Unlimited reels, zero credits: the setup

How it works

• Script: an AI writes the script from one line, using hook formulas.
• Voice: a free text-to-speech model (Kokoro) records the voiceover on your laptop.
• Scenes: every scene is a small web page (HTML + JavaScript) animated by time.
• Frames: a headless browser (Playwright) screenshots every frame.
• Video: FFmpeg joins the frames, voice and sound effects into an MP4.
• Review: the AI looks at sample frames and fixes anything that overlaps or gets cut off.

What to install (all free)

The starter prompt

Open an empty folder in Claude Code and paste this:

Build me a reel engine in this folder.
- Input: projects/<slug>/video.json with scenes (type, say, data).
- voice.py: speak each scene's 'say' with kokoro-onnx, save voice.wav
  and word timings to timeline.json.
- player.html: a 1080x1920 stage; each scene type is a function of time
  t (hook, list, chat, stat, cta). Captions follow the word timings.
  Expose window.renderAt(t) so every frame is deterministic.
- render.js: open player.html with Playwright, call renderAt for every
  frame at 30 fps, pipe screenshots into ffmpeg, add voice.wav.
- Before rendering, screenshot 1 frame per scene to build/review/ and
  check nothing overlaps. Then make a 30-second test reel about [topic].

Tips

• Keep one scene = one idea = one sentence of voiceover.
• Write a brand file once (colours, fonts, handle) and reuse it.
• Always review the frames before posting.