# Rezzo

> A desktop app for cutting vertical clips out of podcasts and talk shows.

I post podcast clips on social media, and the work is the same every time: find an interesting moment, crop a wide shot down to 9:16, keep whoever is talking in frame, add captions. The apps that handle this well either stamp a watermark on the video or put the useful parts behind a monthly subscription. Rezzo does the same job for free, and every feature is in the free version because there is no other version. It's made for Serbian shows first, since that's what I work with.

## How it works

You start with a whole episode. Rezzo transcribes it with a time for every word, always in Latin script, even if the speech model hands back Cyrillic. While the video plays you can search the transcript, and fix the words Whisper got wrong one after another from the keyboard. Words it always gets wrong (it writes "futbal" every single time) go on a replacement list, and every new transcript gets them fixed.

## Framing and captions

Select a few sentences and they become a clip. Rezzo tells the voices in the episode apart, and when you open a clip it looks for camera cuts and people. Press Auto frame and you get one of three layouts for each shot: following the person talking, two people stacked on top of each other (Split), or the whole picture over a blurred copy of itself. What it does is ordinary keyframes, so you can change any of it by hand. Every clip gets captions that appear word by word and mark the word being spoken.

## Under the hood

Electron, React and TypeScript. FFmpeg comes with the app, whisper.cpp does the transcription, MediaPipe finds people and faces, and sherpa-onnx tells voices apart. A single WebGL renderer draws both the preview and the export, so the file looks like what you saw while editing. Before any of it was built I checked that the hard parts work on an ordinary PC.

stack: Electron React TypeScript FFmpeg whisper.cpp MediaPipe WebGL

kind: Desktop app, open source, 2026

status: v0.1.0 · Windows

link: github.com/dzonyv/rezzo // Source on GitHub