kajisho5/ffmpeg-skill v3.8.18

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.17.

Fixed

• What was left of the old grouped issues #402, #403, #404 and #429 (#510).
  • broll.py --insert with a still picture and a --from past 0 shows it for the whole
    cutaway; A showed instead, since the still is one frame at 0 and the trim kept nothing (with
    --from 0 it was one frame through 3.8.11; 3.8.12 holds B's last frame), however short the
    window, and a still that image2 gives a one-frame duration (a TGA) is not refused as too
    short. grid.py with a
    still as one cell fills that cell for the grid's length; the still was counted as 0 s, so the
    grid was 0 s long and
    the run ended in a traceback; with a still as the first cell and no -o, the grid is named
    .mp4. A grid of stills only is refused, naming insert.py.
    overlay.py STILL --image LOGO (a still as the main picture) writes one picture; it ran until
    its timeout, since -shortest never ended the looped layer over a main without a duration.
  • A font file that is not there or is not a font is refused (kind: input) by overlay.py --text and graphics.py, from --font-file or a brand file's font_file, and by
    caption.py --mode burn from a brand file's font_file (not by graphics.py --template progress, which draws no text, nor by caption.py with no input video, which writes only the
    SRT). On FFmpeg 6.1.1 the text was drawn in a
    fontconfig default face and the result said font_used.substituted: false.
  • A brand file key that no tool reads is refused (kind: input, with the nearest known key),
    in every block, as render.py refuses an unknown project key. A misspelt key ("safe_margn",
    "colours") was dropped and the default drawn. loudness, which no tool reads and the shipped
    example carried until 3.8.8, is not refused: it gets a note. A brand file that relied on an
    ignored key being accepted now fails until the key is removed.
  • report.py --pack takes only the table render.py --template all writes (its five columns,
    its delimiter row under them, five cells in every later row, the first table in the file);
    any Markdown with a table was rendered as a
    pack and reported verified: true.
  • A test now pins that an output with no integrated loudness gets a loudness step that is not
    ok and says not_measured (the behaviour since 3.6.7).
• fix: leftovers of the audit, the open items of #510 (#532)

What's Changed

• fix: leftovers of the audit, the open items of #510 by @kajisho5 in #532

Full Changelog: v3.8.17...v3.8.18