-crf 18 -preset 8 -tune film -profile:v high -level 51 -pix_fmt yuv420p -maxrate 24M -bufsize 48M -refs 4 -bf 6 -r 30000/1001 -s 1440x1080 -g 290 -keyint_min 1 -fast-pskip 0 -me_method umh -me_range 32 -subq 10 -aq-mode 2 -aq-strength 0.9 -trellis 2 -psy-rd 0.8:0.05 -ar 48000 -b:a 256k -ac 2 -c:a aac -af loudnorm -max_muxing_queue_size 2222
-c:v libx265 -pix_fmt yuv420p10le -crf 20 -preset 8 -x265-params "open-gop=0:min-keyint=1:keyint=192:rd=4:ref=6:subme=5:rc-lookahead=60:rect=0:rskip=0:aq-mode=3"
ffmpeg -i Main.mp4 -i Refs.mp4 -map v:0 -lavfi ssim -f null -ffmpeg -i XXX.mkv -map v:0 -f hash -hash murmur3 -
- 分别计算帧序列 [60,65) 的Hash值：
ffmpeg -i XXX.mkv -map v:0 -vf trim=start_frame=60:end_frame=65 -f framehash -hash murmur3 -
- 以下适用于4K屏幕场景，观看距离为1.5倍屏幕高度；记录的帧率可不同，但帧数量与分辨率皆相同（默认皆为逐行扫描）：
ffmpeg -r 1 -i Main.mp4 -r 1 -i Refs.mp4 -map v:0 -lavfi [0:v][1:v]libvmaf=model=version=vmaf_4k_v0.6.1:n_threads=16 -f null -
- 以下适用于1080P屏幕场景，观看距离为3倍屏幕高度；记录的帧率可不同，帧未精确对齐、切黑边的情况，并生成CSV文件以供分析：
ffmpeg -r 1 -colorspace bt709 -i Mian.mp4 -r 1 -colorspace bt709 -i Refs.mp4 -hide_banner -map_chapters -1 -map_metadata -1 -an -sn -dn -lavfi "[0:v:0]trim=start_frame=5000:end_frame=15000,crop=1904:1024[main];[1:v:0]trim=start_frame=5000:end_frame=15000,crop=1904:1024[refs];[main][refs]libvmaf=n_threads=16:log_fmt=csv:log_path=Main.csv" -f null -
- 用Powershell计算 1% Low of VMAF
$csv = Import-Csv Main.csv; ([double[]]$csv.vmaf | sort | select -First ($csv.Count/100) | measure -AllStats).Average
for %a in (*.mp4 *.flv) do ffmpeg -i "%a" -crf 20 "output\%~na_cfr-20.mp4"
先建立output文件夹；若要保存为.bat，则需将%替换为%%
Get-ChildItem *.jpg | ForEach-Object { ffmpeg -i $_.Name -lossless 1 "$($_.BaseName).webp" }ffmpeg -i XXX.mkv -map 0 -c copy -f segment -segment_time 17 -reset_timestamps 1 -segment_list XXX.ffcat XXX_%3d.mp4
ffmpeg -i XXX.ffcat -c copy -video_track_timescale 15360 .\XXX-C.mp4
适合 MPEG CFR; 以上命令每17秒切一刀; 切片首帧为关键帧
-segment_times 13,18,55以每个时间点之后的关键帧为切割点，若切割时间点与上一个相同，则顺延到下一个关键帧
保持对某一片段重编码后的拼接完整性：
-pix_fmt yuv420p -crf 1 -preset 1
http://ffmpeg.org/ffmpeg-formats.html#segment_002c-stream_005fsegment_002c-ssegment
ffmpeg -i XXX.m2ts -map v:0 -vf trim=start_frame=5000:end_frame=15000,setpts=PTS-STARTPTS,fps=24000/1001,crop=1920:800 -preset 0 -qp 0 -colorspace bt709 -color_primaries bt709 -color_trc bt709 -color_range tv YYY.mp4
(for %i in (*.flv) do @echo file '%i') > mylist.txt
Get-ChildItem *.mp4 | ForEach-Object { Write-Output "file '$($_.Name)'" } | Out-File mylist.txt ffmpeg -f concat -i mylist.txt -c copy YYY.mkv
ffmpeg -ss 00:00:18.000 -t 15 -i XXX.mp4 -c copy -avoid_negative_ts make_zero YYY.mp4
-avoid_negative_ts make_zero: 从指定起始位置的上一关键帧开始裁切，末端或有缺失帧，起始时间戳或略大于零
若输出容器格式为mkv，可省略
-avoid_negative_ts make_zero; 但mkv默认时间精度为 1k tbn ，较低不适合作为抽取的目标容器
若
to位于[input]之后，则视为时间段t
-ss 18 -t 15 -i ...与-ss 18 -to 33 -i ...等效
假设输入视频的起始时间戳为12.000，若...-ss 14...，则从输入视频的原始时间戳26.000开始裁切
| 优点 | 缺点 | |
|---|---|---|
有 -c copy，不重新编码 |
瞬间提取 | 实际时间范围不太精确 |
无 -c copy，需重新编码 |
实际时间范围准确：18.000s~28.000s | 重编码缓慢，CPU满负载 |
ffmpeg -i input.mkv -map 0:1 -map 0:2 -c copy audio_only.mkv -map 0:0 -c copy video_only.mkv
ffmpeg -i XXX.mkv -map 0 -c copy -disposition:a:0 0 -disposition:a:2 default YYY.mkv
ffmpeg -stream_loop 3 -i XXX.wav -c copy XXX_x4.wav
Set number of times input stream shall be looped. Loop 0 means no loop, loop -1 means infinite loop.
