菜单
Professional video and audio processing - transcode, edit, stream, and analyze media files. Use when transcoding, editing, streaming, or analysing video and audio files.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/DreamLab-AI/agentbox --skill ffmpeg-processing命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | ffmpeg-processing |
| description | Professional video and audio processing - transcode, edit, stream, and analyze media files. Use when transcoding, editing, streaming, or analysing video and audio files. |
FFmpeg 8.0 for video/audio transcoding, editing, streaming, and analysis. The recipes/ directory holds production-grade compound commands for automation pipelines; this file is the quick-reference index.
imagemagickcomfyuiblendermermaid-diagrams or report-builderplaywright or browserFor production patterns (most automation tasks land here), open the matching recipe:
| Recipe | Use when |
|---|---|
| recipes/vertical-reframing.md | Reframe 16:9 → 9:16 with pan-and-scan, multi-output (YT + Shorts in one pass) |
| recipes/kenburns-slideshow.md | Image → video, slideshows with xfade, Ken Burns zoompan |
| recipes/intro-main-outro.md | Concat clips with normalised formats, background music with fade and amix duration=first |
| recipes/styled-subtitles.md | Burn styled SRT with custom font, embed soft subs with default disposition |
| recipes/timed-text-overlays.md | drawtext with alpha=if(...) fade-in, enable= time gating, textfile/fontfile |
| recipes/audio-mixing.md | Replace/mix audio, crossfades, mono panning, dynamic normalisation |
| recipes/thumbnails-storyboards.md | Scene-detected thumbnails, tile=NxM storyboards, keyframe-only extracts |
| recipes/gotchas.md | Input vs output seeking, -c copy traps, hvc1 for Apple H.265, yuv420p for QuickTime, -vsync deprecation |
| recipes/gpu-encoding.md | NVENC, Intel QSV, VAAPI matrix and tradeoffs |
Foundation for everything else.
| Syntax | Meaning |
|---|---|
[0:v] | Video stream from first input |
[1:a] | Audio stream from second input |
0:v:0 | First video stream of first input (0-indexed) |
0:a:1 | Second audio stream of first input |
[name] | Named filter output (used inside -filter_complex) |
-map [name] | Route a named stream to output |
-map 0:v -map 1:a | Take video from input 0, audio from input 1 |
-vf / -af | Simple video / audio filter chain (one input, one output) |
-filter_complex | Multi-input/output filter graph |
-y | Auto-overwrite output (put at start of every command) |
ffmpeg -i input.mp4 # Show file info
ffprobe -v quiet -print_format json -show_format -show_streams input.mp4
# Duration / resolution / bitrate
ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 input.mp4
ffprobe -v error -select_streams v:0 -show_entries stream=width,height -of csv=s=x:p=0 input.mp4
ffprobe -v error -show_entries format=bit_rate -of default=noprint_wrappers=1:nokey=1 input.mp4
ffmpeg -y -i input.avi output.mp4 # Re-encode
ffmpeg -y -i input.mp4 -c copy output.mkv # Remux (no re-encode, fast)
The correct production form -- preserves aspect, pads black, normalises SAR:
ffmpeg -y -i input.mp4 -vf \
"scale=w=1920:h=1080:force_original_aspect_ratio=decrease,\
pad=1920:1080:(ow-iw)/2:(oh-ih)/2:color=black,setsar=1:1" \
output.mp4
For naive scaling without padding (only when source aspect already matches):
ffmpeg -y -i input.mp4 -vf scale=-2:720 output.mp4
ffmpeg -y -i input.mp4 -c:v libx264 -crf 18 -preset veryslow output.mp4 # Archive quality
ffmpeg -y -i input.mp4 -b:v 2M -b:a 192k output.mp4 # Fixed bitrate
CRF: 0 lossless, 17-18 visually lossless, 23 default, 28+ small/lossy. ±6 ≈ ½× or 2× filesize.
