Ozzy Sports Gives Live Streams an AI Heckler

Field notes

The browser captures two or three stream frames every few seconds, sends them to a vision model, and converts the resulting line into speech before mixing it with the broadcast audio. A left-minus-right centre-channel cancellation process is used to reduce the original commentator while retaining stereo crowd ambience. The documented controls include family-friendly, cheeky, and full Ozzy styles, plus pacing, volume, cancellation, and free-text event context. Private network fetching is blocked by default in the resolver and proxy.