Find a Puzzle

Private browser-based recognition

Identify a puzzle from a photograph

Upload or take a photograph and compare it with the Crafty Puzzles image archive. The quick matcher runs entirely in your browser. For difficult cases, you can choose an enhanced local vision model — it is downloaded only after you choose to use it, and your photograph is not uploaded.

Upload a puzzle photograph

No photograph selected.

Quick local match

Uses image fingerprints, colour and edge patterns. No model download and no external connection is required.

Enhanced private recognition

Uses a browser-based vision model to re-rank the strongest archive matches. The model is loaded only if you choose this option.

Use the guided finder insteadDescribe it instead

Privacy: the selected photograph stays in your browser. Enhanced recognition downloads model files and browser code from Hugging Face and jsDelivr after you consent; those services receive normal web-request information such as your IP address, but not your puzzle photograph.

Recognition results

Likely matches

Choose a photograph to begin.

How private photo matching works

The quick matcher compares a small visual fingerprint of your photograph with fingerprints generated from Crafty Puzzles archive photographs. The enhanced matcher first narrows the catalogue, then uses an image-embedding model in the browser to compare your photograph with the strongest reference images.

Neither method needs an OpenAI account or an API key. Your photograph is not submitted to Crafty Puzzles or an image-recognition API. If you choose enhanced recognition, only the model files and recognition software are downloaded to your browser.

Optional enhanced recognition

Enhanced matching uses a compact MobileCLIP vision model that runs on your device. It is not loaded during normal browsing.

  • Your selected puzzle photograph stays in your browser.
  • Model files are downloaded only after you choose Use enhanced recognition.
  • The first use downloads the recognition software and model (typically tens of megabytes); your browser can cache these files for later use.
  • The model and browser library are supplied by Hugging Face/jsDelivr, so those services receive normal download-request information such as your IP address — but not the photo you selected.