Video Editing

This is a complete short-form video pipeline built for Chinese platforms: Xiaohongshu, Douyin, and Weixin Shipin Hao. It takes raw voiceover audio and b-roll, runs transcription with mlx-whisper or faster-whisper, lets you review and correct the transcript with redistributed word timings, then handles rough cuts, scene boundary detection, highlight picking, and LLM-driven script rewriting into five-act story structures. The auto-enrich pass schedules b-roll cutaways, chapter cards, emoji stickers, beat-synced BGM with librosa, and generates abstract concept visuals through Codex's built-in imagegen tool. It runs an 80-plus-rule content linter for platform compliance, renders with heavy CJK fonts and loudness normalization, runs post-render QA, and exports to all three platform aspect ratios with titles, captions, and tags. The storyboard and provider decision modules are especially detailed. You'll need FFmpeg and Python 3.