kajisho5/ffmpeg-skill v3.8.6

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.5.

Fixed

• A number outside a flag's range is refused before ffmpeg runs (kind: input, naming the
  flag and the value; audit A21, A52, A72, A109, A124, A165, A176, A185, A190). Each of these
  completed with something other than what was asked, failed inside ffmpeg as kind: ffmpeg, or
  was a traceback: caption.py --gap below 0 (overlapping cues); sequence.py and fit.py --width/--height of 0 or less (0 was read as "not given"); fit.py --aspect with a negative
  side (0:9 stays "no aspect", as decided; the same rule reaches a render.py project's
  frame.aspect); color.py --peak of 0 or less and --desat below 0; graphics.py --from
  below 0 (--from -1 divided by zero) and --scale of 0 or less; look.py --tiles with a side
  below 1 (0x2 divided by zero) and --width below 2; loudness.py -I/--tp/--lra outside
  loudnorm's -70..-5 / -9..0 / 1..50 (the ranges --versions already checked) and
  --sample-rate of 0 or less; silence.py --min-silence and metadata.py --silence-min below
  0; overlay.py --font-size below 1 (-10 drew nothing and reported verified: true),
  --scale below 1, --scale-percent of 0 or less, --fade and --border below 0; sync.py --max-offset and --analyze-seconds of 0 or less (--max-offset -1 reported an offset it
  never searched for); verify.py --seconds of 0 or less (every step failed and the tools were
  blamed); audio.py --fade-in/--fade-out/--music-fade-out below 0; scenes.py --highlights
  below 0 and, with --highlights, --target/--max-scene of 0 or less. join.py --duration
  below 0 (A52, A185) was already refused. Other numeric flags were not swept here and keep
  their behaviour (docs/design-decisions.md lists the ones looked at and left).
• --timeout below 0 or above 500000 s is refused (A188, A8). A negative value was read as 0
  without a word, which removed every limit; a value whose outer ceiling (4x plus 60 s) did not
  fit the millisecond wait of poll() (1e8, 1e300) raised OverflowError after the child had
  started, so an MCP call reported a failure while the tool ran on and wrote its output.
  FFMPEG_SKILL_TIMEOUT above 500000 is read as 0 (no limit) with the timeout_env_invalid
  warning, as a negative one already was. Measured on Linux only.
• cropdetect.py --limit 1 is the top of its range (A164). FFmpeg reads a limit of 1 or more
  as a pixel code, so the documented top of 0..1 was read as code 1, the opposite end from 0.999;
  it is now written as the top code of the source's bit depth (255 for 8-bit; the 10-bit value
  is computed the same way and was not run).
• A flag the chosen mode does not read is refused instead of ignored (A50, A74, A85, A141,
  A144, A145, A171): caption.py --ass with --animate or --karaoke; audio.py
--music-volume, --music-fade-out or --music-loop without --music, --effects-volume
  without --effects, --denoise-strength without --denoise (a render.py project's
  audio block with these and no music is refused by its audio stage, not before the first
  stage); waveform.py --style spectrum with --color or --waveform-mode; scenes.py --edl,
  --target or --max-scene without --highlights; cut.py --segments with --start, --end
  or --duration; overlay.py --video with --fade (and a render.py video layer's fade, in a run that
  reaches the overlays: not a timeline export, not one stopped before them);
  a render.py template pack with --stop-after, --cache, --from, --work, --keep or
  --export-timeline (--audit-log is now handed to each destination). A flag recognised by
  comparing it with its default (--music-volume, --music-fade-out, --effects-volume,
  --denoise-strength, --max-scene, --color, --waveform-mode) is not refused when the
  default value itself is passed.
• audio.py --voice with --denoise denoises at --denoise-strength (A136). --voice light --denoise dropped the denoise, and the strength was not read under any --voice; --voice medium --denoise at the default strength writes the chain it wrote before.
• caption.py --mode mux --offset shifts every --srt track, not only the first (A61); an
  offset that would leave any track with no cue is refused before a file is written (the first
  track's empty copy used to fail inside ffmpeg and stay behind). --max-lines and
  --min-duration still lay out the first track only; a note now says so when they changed it.
  --text cues that --offset moves before 0, all of them, are refused the same way (an empty
  SRT, or an IndexError when the video's frame rate was known), and a dry run plans a track an
  earlier step has yet to write under its shifted copy's name.
• A project or recipe value no stage reads as written is refused before the first stage (A89,
  A92, A93, A94, A104): a clip's own snap.to other than beats/none on a clip the render
  cuts (it rendered unsnapped); chapters that is neither a file path nor a list (false and 0 stay "no chapters"); an audiogram
  with more than one clip in a render (not in --export-timeline; only the first clip was
  rendered); render.py --template audiogram without --image, unless --write-project (the
  dry run completed and the real run failed at the export); an unknown key in a batch.py
  recipe (a misspelled output_dir, suffix or glob ran with the default; the refusal names
  the nearest valid key when one is close, else lists them).
• fix: a flag value the tool cannot honour is refused before FFmpeg runs (26 audit items) (#504)

What's Changed

• fix: a flag value the tool cannot honour is refused before FFmpeg runs (26 audit items) by @kajisho5 in #504

Full Changelog: v3.8.5...v3.8.6