Tell me a story
Press the mic and talk (Chrome / Android use the browser's speech-to-text), or tap a sample story.
Reader age
Art style
My book
Pages appear here. Each page shows the image prompt a real model would get.
No story yet
Pipeline for the real MVP
Same flow as Kid mode, with the stack named in your post (Google Speech-to-Text, OpenAI, Stable Diffusion, React Native or Flutter).
1. VoiceOn-device record, upload to Google Speech-to-Text (child voice model, profanity filter on).~2-4 s
2. Story shapingOpenAI splits the narration into 4-6 scenes, keeps the kid's words, fixes only grammar for the reading level.~3 s
3. Safety gateModeration pass on text and prompts; blocked words swapped for friendly ones; parent gets a log.~1 s
4. PicturesStable Diffusion with a fixed character sheet so the hero looks the same on every page.~4-8 s per page, parallel
5. BookPages stored per child profile; read-aloud with highlighted words for literacy.offline cache
| Decision | Recommendation for MVP | Why |
|---|---|---|
| App shell | React Native (Expo) | One codebase, fast store review, easy hiring of outsourced devs |
| Backend | Node.js API + queue (one worker per image) | Images run in parallel, the app never waits on a single slow call |
| Character consistency | Seeded prompts + reference image per hero | The number one thing kids notice when it breaks |
| Kids privacy | COPPA: parent account owns data, no voice kept after transcription by default | App store and legal risk before launch, not after |
| Cost per book | Est. USD 0.05 to 0.25 (5 pages) | Depends on image model and resolution; we would measure it in week 2 |
A 90-day lane, 10 to 15 h a week
- Weeks 1-2: architecture doc, stack decision, cost-per-book test with 3 image models, hiring scorecards for the outsourced roles.
- Weeks 3-6: voice to book working end to end on a test build; safety gate and parent account.
- Weeks 7-10: character consistency, read-aloud, offline books; 10 families testing.
- Weeks 11-12: store submission, analytics, roadmap for post-MVP.
Prometio runs build and QA with AI agents alongside the people, so a part-time CTO still ships every week.