kajisho5/ffmpeg-skill v3.8.10

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.9.

Fixed

• Captions and text drawing: the thirteen open items of audit #457, and #505.
  • Gurmukhi, Gujarati, Telugu, Kannada, Malayalam, Sinhala, Khmer and Myanmar text is detected
    (A65). It was read as Latin, so caption.py, graphics.py and overlay.py --text drew it in
    the Latin font and reported the run verified; now a font that covers the script is picked (or
    the job is refused when none is installed), graphics.py renders it through libass, and
    overlay.py --text refuses it as it refuses Devanagari. A grid.py label in one of them is
    drawn with a font that covers it (by drawtext, unshaped, as a Thai label is). doctor --json
    lists the eight under fonts.scripts.
  • --wrap phrase treats the fullwidth ，；：． as the kinsoku table treats 、 and 。 (A54: a
    line opened with ，). Its first fill no longer breaks before one of those characters (or after
    an opening bracket), and the orphan and weak-line passes no longer move one down to open a
    line: both happened in cues of three or more lines and in the pieces of a cue split past
    --max-lines. A break after ，；： is preferred, as after 、. Where no allowed break fits
    (a bracketed run wider than the line, a line of about 3 em, two such characters in a row) a
    forbidden one is still taken. --wrap measured is unchanged. graphics.py's wrapped titles
    go through the same wrap and get the same fix.
  • caption.py: an emoji PNG in a left- or right-aligned caption is placed from the Style's side
    margins, not --margin (A62): on the left it covered the first words.
  • --karaoke with word timings that match the cue's word count waits for the first word (A59) instead of leading the
    speech by the cue's opening silence, and --offset moves the word timings with the cues
    (#505): --offset 0.05 had cut the first word short and 0.2 dropped the timings for an even
    split.
  • --animate slide ends where the line is drawn without it, for every --position, a cue's
    {\anN} and a platform's side margins (A56); it always ended at the bottom centre.
  • --karaoke-style word with --animate: the caption fades out at the cue's end, not at the
    end of the first word, and a slide or a pop carries on across the word events instead of
    jumping (A57).
  • The size fit measures against the --karaoke-style word budget (A58), so a cue reported as
    fitting is not split.
  • --mode mux adds a soft subtitle track without wrapping or splitting it at the burn-in size
    (A55; one cue on a 360x640 video became three); a --max-lines other than the default 2, or --wrap measured,
    with --mode mux is refused (the default is accepted and has no effect there).
  • The emoji placeholder no longer resets the line's own overrides for the text after an emoji
    (A69): the pop animation, a per-cue size and a cue's italics, bold and colour now carry past
    it (caption.py and graphics.py's ASS route).
  • doctor --json fonts.emoji.mode is mono when libass draws an emoji glyph and no colour
    emoji family is installed (A64); it said none, and a caption.py --emoji auto run left the
    emoji out of the text.
  • font_used.file under caption.py --bold names the face fontconfig matches for the family at
    bold weight (A70), on Linux and macOS, where the face is resolved through fontconfig; it named
    the regular face. With --ass, or --fonts-dir and no script font, it stays null.
  • In a burn, --language is the script and wrap hint only (A71): --language ja_JP was refused
    as --srt: 'ja_JP' is not a language code; the wrap now reads ja_JP as ja, as the font
    pick already did. With --mode mux the refusal names --language,
    and a --text/--transcribe track's tag is checked as an --srt one is.
  • An SRT burn draws text in braces (A68): ffmpeg's SRT reader hands {curly} to libass as an
    override block, which draws nothing, while the result said text_unchanged: true. Such cues
    are drawn through the generated ASS, braces included, and a note says so; {\...} blocks on
    their own are left to ffmpeg's reader as before.
• fix: captions and text drawing, the open items of audit #457, and #505 (#511)

What's Changed

• fix: captions and text drawing, the open items of audit #457, and #505 by @kajisho5 in #511

Full Changelog: v3.8.9...v3.8.10