Combines 1_*.mp4, 2_*.mp4, ... in numeric order into one MP4 (or
multiple ~60s chunks), with optional fade transitions, codec/resolution/
fps normalisation.
with:)| key | required | default | notes |
|---|---|---|---|
input_dir | yes | — | Directory containing \d+_*.mp4 segments. |
output_path | yes | — | .mp4 path (full mode) or directory (chunk mode). |
mode | no | full | full or chunk. |
transition | no | 0.5 | Fade duration in seconds. 0 disables. |
fps | no | 24 | Target frame rate. |
crf | no | 22 | x264 CRF (0-51, lower = better). |
preset | no | medium | x264 preset. |
For chunk mode, pass chunk_duration directly to the script — the meta
engine entrypoint above does not template it; use a direct shell call
when chunking is needed.
ffmpeg ≥ 5.0ffprobeInstall hints (or just run the bundled installer):
| OS | One-liner installer | Manual |
|---|---|---|
| Windows (PowerShell) | pwsh -ExecutionPolicy Bypass -File install.ps1 | winget install Gyan.FFmpeg / choco install ffmpeg / scoop install ffmpeg |
| macOS | bash install.sh | brew install ffmpeg |
| Debian/Ubuntu | bash install.sh | sudo apt install ffmpeg |
The Windows installer (install.ps1) defaults to winget but accepts$env:OPENSQUILLA_FFMPEG_INSTALLER="choco"|"scoop"|"skip" to switch
backends. It prints the absolute ffmpeg/ffprobe paths after install so
you can pass them to merge.py via --ffmpeg-path / --ffprobe-path
when subprocess PATH inheritance is unreliable.
The script picks files matching ^\d+_.*\.mp4$ and sorts by the leading
integer. Producers must save segments as 1_shot.mp4, 2_shot.mp4,3_shot.mp4, etc. The meta-short-drama workflow already follows this.
Stdout: progress lines and the final output path. Non-zero exit on
ffmpeg failure.
未找到ffmpeg → install ffmpeg, ensure ffmpeg on PATH or pass--ffmpeg-path directly.
未找到符合命名规则...的MP4文件 → ensure segments are \d+_*.mp4.--resolutionis set; otherwise ffmpeg preserves originals (no normalisation).