ffmpeg -i XXX.mp4 -i XXX.aac -c copy YYY.mp4
合并后时长取较长段。若视频较长，则后半段音量为零；若音频较长，则后半段为视频的最后一帧
ffmpeg -i XXX.mp4 -map_metadata 0 -metadata:s:v rotate="90" -c copy YYY.mp4
ffmpeg -i XXX.mp4 -vf crop=w:h:x:y,scale=3840:-2 YYY.mp4
ffmpeg -i XXX.mp4 -vf "pad=1920:1080:(ow-iw)/2:(oh-ih)/2" YYY.mp4
ffmpeg -i XXX.mp4 -vf scale=1920:1080:force_original_aspect_ratio=decrease,pad=1920:1080:-1:-1 YYY.mp4
ffmpeg -i XXX.mp4 -vf "fade=t=in:st=0:d=5,fade=t=out:st=55:d=5"
0-5秒：淡入； 55-60秒：淡出
ffmpeg -i 30fps.mp4 -lavfi "setpts=0.5*PTS;atempo=2" -r 60 60fps.mp4
https://trac.ffmpeg.org/wiki/How%20to%20speed%20up%20/%20slow%20down%20a%20video
ffmpeg -filter_complex ddagrab=framerate=60,hwdownload,format=bgra,setparams=range=tv:colorspace=bt709:color_trc=bt709:color_primaries=bt709,scale=flags=accurate_rnd+full_chroma_int+bitexact+lanczos,format=yuv420p -c:v h264_amf -profile:v high -quality quality -preset balanced -qp_i 12 -qp_p 12 YYY.mkv
ffmpeg -probesize 64M -f gdigrab -framerate 30 -i desktop -qp 0 -preset 0 -level 51 YYY.mp4
ffmpeg -list_devices true -f dshow -i dummy
ffmpeg -f dshow -i audio="麦克风阵列 (Realtek(R) Audio)" YYY.wav
ffmpeg -f dshow -i audio="立体声混音 (Realtek(R) Audio)" YYY.wav
录制桌面音频需开启麦克风权限并将声音输入设备改为立体声混音
播放音量会影响录制音量
- Encode:
ffmpeg -i XXX.mp4 -c:v h264_nvenc -profile:v high -rc-lookahead 32 -bf 4 -b_ref_mode 2 -temporal_aq 1 -spatial_aq 1 -aq-strength 15 -b:v 0 -bufsize 0 -keyint_min 1 -g 300 -an -preset p7 -qp 16 YYY.mp4
- Decode:
ffmpeg -hwaccel cuda -hwaccel_output_format cuda -i XXX.mkv ......
