kajisho5/ffmpeg-skill v3.8.11

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.10.

Fixed

• Compositing and time tools (#460: A168, A169, A170, A173, A174, A177, A178, A179, A180).
  loop.py --crossfade put a picture that starts 0.2 s after its sound 200 ms behind it in every
  repeat; both streams are now shifted by the picture's start. background.py --angle outside
  0-90 failed as kind: ffmpeg, and inside it the far end of the gradient was a seeded random
  point; C1 now sits on the corner the direction starts from and C2 on the opposite one, at any
  angle (the picture at 0 and 90 changes too). redact.py blur/pixelate at an odd --x/--y
  left the rectangle's last column and row as they were and changed one outside it; crop.py cut
  one pixel left of / above an odd position. An MP4/MOV's chapter text track was copied as a stray
  data track holding the old titles by metadata.py (also --clear-chapters), color.py --retag/--strip-dovi, export's data ladder and the tools that keep subtitle/data streams
  through run_keeping_subtitles; the muxer's own chapter track is now the only one. reverse.py's memory_heavy warning is
  keyed on the decoded frames (size, chroma, depth, count; the estimate loop.py --boomerang
  uses) instead of a length over 60 s. A graphics.py --template sticker at a bottom position
  dropped into place on the drawtext route; it rises, as on the ASS route. grid.py labels keep
  an apostrophe and a percent sign. broll.py judges cutaway overlap by time, so windows given
  out of order are no longer refused as overlapping. freeze.py holding the last frame (no
  --at) keeps the subtitle track, as --mode extend does.
• Edge-case inputs (#461: A234, A235, A238, A255, A256, A258). reverse.py on a variable
  frame rate wrote 51 frames for 49 that ran 81 ms past its sound; the graph now closes on the
  rate the output is written at. straighten.py --fit crop wrote 1920x1080 as 1918x1078 (and a
  SAR such as 6112:6111); it writes the displayed size, even, with square pixels. Matroska
  attachments (a subtitle track's fonts) are kept into .mkv/.mka/.mks by the tools that keep
  subtitles (run_keeping_subtitles, export.py including --preset copy, caption.py --mode mux, color.py's re-encoding retag); source_carry gains attachments and video_streams (a
  second playable picture the tool left out), both with the source_items_dropped warning when
  dropped. render.py's result carries source_carry for a project whose clips are one file with
  no other picture or sound source. README and the contract's commands description say that a
  line naming a file the tool removed at exit (a concat list, an ffmetadata file, ...) cannot be
  run again as printed.
• fix: compositing, time and edge-input findings of the audit (#460, #461) (#513)

What's Changed

• fix: compositing, time and edge-input findings of the audit (#460, #461) by @kajisho5 in #513

Full Changelog: v3.8.10...v3.8.11