cobanov/autocut

Autocut is a desktop app that removes silent pauses from a video clip. You drag a video file into the window, the app figures out where the speech is and where the silence is, and you get back a tighter cut. It is meant for podcasters and anyone editing talking head footage who wants to skip the step of trimming dead air by hand. The project is written in Rust with a Tauri shell and a Svelte 5 front end, so it runs as a native desktop application. Supported input formats include MP4, MOV, MKV, WebM, and AVI. After the analysis pass the timeline shows spoken sections in green and silent sections in red, and a preview player jumps over the removed parts so you can hear the final cut while you are still editing. You can drag the green edge handles to nudge a cut, edit exact in and out timestamps in a side panel, or click an X to mute a particular keep without losing it. Four sliders control the detection: threshold, pad, minimum silence, and minimum speech. Holding shift gives finer steps. When you are happy, you export. The MP4 export gives you a finished video file you can share right away. The FCPXML export instead writes out a timeline that DaVinci Resolve and Adobe Premiere can open, keeping the cuts in place and preserving the source timecode so the clip relinks to the original media file without a media offline warning. The README also notes a preview range only option for long videos, which lets you tune the sliders on a short slice before processing the whole file. Install paths differ per platform. On Apple Silicon Macs the recommended way is Homebrew with brew install cask cobanov tap autocut, since Homebrew handles the macOS Gatekeeper quarantine flag automatically. A manual download of the dmg also works, but you need to run xattr cr on the app once because the bundle is not yet notarized. On Windows there is a 64 bit NSIS installer and an MSI, and Windows SmartScreen will warn on first launch because the bundle is unsigned, so you click More info then Run anyway. Linux users currently have to build from source. The README states that everything runs locally, with no account, no upload step, no Python, and no separate ffmpeg install. It is credited to mert cobanov, dated 2026.