如何用FFmpeg精准分割H.264+MP3编码的MP4视频？

Hi there! Let's break down your questions about getting accurate cuts on your MP4 video step by step:

1. Manual Keyframe Insertion: How to Do It & Does It Work? #

The reason your first command (-c copy) didn't give precise cuts is because H.264 is a frame-interleaved codec—only keyframes are self-contained, while non-keyframes depend on prior frames. When copying streams directly, FFmpeg can only start cutting at the nearest existing keyframe before your desired start time.

To fix this by inserting keyframes manually:

• First, add a keyframe exactly at (or just before) your intended start time with this command:

  ffmpeg -i input-file.mp4 -force_key_frames "expr:gte(t,<your-start-time>)" -c:v libx264 -c:a copy temp-keyframed.mp4
  

  Replace <your-start-time> with the exact timestamp you want to cut from (e.g., 30 for 30 seconds).
• Then use your original copy command on the temp file:

  ffmpeg -ss <starttime> -i temp-keyframed.mp4 -to <duration> -c copy out-file.mp4
  

This method works reliably: inserting a keyframe at your target start point lets FFmpeg copy from that exact spot without re-encoding the entire video (only the small segment around the inserted keyframe gets reprocessed).

2. Precision of Re-Encoding Video (Your Second Command) #

When you re-encode the video (pro tip: use libx264 explicitly instead of just h264 for better control), FFmpeg decodes all frames first, then starts encoding from your exact specified <starttime>. This gives you millisecond-level precision, since it's no longer constrained by existing keyframes. You'll get the exact start and end points you're aiming for.

Do You Need to Re-Encode Audio? #

Your audio is MP3, which uses intra-frame encoding (each MP3 frame is independent). Copying the audio stream (-acodec copy) is almost always sufficient:

• If your audio start time aligns with an MP3 frame boundary, it'll be perfectly precise.
• If not, the deviation is tiny (usually less than 20ms) and undetectable to the human ear.
  Only re-encode audio if you need to adjust its format, bitrate, or fix sync issues—otherwise, copying is faster and preserves original quality.

3. Is the Trim Filter a Better Option? #

Absolutely! The trim filter is another excellent choice for precise trimming, especially if you want to guarantee perfect audio-video sync. It processes frames through FFmpeg's filtergraph, letting you cut both video and audio exactly at your desired timestamps.

Example command:

ffmpeg -i input-file.mp4 \
  -filter:v "trim=start=<starttime>:end=<endtime>,setpts=PTS-STARTPTS" \
  -filter:a "atrim=start=<starttime>:end=<endtime>,asetpts=PTS-STARTPTS" \
  -c:v libx264 -c:a copy out-file.mp4

• The setpts/asetpts filters reset the timestamp counters so the output video/audio starts at 0, avoiding any sync drift.
• Like re-encoding, this requires processing the video stream, so it'll take longer than copying—but it delivers maximum precision.

Quick Option Recap #

• Fast, minor precision tradeoff: Stick with your first command (-ss before -i, -c copy).
• Precise, minimal re-encoding: Insert keyframes first, then copy streams.
• Ultimate precision & sync: Use optimized re-encoding or the trim filter.

内容的提问来源于stack exchange，提问作者user9876266