Green dominance g − max(r,b) > threshold ⇒ transparent. Near-edge pixels get despill correction, not a hard cut, to avoid ghosting/fringing.
Preview
Render
Chroma key, blur/darken and karaoke captions are composited frame-by-frame in 32-bit RGBA canvas space, then encoded client-side with WebCodecs (avc1) + AAC audio and muxed into a standard, moov-first MP4 — no cloud transcoding step.
Layer 3 · Subtitles Overlay
Karaoke: unspoken words render white, the currently-spoken word renders in the highlight color, synced to canvas frame time.