github kajisho5/ffmpeg-skill v3.10.1

43 minutes ago

Automated release: version and notes generated from pull requests merged since 3.10.0.

Fixed

  • A data stream the container refuses no longer costs the subtitle track beside it (#539). The
    tools that keep a source's subtitle and data streams through one helper (fit, color, crop,
    stabilize, graphics, overlay, audio, loudness, proxy, deinterlace, denoise,
    redact, sphere, straighten, pad --end, freeze of the last frame) retried without all of
    them when one failed: an MP4 with a mov_text track and a start timecode lost the track when
    written to .mkv, and on FFmpeg 6.1.1 to .mp4 too. The data stream is now retried away alone,
    as export.py does; the non_av_streams_dropped fallback carries dropped, the stream types
    left out, and for a source with a subtitle and a data stream dropped_non_av_streams is false
    when the MP4 muxer wrote the timecode track itself (a source with a data stream and no subtitle
    still reports true after its retry). color.py --retag's re-encode fallback has its own copy step and is not changed.
  • metadata.py refuses a stream the output container cannot take as a copy, before ffmpeg runs
    (#539).
    A SubRip MKV written to .mp4 (a mov_text MP4 to .mkv, a timecode track to Matroska,
    an attachment outside Matroska) failed inside ffmpeg after a dry run that said completed; it is
    kind: input now, a dry run included, with a hint naming export.py --preset copy. An MP4 with
    a start timecode, which FFmpeg 6.1.1 would not copy even into .mp4, is written: the timecode
    track alone is left out of the copy and the muxer writes it again (.m4v too). .m4a is judged
    as the MPEG-4 family it is: sound with a SubRip track is refused there as well.
  • A tool that writes a picture refuses an audio-only output extension (#539). fit, pad,
    speedramp, reverse, loop and the other picture tools completed with H.264 inside -o x.m4a
    and failed in the muxer for -o x.wav. Both are refused before ffmpeg runs (kind: input,
    reason output_extension_audio_only), dry runs included; metadata.py refuses it for a source
    with a picture, and waveform.py's own refusal carries the same reason. A stream copy by
    sync.py --replace-audio is not covered.
  • An audio output of audio.py or loudness.py keeps a late sound in place (#539). A source
    whose sound starts after its picture was written from the sound's first sample, so a .wav,
    .flac or .m4a extraction played that much early. The lead is written as silence, as
    cut.py --accurate writes it, for the track --audio-stream selects, and reaches the encoder in
    frames of 4096 samples (FFmpeg 9's FLAC encoder refused the one large frame the pad produced).
    Beside a copied picture nothing changes.
  • graphics.py's 0.3 s fades and the lower third's slide fit a short window (#539). In a
    --start/--end window under 0.3 s the text never faded out, and between 0.3 and 0.6 s its
    brightness jumped; the lower third's bar jumped back in a window under 0.7 s. Each fade is cut
    to half the window and the slide shares it 4:3, on drawtext and on the ASS route; the sticker's
    pop is cut with its fade on both routes, and a lower third whose slides fill the window draws no
    rest phase (its emoji showed for one frame).

Checked on FFmpeg 6.1.1 (Linux) only. Left open in #539 for a decision: caption.py's burn and
waveform.py drop the source's subtitle track, and cut.py -o x.mka re-encodes a picture into the
.mka.

  • fix: a refused stream costs only itself, and an output is what its name says (#554)

What's Changed

  • fix: a refused stream costs only itself, and an output is what its name says by @kajisho5 in #554

Full Changelog: v3.10.0...v3.10.1

Don't miss a new ffmpeg-skill release

NewReleases is sending notifications on new releases.