Preserve every media stream¶

Use this when remuxing a rich media file and losing a second audio track, subtitle, attachment, or data stream would be a failure.

FFmpeg's default stream selection normally chooses one stream of each supported type. That behavior repeatedly surprises users working with multilingual and captioned files: see the independent reports about lost secondary tracks and subtitles, keeping every subtitle stream, and retaining multiple audio tracks.

Inspect, explain, then copy¶

pyffmpegcore probe --input source.mkv --json

pyffmpegcore convert \
  --input source.mkv \
  --output preserved.mkv \
  --preserve-all-streams \
  --explain

pyffmpegcore convert \
  --input source.mkv \
  --output preserved.mkv \
  --preserve-all-streams \
  --receipt preserved.receipt.json

pyffmpegcore probe --input preserved.mkv --json

The plan explicitly maps input 0 and uses stream copy for every mapped stream. Video and audio samples are not decoded or re-encoded. Compatible format metadata and chapters are copied too, and an existing output is still refused unless --force is explicit.

Input and output contract¶

Choose an output container that supports every input codec. Keeping the same container is the safest default; Matroska is usually a practical container for mixed stream types.
--preserve-all-streams intentionally rejects --audio-only, codec, bitrate, thread, and hardware-acceleration options. Use the normal conversion path when re-encoding is the goal.
Attachments and data streams are mapped as well as video, audio, and subtitles. Inspect the plan before writing.
A container incompatibility is reported as a failed job; PyFFmpegCore does not silently drop the unsupported stream.

Verify¶

Compare the before/after probes. The maintained real-media test asserts one video stream, two differently tagged audio streams, one subtitle stream, Unicode container metadata, and two chapters after the copy. The exact wheel also runs this recipe on Linux, macOS, and Windows during release validation.

If you need to re-encode while making custom per-stream choices, use reviewed raw FFmpeg for now and include the complete stream map in any recipe proposal.