Product Build
LuxeTide Clipper
2-minute walkthrough, recorded on the app itself.
Overview
Personal product, native macOS app
A free, on-device alternative to OpusClip. Drop in a long video or paste a YouTube, TikTok or Instagram link, and it finds the moments worth posting, writes the title hooks, and exports ready-to-post clips with burned-in captions. Nothing leaves the Mac.
The Problem
AI clipping tools charge by the minute and upload every hour of footage to someone else's servers. For a creator clipping long streams every week, that is a recurring bill and a privacy trade-off for a job a modern laptop can do on its own.
Finding the Moments
Every video is transcribed locally with Whisper (large-v3-turbo via whisper.cpp). A local language model, Qwen3 running in Ollama, reads the transcript, scores each candidate moment from 0 to 100 for how likely it is to stop the scroll, drops anything under the cutoff, and writes a short title hook for each clip. Audio event detection flags laughter and cheering, and on-screen text detection catches gameplay moments like eliminations and wins.
Ask in Plain English
Instead of scrubbing a timeline, you can ask for what you want: the funniest moments, every time a topic comes up, or the exact spot where a phrase was said. The model turns the request into a search plan, and the app runs it against the transcript, the audio events and the screen, then snaps each result to clean start and end points.
Edit and Export
A click-to-clip transcript editor with frame-accurate trims. FFmpeg renders each clip to 9:16, 4:5, 1:1 or 16:9 with layout options and burned-in captions, and finished clips sync to the phone through iCloud so they are ready to post.
Private by Default
Everything runs locally, from transcription to scoring to rendering. An optional cloud scorer exists, but it is off unless explicitly enabled, and the app falls back to the local model whenever there is no key or no connection.
The Engineering
A Python backend served locally, a lightweight studio interface, and a native macOS app built with Tauri, kept alive by a self-healing background service. The hard parts were the real-world ones: detecting files that iCloud has offloaded and pulling them back before processing, keeping multi-gigabyte working files off synced folders, and queueing requests to a local model that answers one at a time so waiting never counts against a timeout. Built with AI-assisted development, at zero cost per clip.
Stack