
Mission 3 of 7
Make a reel from real footage, a real voice and no slop.
A text-card reel with an AI voice over stock footage is easy to spot from across the room. You'll build one from your own screen, a real voice and readable captions, using free tools and a few short scripts.
What you'll do
Set up three free tools that cut and encode video, fetch demos you're allowed to use, and listen to your voice.
Capture your screen clean, crop it to phone shape, make a reusable hook clip and record the voice track.
Generate word-level captions from your own voice, then join everything into one finished, phone-ready video.
- Know where footage comes from, and the one place it must never come from
- The rules on music, AI voice and labelling that keep the account safe
- One scored hook from the last mission and the tool or repo it's about
- A computer with Python 3.10+ and a terminal. Terminal survival kit if that's new
- A microphone that isn't your laptop's, or a voice clone of yourself. Phone earbuds are fine to start
- 75 minutes for the first build. It gets much faster
Your progress0/7
- 01Install the three tools6 min
- 02Where real footage comes from10 min
- 03Record your screen clean, then crop to phone12 min
- 04Make your blurred-presenter hook clip, once12 min
- 05Record or generate the voice10 min
- 06Captions from the voice, three words at a time8 min
- 07Assemble: hook, body, voice, captions, speed, export12 min
Install the three tools
ffmpeg cuts and encodes, yt-dlp fetches, faster-whisper listens. All free, all local.
You'll set these three tools up first, so the commands in the later steps will run. Each one does a single job in the build.
Install all three from a terminal, using the lines for your system:
Terminal# Windows (PowerShell), with winget winget install ffmpeg winget install yt-dlp pip install faster-whisper pillow # Mac brew install ffmpeg yt-dlp pip3 install faster-whisper pillowCheck them:
ffmpeg -version,yt-dlp --version,python -c "import faster_whisper".If all three answer, you've got the whole studio. No CapCut, no subscription, no watermark.
TipLinux:sudo apt install ffmpegandpip install yt-dlp faster-whisper pillow. Ifpipcomplains about an externally managed environment, make a venv first:python3 -m venv ~/reels && source ~/reels/bin/activate.
If it breakswinget isn't recognised + 1 more
winget isn't recognised
bin folder on your PATH.faster-whisper is slow on first run
small model is about 500MB). After that it's fast. On a laptop with no GPU, small transcribes a minute of audio in about 20 seconds.