build-small-hackathon/attenborough-junk-drawer

🎙️ Attenborough for Your Junk Drawer

A wildlife documentary about the things on your desk.

Point your webcam at anything gloriously mundane — a messy desk, a fruit bowl, a tangle of charging cables — and a tiny vision model narrates it as an epic nature documentary. Your stapler becomes an apex predator. Your houseplant, an ancient sentinel of the canopy. The app even reads the narration aloud in a hushed British documentary voice.

The whole thing runs on OpenBMB MiniCPM-V 4.6 (~1.3B params) — proof that a model small enough to live on your laptop can still make you laugh out loud.

🍄 Track

An Adventure in Thousand Token Wood — the AI is load-bearing. The joke only works because a vision model actually understands the photo and then reimagines it. There's no experience without the model.

🎖️ Badges & prizes targeted

• —Tiny Titan — runs on a 1.3B model (≤4B cap).
• —OpenBMB · Best MiniCPM Build — MiniCPM-V 4.6 is the core of the experience.
• —Off-Brand — fully custom cinematic UI (lower-third captions, film grain, REC ticker).
• —Best Use of Modal — the vision model is served on Modal GPU.
• —Best Demo — built to be filmed: it's funny in 30 seconds and reads itself aloud.

How it works

Gradio Space (app.py)              Modal endpoint (modal_app.py)
  webcam capture  ──base64 JPEG──▶   MiniCPM-V 4.6 on an A10G GPU
  cinematic UI   ◀── narration ───   nature-doc persona prompt
  browser TTS reads it aloud (en-GB, fully client-side)

The persona prompt instructs the model to treat ordinary objects as wild creatures and never break character. Four selectable "moods" (classic, deep-sea, savanna, polar) reshape the scene. Narration is spoken via the browser's own speechSynthesis — no audio model needed, so it stays free and instant.

Run locally

pip install -r requirements.txt
export MODAL_ENDPOINT_URL="https://...modal.run"   # optional for preview
export MODAL_TOKEN="your-shared-secret"            # optional for preview
python app.py

Without the Modal variables set, the app shows a sample narration so you can preview the UI and voice offline.

Upgrade ideas

• —Swap the browser voice for OpenBMB VoxCPM2 TTS to deepen the OpenBMB story.
• —Add short video clips (MiniCPM-V 4.6 also does video) for a moving documentary.

Demo video & social post

• —🎬 Demo: <add your YouTube / uploaded video link>
• —📡 Post: <add your social post link>