Tiny Narrator

Build Small Hackathon concept

A tiny model reader that turns articles into guided narration

Tiny Narrator is an accessibility-first article experience. It can describe generated images, build a reading path, and speak each part of the page with a lightweight local voice.

Illustration of an article reader with narration controls and a highlighted reading path.
The article view doubles as the demo surface, so every feature has a real reading task.

Why it belongs in a tiny-model hackathon

Accessibility tools should feel immediate, private, and personal. Small models help because they can run closer to the reader, keep latency low, and make the experience easier to inspect.

The goal is not to read the whole page at the user. The goal is to make the page navigable by sound.

The model map

The reader brain runs through llama.cpp, the vision model writes practical alt text, Kokoro speaks the final narration, and a four-billion-parameter image model creates article illustrations.

Diagram showing the reader brain, vision, speech, and image generation models working together.
Each model stays at or below four billion parameters for Tiny Titan eligibility.