kajisho5/ffmpeg-skill v3.1.1 on GitHub

Automated release: version and notes generated from pull requests merged since 3.1.0.

The GitHub Pages site opens on a landing page (site/index.html, one static HTML file in the
Velisa design, English and Japanese) instead of a copy of the README. The README moved to
readme.html; the other pages are unchanged. build_pages.py copies site/ to the site root
and fails on a relative link in it that points nowhere, as it already did for the Markdown.
Two demos from AI-generated footage: reframe_talk (a speaker kept in a 9:16 crop with
--crop-x) and filler_en (two hesitations cut from English speech). The README and the demo
gallery show GIFs; the landing page plays both as MP4 with sound (site/media/), built by
demos/pages_videos.py with the skill's own grid.py, graphics.py, join.py and look.py.

silence.py --filler --transcribe found no filler words on clean speech with a Whisper engine:
Whisper writes tidy text and leaves "uh"/"um" out unless its context shows them. The filler list
of the language asked for (English under auto) is now passed as the initial prompt (whisper.cpp
--prompt, faster-whisper initial_prompt, openai-whisper --initial_prompt). On the new demo
clip, faster-whisper base found neither hesitation without it and both with it.
Prime Whisper with the filler list; two demos from AI-generated footage (#391)
Pages: open on a landing page instead of a copy of the README (#390)

What's Changed

Pages: open on a landing page instead of a copy of the README by @kajisho5 in #390
Prime Whisper with the filler list; two demos from AI-generated footage by @kajisho5 in #391

Full Changelog: v3.1.0...v3.1.1