kajisho5/ffmpeg-skill v3.8.19

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.18.

Fixed

• An output named like a file the run reads but does not hand to ffmpeg as -i is refused
  (#522). sync.py REF SECOND --trim-second --overwrite replaced REF when REF was named
  SECOND_synced.mp4 (the default output name) or given as -o: the command reads SECOND alone,
  so the output-is-input guard never saw REF. The same held for cut.py --snap-source,
  metadata.py --chapters, graphics.py and overlay.py --brand/--font-file, silence.py --words/--filler-words and join.py --list. Each is now refused, kind: input,
  reason: output_is_input, before anything is written and in a dry run.
• verify.py no longer replaces an existing <stem>_cues.txt in --out without
  --overwrite (#522). The caption step fails on it (reason: output_exists) and the file is
  left as it was; a link or a folder at that name is never written through. A --report that
  names a file the steps write is refused before any step runs (reason: output_path_taken); it
  replaced that file with the Markdown.
• A blurred pad on a small frame no longer fails inside ffmpeg (#527). fit.py --fit blur,
  --pad-fill blur and waveform.py --image-fit blur gave the chroma planes the luma's blur
  radius, which a 4:2:0 frame's half-size planes cannot hold (a 160x120 source to 9:16: "radius 20
  ... must be <= 17"). redact.py's blur assumed 4:2:0 and failed on a 4:1:1 or 4:1:0 source.
  The chroma radius is now measured on the planes' own size; on a frame large enough the output
  is unchanged.
• fit.py --smooth interpolate --fps N makes its in-between frames at N (#527). They were
  made at the source's rate and then repeated to reach N: a 2x slow motion to 60 fps from 30 had
  every other frame a copy. A VFR source whose speed change conforms the picture now says so on
  stderr, as one conformed by the fps filter does.
• sequence.py refuses a frame that does not decode (#529). Only the first frame was probed;
  a later file named as an image that is not one (a text file called .png, a truncated PNG) was
  left out while the frame before it was shown in its place, and the run reported success. Every
  frame is decoded once before the encode (ffprobe, so a dry run checks too) and the run is
  refused, kind: input, reason: still_unreadable, naming the frames. When ffprobe cannot read
  the frame list back, a note says the frames were not checked and the encode runs.
• A side file named like a brand file's font or logo is refused (review of #534).
  graphics.py --write-ass, caption.py --write-srt/--write-ass (and its --plan), and
  render.py --template --write-project with --overwrite replaced the brand's font_file or
  logo with the text they write; only the brand file itself was guarded.
• fix: sync --trim-second and other side inputs as outputs, the blurred pad on a small frame and sequence's bad frames (#534)

What's Changed

• fix: sync --trim-second and other side inputs as outputs, the blurred pad on a small frame and sequence's bad frames by @kajisho5 in #534

Full Changelog: v3.8.18...v3.8.19