In-house tool (not public) AI video production
MiVid
One podcast in, a week of short videos out.
Turns a long podcast into short vertical clips with animated captions, or a voiceover into a finished motivational video.
In-house tool · live demo on request
The problem
What was getting in the way
Cutting a podcast into social clips takes hours of finding moments, reframing, captioning and styling each one.
What I built
What it does
- AI finds the most shareable moments in a long podcast and cuts them out
- Converts landscape to vertical by detecting shots and following faces, including two-speaker shots
- Word-by-word animated captions that avoid covering faces
- 1,800+ fonts, 56 animations, 30 text styles, one-click themes and color grades
- Caption styles learned from a screenshot of a look you like
- Voiceover mode: upload a voiceover and get a full video with B-roll or 16 animated scenes matched to the words
- Background music mixed at the right volume automatically
How it's built
Under the hood
A Python pipeline (ffmpeg, Whisper transcription, face detection, Gemini) that makes every caption decision, and Remotion (React) that renders the final video frame by frame.
Python
FastAPI
ffmpeg
faster-whisper
OpenCV
Google Gemini
Remotion
Proof
Results so far
- Two complete modes: podcast clips and voiceover videos
- Deterministic: the same input and settings always produce the identical video
Proves these services