QVAC

Education

TaleTrip

Bilingual storybooks generated and illustrated for traveling families, read offline on iPad after the first sync

image-generationragp2p
TaleTrip

Kids' AI content tends to be generic, cloud-bound, and rarely about the child's own life. TaleTrip turns a real family trip into a bilingual, illustrated storybook starring the child, generated entirely on a parent's Mac and delivered to the kid's iPad over peer-to-peer - no cloud, no accounts, and nothing about the child ever leaves a device.

Plan, research, write, paint

A Qwen3-4B agent plans the book and researches the destination with EmbeddingGemma RAG grounded in real destination facts and the family's favorite picture-book style. Llama-3.2-1B writes the bilingual page text, and SDXL paints a matching illustration for every page. The result - a StoryPack - is seeded straight to the kid's iPad over a pear:// Hyperdrive/Hyperswarm link.

how it works (README.md)
Parent Studio (Mac)  --Qwen3 4B plans the book-->  Research (EmbeddingGemma RAG)
  --Llama 3.2 1B writes-->  Bilingual page text + vocab
  --SDXL paints-->  Per-page illustrations
  both --> StoryPack --seed over pear:// P2P--> iPad (Kid app)
  iPad --> Read aloud, tap-to-learn Spanish, TTS
  iPad --> Scavenger Hunt (SmolVLM)
  iPad --> Photo Story
  iPad --> Sleep Coach (jet-lag bedtime plan)

Beyond the book

The same on-device models power what happens after the story: a Scavenger Hunt where SmolVLM2-500M recognizes real-world objects through the camera, a Photo Story mode that captions photos and writes a diary entry, and a Sleep Coach that drafts a jet-lag bedtime plan with MedPsy-1.7B guardrails.

Demo video

Why it's more than a bedtime-story generator

  • Generation, illustration, RAG, and speech all run on-device - nothing about the trip reaches a server

  • The finished storybook is delivered to the kid's iPad over a direct pear:// P2P link, not the cloud

  • A bundled demo book ships as a fallback for a known P2P image-transfer edge case

Build this

Ship your own local AI

One install in JavaScript or Python. 10+ AI tasks. Private, offline and free - on the hardware you already own.