるので、ビットレート指定の場合は増やすと画質が向上する

-q:v 指定の場合、増やすとほぼ画質は変わらずビットレートが下がる

出力後のキーフレーム間隔を短くする必要がなければ、ある程度上げとくと良さげ

level, profile について

軽く試した限り、上記に加えて -level 52 -profile:v high を指定したときと指定しないときの違いは無さそうだった

エンコーダの固有オプションリスト

ffmpeg 7.1 (homebrew) の ffmpeg -h full より

code:avoptions.txt

h264_videotoolbox AVOptions:

-profile <int> E..V....... Profile (from -99 to INT_MAX) (default -99)

baseline 66 E..V....... Baseline Profile

constrained_baseline 578 E..V....... Constrained Baseline Profile

main 77 E..V....... Main Profile

high 100 E..V....... High Profile

constrained_high 612 E..V....... Constrained High Profile

extended 88 E..V....... Extend Profile

-level <int> E..V....... Level (from 0 to 52) (default 0)

1.3 13 E..V....... Level 1.3, only available with Baseline Profile

3.0 30 E..V....... Level 3.0

3.1 31 E..V....... Level 3.1

3.2 32 E..V....... Level 3.2

4.0 40 E..V....... Level 4.0

4.1 41 E..V....... Level 4.1

4.2 42 E..V....... Level 4.2

5.0 50 E..V....... Level 5.0

5.1 51 E..V....... Level 5.1

5.2 52 E..V....... Level 5.2

-coder <int> E..V....... Entropy coding (from 0 to 2) (default 0)

cavlc 1 E..V....... CAVLC entropy coding

vlc 1 E..V....... CAVLC entropy coding

cabac 2 E..V....... CABAC entropy coding

ac 2 E..V....... CABAC entropy coding

-a53cc <boolean> E..V....... Use A53 Closed Captions (if available) (default true)

-constant_bit_rate <boolean> E..V....... Require constant bit rate (macOS 13 or newer) (default

false)

-max_slice_bytes <int> E..V....... Set the maximum number of bytes in an H.264 slice. (fr

om -1 to INT_MAX) (default -1)

-allow_sw <boolean> E..V....... Allow software encoding (default false)

-require_sw <boolean> E..V....... Require software encoding (default false)

-realtime <boolean> E..V....... Hint that encoding should happen in real-time if not f

aster (e.g. capturing from camera). (default false)

-frames_before <boolean> E..V....... Other frames will come before the frames in this sessi

on. This helps smooth concatenation issues. (default false)

-frames_after <boolean> E..V....... Other frames will come after the frames in this sessio

n. This helps smooth concatenation issues. (default false)

-prio_speed <boolean> E..V....... prioritize encoding speed (default auto)

-power_efficient <int> E..V....... Set to 1 to enable more power-efficient encoding if su

pported. (from -1 to 1) (default -1)

-max_ref_frames <int> E..V....... Sets the maximum number of reference frames. This only

has an effect when the value is less than the maximum allowed by the profile/level. (from 0 to INT_

MAX) (default 0)

hevc_videotoolbox AVOptions:

-profile <int> E..V....... Profile (from -99 to INT_MAX) (default -99)

main 1 E..V....... Main Profile

main10 2 E..V....... Main10 Profile

-alpha_quality <double> E..V....... Compression quality for the alpha channel (from 0 to 1

) (default 0)

-constant_bit_rate <boolean> E..V....... Require constant bit rate (macOS 13 or newer) (default

false)

-allow_sw <boolean> E..V....... Allow software encoding (default false)

-require_sw <boolean> E..V....... Require software encoding (default false)

-realtime <boolean> E..V....... Hint that encoding should happen in real-time if not f

aster (e.g. capturing from camera). (default false)

-frames_before <boolean> E..V....... Other frames will come before the frames in this sessi

on. This helps smooth concatenation issues. (default false)

-frames_after <boolean> E..V....... Other frames will come after the frames in this sessio

n. This helps smooth concatenation issues. (default false)

-prio_speed <boolean> E..V....... prioritize encoding speed (default auto)

-power_efficient <int> E..V....... Set to 1 to enable more power-efficient encoding if su

pported. (from -1 to 1) (default -1)

-max_ref_frames <int> E..V....... Sets the maximum number of reference frames. This only

has an effect when the value is less than the maximum allowed by the profile/level. (from 0 to INT_

MAX) (default 0)

prores_videotoolbox AVOptions:

-profile <int> E..V....... Profile (from -99 to 5) (default auto)

auto -99 E..V....... Automatically determine based on input format

proxy 0 E..V....... ProRes 422 Proxy

lt 1 E..V....... ProRes 422 LT

standard 2 E..V....... ProRes 422

hq 3 E..V....... ProRes 422 HQ

4444 4 E..V....... ProRes 4444

xq 5 E..V....... ProRes 4444 XQ

-allow_sw <boolean> E..V....... Allow software encoding (default false)

-require_sw <boolean> E..V....... Require software encoding (default false)

-realtime <boolean> E..V....... Hint that encoding should happen in real-time if not f

aster (e.g. capturing from camera). (default false)

-frames_before <boolean> E..V....... Other frames will come before the frames in this sessi

on. This helps smooth concatenation issues. (default false)

-frames_after <boolean> E..V....... Other frames will come after the frames in this sessio

n. This helps smooth concatenation issues. (default false)

-prio_speed <boolean> E..V....... prioritize encoding speed (default auto)

-power_efficient <int> E..V....... Set to 1 to enable more power-efficient encoding if su

pported. (from -1 to 1) (default -1)

-max_ref_frames <int> E..V....... Sets the maximum number of reference frames. This only

has an effect when the value is less than the maximum allowed by the profile/level. (from 0 to INT_

MAX) (default 0)