https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new
https://developer.nvidia.com/blog/nvidia-ffmpeg-transcoding-guide/
https://github.com/Xaymar/obs-StreamFX/wiki/Encoder-FFmpeg-NVENC
https://blog.xaymar.com/2020/06/24/the-art-of-encoding-with-nvidia-turing-nvenc/
https://github.com/rigaya/NVEnc/blob/master/NVEncC_Options.zh-cn.md
ffmpeg -i XXX.gif -vf scale=320:-1,fps=15 -loop 0 -lossles 1 -y YYY.webp
ffmpeg -ss 5 -t 7 -i XXX.mp4 -vf fps=10,scale=480:-2 -loop 0 YYY.gif
- 高质量版：
ffmpeg -i XXX.mp4 -vf "fps=10,scale=480:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 YYY.gif
ffmpeg -i XXX.mp4 -vf "drawtext=fontfile=C\\:/Windows/fonts/consola.ttf:text='Abg0123':x=20:y=H-th-20:fontsize=30:fontcolor=white:borderw=3:bordercolor=black" YYY.mp4
ffmpeg -f lavfi -i "color=c=0x333333:s=1920x1080:r=10,drawtext=fontfile=C\\:/Windows/fonts/consola.ttf:fontsize=96:fontcolor='white':timecode='00\:00\:00\:00':rate=10:text='TCR\:':boxcolor=0x000000AA:box=1:x=960-text_w/2:y=540" -g 100 -keyint_min 100 -t 60 YYY.mp4
rate = video fps
ffmpeg -i XXX.mp4 -vf "drawtext=fontfile=C\\:/Windows/fonts/consola.ttf:text='test0test':x=100:y=100:enable=lt(mod(t\,1.6)\,0.8):fontsize=30:fontcolor=blue" YYY.mp4
ffmpeg -i XXX.mp4 -i XXX.png -filter_complex overlay=20:20:enable='between(t,10,16)' YYY.mp4
右上：overlay=W-w-20:20
右下：overlay=W-w-20:H-h-20
左下：overlay=20:H-h-20
居中：overlay=(W-w)/2:(H-h)/2
ffmpeg -i XXX.mp4 -i XXX.png -filter_complex "overlay='if(gte(t,5), -w+(t-5)*400, NAN)':(H-h)/2" YYY.mp4
ffmpeg -i XXX.mp4 -ignore_loop 0 -i XXX.gif -filter_complex overlay=20:20:shortest=1 YYY.mp4
ffmpeg -i XXX.mp4 -i XXXX.mp4 -filter_complex "overlay=x='if(gte(t,2), -w+(t-2)*400, NAN)':y=0" YYY.mp4
ffmpeg -colorspace bt709 -i XXX.mkv -sws_flags accurate_rnd+full_chroma_int+bitexact -map v:0 -vf select='not(mod(n\,2143))',crop=1920:800,format=rgb24 -fps_mode passthrough -pred 2 -frame_pts 1 -v 16 -stats Frame.%06d.png
ffmpeg -loop 1 -t 60 -framerate 30 -i XXX.bmp -vf setparams=range=tv:colorspace=bt709:color_trc=bt709:color_primaries=bt709,scale=flags=accurate_rnd+full_chroma_int+bitexact+lanczos,format=yuv420p YYY.mp4
-vf zscale=range=limited:matrix=709,format=yuv420p -color_trc bt709 -color_primaries bt709
-vf zscale=t=linear:npl=100,format=gbrpf32le,zscale=p=bt709,tonemap=tonemap=hable:desat=0,zscale=t=bt709:m=bt709:r=tv,format=yuv420p,zscale=1920:-2
Requirement: https://github.com/jellyfin/jellyfin-ffmpeg/releases
https://www.reddit.com/r/ffmpeg/comments/yn5krm/comment/iv7a3ui/
-init_hw_device opencl:0 -i XXX.mp4 -vf hwupload,tonemap_opencl=tonemap=bt2390:desat=0:peak=100:format=nv12,hwdownload,format=nv12
ffmpeg -h encoder=prores_ks
ffmpeg -i XXX.mp4 -c:v prores_ks -profile:v 4 -pix_fmt yuva444p10le -c:a pcm_s16le YYY.mov
https://video.stackexchange.com/questions/14712/how-to-encode-apple-prores-on-windows-or-linux
相比于prores_ks, prores牺牲了压缩率提升了编码速度（
profile:v也适用于prores）
- RMS:
ffmpeg -i XXX.mp4 -af volumedetect -f null nul
- EBU R128:
ffmpeg -i XXX.mp4 -map a:0 -af ebur128=peak=true:framelog=verbose -f null -
ffmpeg-normalize audio.m4a -vn -sn -mn -cn --keep-loudness-range-target -t -17 -tp -1 -ar 48000 -o output.wav
ffmpeg -i XXX.mp4 -c:v copy -af volume=1.5:enable='between(t,10,16)' YYY.mp4
