LeeClip
Turns a 90-minute podcast into ready-to-post vertical clips, entirely on my own PC. No cloud. No subscription. My footage never leaves the machine.
Click play to load the player — nothing is downloaded until you do.
The clipping pipeline
- 01Drop in a long video (podcast, interview, stream, YouTube)
- 02It transcribes the whole thing locally, word by word
- 03It finds the moments worth clipping on its own
- 04Each one gets scored — how strong the hook is, whether it makes sense without the rest of the video, whether the payoff actually lands, and whether the speaker gets genuinely animated
- 05I review, nudge the start/end points, approve what I like
- 06It reframes to 9:16 and follows whoever is talking
- 07Captions burned in, plus a title I drag into place myself
- 08Export the MP4
The part I'm happiest with
The crop follows the speaker. Not a lazy centre crop that frames the middle of the screen while the interesting person talks off to the side. It listens to who's talking and glides the frame onto their face.
Three things in here aren't really about clipping at all
Caption burn
Pick any stretch, or the whole video, and burn word-by-word captions into it at the original resolution and aspect ratio, with my own colour palette. Same transcript the clips use. This is what made my own uploads watchable with the sound off.
Voice cloning
Record half a minute of yourself once, and LeeClip can then speak new lines in your voice, with dials for emotion and delivery so it doesn't come out flat and robotic.
Publishing
Connect a YouTube channel and push a finished export straight up with title, description, privacy and a schedule.
What's under the hood
- FFmpeg for the video processing
- Whisper for transcription — runs fine on CPU, uses the GPU if there is one
- A small local model via Ollama (Qwen / Llama / Gemma) to judge and rank the clips, with an option to plug in any OpenAI-compatible API instead
- OpenCV's YuNet for face detection, pyannote for telling speakers apart
- Chatterbox Turbo for the voice cloning
- Next.js, React, TypeScript and Tailwind for the interface, Node/Express and SQLite behind it