Scrabblecam

Scrabblecam solves a very specific and fiddly problem: turning a photograph of a real, physical Scrabble board into a machine-readable game state. It uses state-of-the-art deep learning models for board recognition, which lets it handle boards from various manufacturers and different visual styles rather than only a single official set. Separate recognition is also available for the player's tile rack.

Once the board state and rack are known, moves are generated by a custom implementation of an algorithm described in academic research, which finds the best-scoring plays for any valid input in a fraction of a second. The tool supports standard 15x15 boards and multiple official word lists, including English CSW21, English NWL23, French ODS9, and Italian.

Beyond the interactive demo, Scrabblecam exposes public APIs for board state recognition, rack recognition, and move generation, each returning structured JSON. The service is free to use, though the API is subject to strict rate limiting for heavy consumers. The project is explicitly independent and unaffiliated with the trademark holders of the game.