Your camera roll, turned into a personality card.
Rollcard looks through the photos on your phone, works out the themes that say the most about you, and turns them into a card you can share.
- Finds your themes. Rollcard analyses your camera roll with an AI vision-language model and matches it against a library of 180+ personas, such as Battlestation, Summit Chaser, Float Like a Butterfly and Loud Pipes Save Lives.
- Shows the evidence. Each theme on the card is paired with one of your own photos that matched it.
- Adds rarity. Every theme gets a rarity, such as Common or Rare, and so does the card as a whole.
- Lets you share it. One tap to share your Rollcard.
- Read the gallery. With the user's permission, the app loads photos from the camera roll.
- Match photos to personas. Each persona has a set of text descriptions of what its photos look like. SigLIP, a vision-language model, scores how well each photo matches each description. Because SigLIP compares images with text directly, adding a new persona only needs new descriptions, not new training data.
- Apply thresholds. A photo counts towards a persona only if its score clears that persona's confidence threshold. One photo can count towards several personas, so this is multi-label classification.
- Filter out false matches. A photo can match a description without telling you anything about the person, so two rules handle that:
- Utility photos don't count. Screenshots, receipts and notes are excluded from scoring.
- Personas have a signal type.
- Embodiment personas need the photo to show the user actually doing the thing. A saved picture of someone else's tattoo shouldn't make you "the tattooed one".
- Affinity personas accept saved images, because saving anime art still says you're into anime.
- Build the card. The top themes go onto the card, each shown with a photo that matched it.
Each persona is defined in JSON with:
- a display name
- a tagline
- a set of detection prompts
- a confidence threshold
- a signal type (embodiment or affinity)
This keeps the library easy to grow and tune without touching the app code.
| Part | Technology |
|---|---|
| Mobile app | React Native, Expo |
| Image–text matching | SigLIP |
| Backend | Firebase |
| Image hosting | Cloudinary |
You'll need Node.js, the Expo Go app on your phone, and your own Firebase and Cloudinary accounts, with their config added to the app.
git clone https://github.com/Rokuu010/rollcard.git
cd rollcard
npm install
npx expo startScan the QR code with Expo Go. Your phone and computer need to be on the same Wi-Fi network; if they can't be, run npx expo start --tunnel instead.
Complete: working end to end, from gallery scan to shareable card.