kajisho5/ffmpeg-skill v3.6.3

on GitHub

Automated release: version and notes generated from pull requests merged since 3.6.2.

Fixed

• join.py cuts every clip to whole frames and its sound to whole samples (#455). Each clip
  was trimmed to its length in milliseconds (trim=duration=0.167 for 5 frames at 30 fps), and the
  next clip started after it: 60 five-frame clips had one-frame holes from about 8 s. The picture
  is now cut with trim=end_frame, the sound with atrim=end_sample at the samples those frames
  last (counted from the running total, so the sound stays within a sample of the picture), at the
  first clip's exact rate (30000/1001, where the fps filter used to run at 29.97). The result
  reports frame_rate, clip_frames and transition_frames.
• A join.py transition starts on its frame (#468). The xfade offset was written to the
  millisecond, and xfade starts on the first frame at or after it: 2.566667 s written 2.567
  started the fade a frame late (138 frames where 137 were due). Offsets are now frames of the
  output rate, written rounded down to the microsecond, the transition is a whole number of frames
  (at least one: --duration 0 with a transition built acrossfade=ns=0, which FFmpeg refuses)
  and each acrossfade overlaps the samples of those frames, taken from the output's rounded frame
  positions so the sound never runs a sample longer per transition.
• join.py reads a short clip without a last-frame duration at its own rate (#475). FFmpeg 7.1
  read a 5-frame MOV that stores no duration for its last frame as 0.1333 s at 37.5 fps; join.py
  ran the whole join at that rate and held every clip to its 0.1667 s sound, 7 frames each (60
  clips: 418 frames for 300). The clip's declared r_frame_rate is used when the average is short
  by exactly the last frame, and its picture counts every frame it holds.
• A slow latest on npm no longer stops the MCP Registry publish. The release job waited five
  minutes for npm's latest to read the new version and then failed, which skipped every later
  step: 3.6.2 was on npm but not on the MCP Registry until the job was re-run. It now waits for
  npm to have the version (that is what the registry steps need), runs the registry steps, and
  only then fails the job if latest is still behind after ten minutes.
• fix(join): whole-frame clip lengths and frame-exact xfade offsets; release waits for the npm version, not for latest (#473)

What's Changed

• fix(join): whole-frame clip lengths and frame-exact xfade offsets; release waits for the npm version, not for latest by @kajisho5 in #473

Full Changelog: v3.6.2...v3.6.3