FFmpeg Video Editing Assistant

Problem

Video editing requests often describe outcomes, such as removing the first 30 seconds, adding a cinematic look, or joining several clips, while the actual work still involves ffmpeg arguments, ffprobe inspection, filter chains, encoders, and error diagnosis. This skill turns natural-language editing tasks into a confirmable FFmpeg workflow.

How It Works and Limits

The core flow starts by scanning the working directory and generating MEDIA_INVENTORY.md, which records file resolution, codec, duration, audio tracks, and other key facts. It then selects the right operation:
- Media preparation: uses ffprobe or npx remotion ffprobe to reduce failures caused by mismatched codecs, wrong durations, or bad filter assumptions.
- Cutting and audio: supports trimming, joining, splitting, audio extraction, replacement, mixing, volume control, fades, and noise cleanup.
- Visual and compositing work: covers scaling, cropping, rotation, speed changes, color grading, stabilization, watermarks, subtitles, picture-in-picture, chroma key, and transitions.
- Export and streaming: handles GIFs, thumbnails, slideshows, bitrate, frame rate, metadata, screen capture, webcam capture, RTMP, and HLS.
It does not create creative assets by itself; it translates editing intent into commands and explains tradeoffs before execution, such as fast stream copying versus precise re-encoding. If FFmpeg is absent, the npx fallback may work but only supports a limited codec set, while AV1, FLAC, Opus, and other codecs may require a full FFmpeg install.