Volume与CRF算法相似，volume=0.5 相当于 volume=-6dB
ffmpeg -i XXX.wav -filter_complex "showwavespic=s=1920x1080:split_channels=1" -frames:v 1 YYY.png
ffmpeg -i XXX.mp4 -y -v 16 -lavfi showspectrumpic=s=3584x2048 XXX.png
高度需为2^n; 默认值
4096x2048可能导致频谱轻微变形
ffmpeg -i 01.wav -i 02.wav -filter_complex amix=inputs=2:duration=first:dropout_transition=2 mix.wav
ffmpeg -i XXX.m2ts -map 0:2 -c:s copy YYY.sup
ffmpeg -ss 12:34 -copyts -i XXX.mkv -ss 12:34 -t 56 -vf "subtitles=XXX.srt:force_style='Fontname=Source Han Serif,Fontsize=28,Outline=2,MarginV=20,PrimaryColour=&H8515C7'" YYY.mkv
&H8515C7 = #C71586
ffmpeg -analyzeduration 100M -probesize 100M -ss 38:10 -t 20 -i XXX.mkv -lavfi "[v:0]setpts=PTS-STARTPTS,pad=3840:2160:(ow-iw)/2:(oh-ih)/2[v1]; [s:0]setpts=PTS-STARTPTS,scale=3840:2160[s1]; [v1][s1]overlay[out]" -map "[out]" -map a:0 YYY.mkv
ffmpeg -i XXX.mp4 -i XXX.srt -map 0:v:0 -map 0:a:0 -map 1 -c:v copy -c:a copy -c:s mov_text -metadata:s:s:0 language=chi YYY.mp4
ffprobe -v error -show_format -show_streams XXX.mp4
ffprobe -v error -select_streams v:0 -show_entries frame=pts_time,pkt_size,pict_type -of csv=p=0 XXX.mp4 >XXX.csv
ffprobe -v 16 -select_streams v:0 -skip_frame nokey -show_entries frame=pts_time -of csv=print_section=0 XXX.mp4
ffprobe -v 8 -count_frames -select_streams v:0 -skip_frame nokey -show_entries stream=nb_read_frames -of default=nokey=1:noprint_wrappers=1 XXX.mp4
-skip_frame nokey: Keyframes
-skip_frame nointra: I frames
-skip_frame bidir: except B frames
https://stackoverflow.com/questions/2017843/fetch-frame-count-with-ffmpeg
ffprobe -read_intervals 01:25:31%+#1 -show_entries frame=key_frame,pts_time -of json -v 16 XXX.mkv
ffprobe -v 16 -f lavfi movie=XXX.mp4,trim=start=20:end=40,signalstats -show_entries frame=pts_time:frame_tags=lavfi.signalstats.YAVG -of csv=p=0 >YAVG.csv; $Data = Import-Csv YAVG.csv -Header pts_time,YAVG; $Data | Export-Csv YAVG.csv -NoTypeInformation
ffmpeg -i XXX.mp3 -i XXX.png -map 0:0 -map 1:0 -c copy -id3v2_version 3 -write_id3v1 1 -metadata title="?" -metadata artist="?" -metadata album="?" -metadata comment="Cover (front)" YYY.mp3
ffplay XXX.mkv -fs -ast 2 -vf subtitles=XXX.mkv:si=0
ffplay -v 16 -fs -f lavfi "movie=XXX.mp4,fps=source_fps,format=gbrp10le[A];movie=YYY.mp4,fps=source_fps,format=gbrp10le[B];[A][B]blend=all_mode=difference,eq=gamma=1.5"
对比立体声的FR声道：
ffplay -v 16 -fs -f lavfi "amovie=XXX.m4a,pan=stereo|c0=FR,showspectrumpic=s=3584x2048,drawbox=y=2112:t=fill,format=gbrp[B];amovie=XXX.wav,pan=stereo|c0=FR,showspectrumpic=s=3584x2048,drawbox=w=142:t=fill,format=gbrp[C];[B][C]blend=all_mode=difference"
对比5.1(side)的FC声道：
ffmpeg -v 16 -i XXX.m4a -i XXX.wav -lavfi "[0:a]pan=5.1(side)|c0=FC,showspectrumpic=s=7226x4096:stop=20000:fscale=log,drawbox=y=4160:t=fill,format=gbrp[B];[1:a]pan=5.1(side)|c0=FC,showspectrumpic=s=7226x4096:stop=20000:fscale=log,drawbox=w=142:t=fill,format=gbrp[C];[B][C]blend=all_mode=6,format=gbrp,drawbox=y=1020:h=4:c=yellow,drawbox=y=2660:h=4:c=cyan" XXX.qoi
https://ffmpeg.org/ffmpeg-utils.html#toc-Channel-Layout
https://qoiformat.org/
ffplay -vcodec hevc_cuvid -an -x 960 -y 540 XXX.mp4
metaflac --remove-all XXX.flac
metaflac --import-tags-from=FlacTags.txt --import-picture-from=cover.jpg XXX.flac
FlacTags.txt (ANSI编码)