Tuft: a nature quest that turns your screen black, judged by an open model on your own device

Description and use

Tuft is an offline mobile app that hands users a brief nature quest, turns the screen black to reduce distraction, and after a timer ends takes a single photo. An on‑device vision model (Ollama or CLIP) judges the photo against the quest description, gives a spoken verdict, and logs the result locally.

Problem

Users need a simple way to encourage outdoor activity without constant phone use and to verify completion of nature quests locally without relying on cloud services.

Outcome

Tuft enables outdoor engagement by eliminating phone distraction, providing local vision-based quest validation, and requiring no backend or internet after initial asset download.

Workflow

• 1. User selects quest duration and taps "Go outside".
• 2. Screen turns black and a timer starts with wake lock and voice cue.
• 3. After timer ends, phone buzzes and prompts for a photo.
• 4. User takes one photo; no retakes allowed.
• 5. Vision model judges photo against quest description and "not this" labels.
• 6. Spoken verdict is given and result logged locally with thumbnail.

Evidence

Tuft gives you one small quest, like find moss or lichen, find sky between branches, or find a shadow with a shape.

Source

When the timer ends the phone buzzes and talks to you. You take one photo, no retakes. A vision model decides whether it completes the quest, says what it saw, and drops the result into a small field log on your device, thumbnail included.

Source

If Ollama is running and has a vision model, Tuft uses it. It picks up Gemma, LLaVA and similar models on its own, and the same model can write new quests.

Source

A service worker caches the page, the CLIP library and its files the first time they load, and transformers.js caches the model weights.

Source