mintsy: turning any book into a five-minute video series
One chapter, one episode, five minutes. I designed and built the whole system: a pipeline that turns a book PDF into narrated, animated episodes, the tools to review and publish them, the API, and the iPhone app that streams them.



Books take hours. People have minutes.
Most of us have a stack of books we keep meaning to read. The time we actually have comes in five-minute pockets: a commute, a coffee, the end of the day. Video fits those pockets, but making a good narrated, animated video for every chapter of every book by hand would take a studio, not one person.
Problem statementHow might one person turn a whole book into a chapter-by-chapter video series people can finish five minutes at a time, without a production team?
- Time: a book is hours; a sitting is minutes
- Scale: writing, voicing and animating every chapter by hand doesn't scale
- Trust: every episode has to stick to what the chapter actually says
Find the format first, then the machine that can make it
Before writing code I worked out what an episode should be, which books could be used, and which tools could produce it reliably on one machine.
- A
The format
How long a sitting really is, and what a viewer should walk away with.
One chapter per episode: about 700 words of narration, a one-line summary and three takeaways.
- B
The books
Which titles can be adapted and published without rights problems.
Start with public-domain books from Project Gutenberg, behind a rights screen.
- C
The production tools
Writing, narration and animation that a script can drive end to end.
Claude writes, Kokoro narrates locally with word timings, Remotion renders video from code.
- D
The viewing experience
How to stream smoothly on a phone and let people pick who reads to them.
Adaptive HLS streaming, three narrator voices, and captions drawn by the app instead of burned into the video.
A studio in a pipeline, and an app to watch it in
Upload a book PDF and the studio does the rest: it finds the chapters, writes each episode, animates it, voices it in three narrators, renders and uploads it. A person reviews before anything goes live. Here's what was built.
- AI step
- Human review
- Ingest · Claude reads the PDF once: title, category, chapters (AI step)
- Script · ~700-word narration, summary, 3 takeaways (AI step)
- Scenes · 10–16 collage scenes, timed to cue phrases (AI step)
- Voice · Kokoro, local: 3 narrators with word timings
- Render · Remotion: one video per voice + thumbnail
- Upload · HLS (1080p / 720p / 360p) to Cloud Storage, draft saved

- RevenueCat subscriptions
- Sign in with Apple
- Push notifications
- Resend email
- getmintsy.com
Every step leaves a file
Each stage writes something you can open and edit. Hand edits survive, and any step can be re-run on its own.
People approve what goes live
Scripts and videos are reviewed in the dashboard. Nothing is public until someone publishes it.
Checked, not hoped
Narration is copied word for word into scenes, word counts and takeaways are validated, and a failed step saves nothing.
What was delivered
- A studio pipeline that turns a book PDF into narrated, animated episodes, one per chapter
- An admin dashboard to upload, review, edit and publish, plus an autopilot for public-domain books
- A GraphQL backend on Cloud Run with accounts, subscriptions, push and email
- A native iPhone app in SwiftUI, with narrator choice and captions (first milestone done)
- A brand system and website: the Ink & Coral theme, logos and getmintsy.com
From the studio