ffmpeg -y -i video.mp4 -vn -acodec libmp3lame -q:a 2 audio.mp3 # MP3 high quality
ffmpeg -y -i video.mp4 -vn -c:a copy audio.aac # AAC remux (fastest)
ffmpeg -y -i video.mp4 -map 0:a:0 -acodec copy audio.aac # Specific stream
ffmpeg -y -i input.mp4 frame%04d.png # All
ffmpeg -y -i input.mp4 -vf fps=1 frame%04d.png # 1 fps
ffmpeg -y -ss 00:01:30 -i input.mp4 -frames:v 1 -q:v 2 thumb.jpg # At timestamp
Output seeking (frame-accurate, requires re-encode for clean cuts):
ffmpeg -y -i input.mp4 -ss 00:00:10 -to 00:00:30 output.mp4
Input seeking (fast, keyframe-aligned, may produce black frames with -c copy):
ffmpeg -y -ss 00:00:10 -to 00:00:30 -i input.mp4 -c copy output.mp4
See recipes/gotchas.md for the full tradeoff.
# list.txt:
# file 'a.mp4'
# file 'b.mp4'
ffmpeg -y -f concat -safe 0 -i list.txt -c copy output.mp4
For mixed sources, see recipes/intro-main-outro.md.
Good baseline for VOD / archival / multi-device playback:
ffmpeg -y -i input.mp4 \
-c:v libx264 -crf 18 -preset veryslow -tune fastdecode \
-pix_fmt yuv420p -movflags +faststart \
-c:a aac -b:a 192k \
output.mp4
ffmpeg -y -i input.mp4 \
-c:v libx264 -preset slow -crf 18 -pix_fmt yuv420p \
-c:a aac -b:a 192k -movflags +faststart \
youtube.mp4
ffmpeg -y -i input.mp4 -vf "fps=10,scale=480:-1:flags=lanczos,palettegen" palette.png
ffmpeg -y -i input.mp4 -i palette.png \
-lavfi "fps=10,scale=480:-1:flags=lanczos[x];[x][1:v]paletteuse" output.gif
# HLS
ffmpeg -y -i input.mp4 -codec: copy -start_number 0 -hls_time 10 -hls_list_size 0 -f hls playlist.m3u8
# RTMP live
ffmpeg -re -i input.mp4 -c:v libx264 -preset veryfast -maxrate 3000k -bufsize 6000k \
-pix_fmt yuv420p -g 50 -c:a aac -b:a 160k -ac 2 -f flv rtmp://server/live/streamkey
ffmpeg -y -hwaccel cuda -hwaccel_output_format cuda -i input.mp4 \
-c:v h264_nvenc -preset p4 output.mp4
For full GPU matrix (NVENC, QSV, VAAPI) and tradeoffs, see recipes/gpu-encoding.md.
for f in *.mkv; do ffmpeg -y -i "$f" -c copy "${f%.mkv}.mp4"; done
for f in *.mp4; do ffmpeg -y -ss 00:00:05 -i "$f" -frames:v 1 "${f%.mp4}.jpg"; done
libx264) -- universal compatibility, default for MP4libx265) -- ~30-50% smaller than H.264; add -tag:v hvc1 for Apple/iOS playbacklibvpx-vp9) -- WebM, 20-50% smaller than H.264 at same quality, slower encodelibaom-av1) -- best compression, slowest encode; libsvtav1 is the fast alternativeaac) -- universal, default for MP4libmp3lame) -- legacy compatibilitylibopus) -- best quality/bitrate; default for WebMflac) -- lossless-c copy whenever possible (no re-encode) -- but see recipes/gotchas.mdultrafast/superfast/veryfast/faster/fast/medium/slow/slower/veryslow -- slower = better compression at same quality-threads 0 lets FFmpeg pick (default); only override for specific reasons-progress pipe:1 for parseable progress output in pipelines-filter_complex with split produces multiple outputs in one decode pass -- always cheaper than running ffmpeg twice# Repair attempt
ffmpeg -y -err_detect ignore_err -i corrupted.mp4 -c copy repaired.mp4
# Fix A/V sync (offset audio +500ms)
ffmpeg -y -i input.mp4 -itsoffset 0.5 -i input.mp4 -map 0:v -map 1:a -c copy output.mp4
# Force pixel format compatibility (QuickTime, iOS, older players)
ffmpeg -y -i input.mp4 -c:v libx264 -pix_fmt yuv420p -c:a copy output.mp4
-y for non-interactive automation