Ch.1
Capture
Frame-by-frame vision
Sampled the video in sequence and sent the frames to Gemini for context-aware analysis of the play.
Off air
Sunset archive
Commentrix
Commentrix watched game film, wrote play-by-play commentary, and voiced it in real broadcaster style — an AI sports-commentary experiment that's now off the air.
Reel — what it explored
Automated commentary for sports video
Commentrix tested whether a video model could spot the moments that matter, write commentary for them, and hand back a voice track that lines up with the footage.
Ch.2
Commentary
Synthetic narration
Turned the selected commentary lines into spoken audio with ElevenLabs voice synthesis.
Ch.3
Broadcast
Video and subtitles
Rendered a final video track with an SRT file that keeps commentary in sync with the source footage.
Reel — prototype notes
The MVP code stays in the repository
The FastAPI MVP is kept for reference, but it isn't deployed on this site. The live Cloudflare Pages surface serves only this static landing archive.
Running the MVP locally needs your own API keys and a local input video. Generated videos, subtitles, virtual environments, and environment files are intentionally left out of the public repo.
Off air
Current status
Commentrix has been sunset.
The project isn't taking on customers, demos, pilots, or support requests anymore. This site stays up as a public record of the concept and the prototype that ran it.
Reel — FAQ
Common questions
Is Commentrix available to use?
No. Commentrix is a sunset project and isn't offered as a commercial product.
What models did the prototype use?
The MVP was built around Gemini for video-frame understanding and ElevenLabs for text-to-speech output.
Can the MVP run from this website?
No. Cloudflare Pages serves only the static archive in the docs/ folder.