kajisho5/ffmpeg-skill v3.9.0

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.23.

Added

• fit.py says where the picture sits in the frame it wrote (#519). After a reframe
  (--aspect, --width or --height) the result carries picture: the picture's rectangle
  {x, y, width, height} in the written frame, frame_share (its share of the frame's area),
  scale (its width over the source's) and the source pixels --fit crop cut on each side
  (cropped). It is measured, not judged: ffmpeg runs the frame's own filters on a white and a
  black frame of the source's size and pixel format, and the picture is where the two differ, so
  --aspect 9:16 --width 360 on a 640x360 clip reports 360x202 at y 218 in a 360x640 frame, as
  the written file has it (no frame of the source is decoded). A picture scaled down with no --width/--height
  asked gets a note. subject_framing stays in unverified.

Fixed

• A still picture given where a clip with a length is needed is refused as an input (#531).
  cut.py wrote 0.04 s with verified: true from --start 0 and failed as kind: output past
  the one frame; fit.py --duration refused a speed of 0.00x or wrote one frame as verified
  (--method trim); freeze.py, speedramp.py and loop.py quoted a length of 0.000;
  metadata.py --chapters failed inside ffmpeg. Each is now kind: input, naming the still and
  insert.py, which makes a clip of one. freeze.py --at 0 still holds a still's one frame.
• look.py --at past a still's one frame, and --compare past either clip's end, is refused
  (#531). Both were kind: output ("not written"), where a plain clip with a time past its end
  got a kind: input refusal; --compare checked neither picture's length. --at 0 on a TGA
  still (which image2 gives a length of one frame) wrote nothing and now shows it.
• stabilize.py takes an RGB source (#531). vid.stab aborts on RGB (a PNG still, a qtrle
  or RGB .mov: kind: ffmpeg); both passes now stabilize a YUV 4:4:4 copy, converted to BT.709
  as every SDR RGB re-encode is (an HDR or BT.2020 one by FFmpeg's own conversion, as on every
  other line), and a note says so. The output's pixel format is the encoder line's,
  as before (probe.video.pix_fmt).
• caption.py --fonts-dir DIR --font NAME reports the font libass drew with (#531). With no
  face named NAME in DIR, libass asks fontconfig, and the result said
  font_used.substituted: false whatever it got. DIR is now read for the face (fc-scan); without
  one, the font is reported as a run without --fonts-dir reports it: substituted: true, with a
  font_substituted fallback, only when fontconfig gives another family. A face counts as found
  only as libass 0.17 matches one: the folder's own files (not subfolders), by family name, or by
  full name for TrueType and PostScript name for CFF. A file is named only for a face in exactly
  the asked style (upright, regular or --bold) or one asked by its full or PostScript name; a
  folder holding the family in other styles only gets file: null and a note, since libass then
  chooses among those and the installed faces. font_used names one face: libass draws a glyph
  it lacks from another face, not recorded.
• feat: fit reports where the picture sits; a still where a clip is expected is refused, stabilize takes RGB input (#543)

What's Changed

• fix: a still where a clip is expected is refused as an input, stabilize takes RGB input, and fit reports where the picture sits by @kajisho5 in #543

Full Changelog: v3.8.23...v3.9.0