VX — AI Video Editor
CLI tool that turns raw trip footage into polished vlogs using AI. Automates editorial thinking — clip review, story structure, and rough cut assembly.
VX exists because most people come back from trips with hours of raw clips and never edit them. The gap between "raw footage on a hard drive" and "a video worth sharing" is enormous. VX automates the editor's thinking, not just the cutting — it reviews every clip, understands the story the footage can tell, and assembles a structured edit.
The pipeline starts with ffmpeg preprocessing (proxy generation, scene detection, audio extraction), then uses Google Gemini's multimodal AI to review each clip, understand context through an interactive briefing, and produce a complete editorial storyboard with precise timestamps. The output is a structured EDL that assembles into a rough cut without human intervention.
Key technical decisions include software decode with hardware encode (VideoToolbox), forced IDR keyframes at segment boundaries, multi-layer validation before concatenation, and a versioned library system that preserves every analysis run. The interactive HTML preview lets you adjust cut points and re-export without re-running the AI.
// Like this project?
// Discussion