使用 ffmpeg 修剪 HEVC 视频：正确的持续时间，但 output 文件大小太大

Question

I have a digitized 8mm 16fps HEVC.mp4 that I'm trying to split into clips using ffmpeg (5.1.2) on MacOS terminal. The output file trims to the correct duration but the filesize is too large.

ffmpeg -ss 05:09 -i input.mp4 -t 02:19 -an -sn -dn -c copy output.mp4

input.mp4 : 11:56 in duration and 8.24 GB in filesize (bitrate 674251 kbits/s).
output.mp4 : 02:19 in duration and 5.15 GB in filesize (too large).

5.15 GB is the filesize I'd expect for a clip that is 05:09 + 02:19 in duration at input.mp4 's 674251 Kbits/s:
(05:09+02:19) * 674251 kbits/sec = 5.16 GB

I've tried various -ss start_pos and -t duration values with similar results. The output file is correctly duration in time and unexpectedly ( start_pos + duration )*bitrate in filesize.

I've consulted this ShotStack.io article on trimming with ffmpeg and this ffmpeg wiki section on cutting small sections from files. Neither specifically states that the resulting file size will be trimmed to match the duration of the clip, but I'd think that'd be useful.

Addendum :
My input file has only one keyframe at t=0.000s. Likely the root of all my problems.

ffmpeg stdout :

[mov,mp4,m4a,3gp,3g2,mj2 @ 0x14ff04760] Using non-standard frame rate 16384/1024
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'input.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2mp41
    creation_time   : 2021-07-24T11:23:44.000000Z
    encoder         : Lavf58.45.100
  Duration: 00:11:56.13, start: 0.000000, bitrate: 92041 kb/s
  Stream #0:0[0x1](und): Video: hevc (Main) (hvc1 / 0x31637668), yuv420p(tv, bt709/bt709/unknown, progressive), 5120x3840 [SAR 1:1 DAR 4:3], 91715 kb/s, 16 fps, 16 tbr, 16384 tbn (default)
    Metadata:
      creation_time   : 2021-07-24T11:23:44.000000Z
      handler_name    : VideoHandler
      vendor_id       : [0][0][0][0]
      timecode        : 17:15:11:05
  Stream #0:1[0x2](und): Audio: aac (LC) (mp4a / 0x6134706D), 48000 Hz, mono, fltp, 159 kb/s (default)
    Metadata:
      creation_time   : 2021-07-24T11:23:44.000000Z
      handler_name    : SoundHandler
      vendor_id       : [0][0][0][0]
  Stream #0:2[0x3](und): Audio: aac (LC) (mp4a / 0x6134706D), 48000 Hz, mono, fltp, 159 kb/s (default)
    Metadata:
      creation_time   : 2021-07-24T11:23:44.000000Z
      handler_name    : SoundHandler
      vendor_id       : [0][0][0][0]
  Stream #0:3[0x4](eng): Data: none (tmcd / 0x64636D74)
    Metadata:
      creation_time   : 2021-07-24T11:23:44.000000Z
      handler_name    : TimeCodeHandler
      timecode        : 17:15:11:05
File 'output.mp4' already exists. Overwrite? [y/N] y
[mp4 @ 0x14ff08660] Using non-standard frame rate 16/1
    Last message repeated 1 times
Output #0, mp4, to 'output.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2mp41
    encoder         : Lavf59.27.100
  Stream #0:0(und): Video: hevc (Main) (hvc1 / 0x31637668), yuv420p(tv, bt709/bt709/unknown, progressive), 5120x3840 [SAR 1:1 DAR 4:3], q=2-31, 91715 kb/s, 16 fps, 16 tbr, 16384 tbn (default)
    Metadata:
      creation_time   : 2021-07-24T11:23:44.000000Z
      handler_name    : VideoHandler
      vendor_id       : [0][0][0][0]
      timecode        : 17:15:11:05
Stream mapping:
  Stream #0:0 -> #0:0 (copy)
Press [q] to stop, [?] for help
frame= 4290 fps=0.0 q=-1.0 Lsize= 2988507kB time=00:02:18.93 bitrate=176207.6kbits/s speed= 189x    
video:2988420kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.002924%

ffprobe output.mp4

[mov,mp4,m4a,3gp,3g2,mj2 @ 0x130804080] Using non-standard frame rate 16384/1024
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'output.mp4':
  Metadata:
    major_brand     : isom
    minor_version   : 512
    compatible_brands: isomiso2mp41
    encoder         : Lavf59.27.100
  Duration: 00:02:19.31, start: 0.000000, bitrate: 175732 kb/s
  Stream #0:0[0x1](und): Video: hevc (Main) (hvc1 / 0x31637668), yuv420p(tv, bt709/bt709/unknown, progressive), 5120x3840 [SAR 1:1 DAR 4:3], 91304 kb/s, 16 fps, 16 tbr, 16384 tbn (default)
    Metadata:
      handler_name    : VideoHandler
      vendor_id       : [0][0][0][0]
      timecode        : 17:15:11:05
  Stream #0:1[0x2](eng): Data: none (tmcd / 0x64636D74)
    Metadata:
      handler_name    : VideoHandler
      timecode        : 17:15:11:05
Unsupported codec with id 0 for input stream 1

Answer 1

input.mp4 is unusual in that it only has one keyframe at t=0.000s, all others are B-frames (verified with ffprobe ).

The ffmpeg command is correct and the output is a result of there only being one keyframe. When using the -c copy coder, all trims/ splices/cuts have to be made at the prior keyframe, which for this input file is only at t=0.

input.mp4 and desired trim clip:

Frame type:    Ibbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb...
Keyframe:      ^
Trim Points:                        |-----------------|
Time:          0                    ^start_pos        ^start_pos+duration

ffmpeg -ss start_pos -i input.mp4 -t duration -c copy output.mp4

output.mp4 :

Frame type:    Ibbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb???
Keyframe:      ^
Trim Points:                        |-----------------|
Time:          ^-start_pos          0                 ^duration

All frames from start_pos back to the prior keyframe (which in this file is the first frame) have to be included in output.mp4 .

Decoding of output.mp4 starts at the beginning (the keyframe) but frames are not displayed until t=0. In a large file, that could incur significant delay before displaying video.

I'm not sure how many extra frames are needed after start_pos+duration , but some will be because the end of the clip has B -frames that can reference several frames B i-directionally (backwards and forwards).

Alas, my unsatisfying solution for this problem is to lossless re-encode the input.mp4 (adding in more keyframes), trim out my desired clips, and archive the outputs.