The idea is simple: upload a screenshot of your game, and the solver reads the board for you, then shows the best move for each of your 3 pieces, step by step.

Try it live: https://blockblastsolversi.com

Here's how it works.

Two problems, not one

1. Vision. A screenshot has to become data: an exact 8×8 board plus the shapes of the 3 pieces in the tray.

2. Search. Given that board, try the placements for all 3 pieces and keep the sequence that leaves the board healthiest.

Reading the board with Python

The screenshot analysis runs on a small FastAPI service:

• Find the board area in the image and split it into an 8×8 grid
• Classify each cell (filled / empty) by sampling colors near the cell center
• Extract the tray pieces the same way, as mini shape grids

Themes and skins change the block colors, so fixed RGB matching failed constantly. Comparing relative color differences instead fixed most of it — on 71 real test screenshots, board-cell detection lands around 99% accuracy.

Planning the moves

Once the board is data, the search is straightforward:

1. Generate every legal placement for piece 1, 2 and 3
2. Simulate the game's line clears (rows + columns) after each sequence
3. Score the final board: reward clears and combos, punish rough surfaces, trapped holes and clutter

The search runs in the browser, so the scoring function has to be cheap and bad branches get pruned early. The result feels instant, even on a phone.

Stack

• Next.js + TypeScript on the front end (search runs client-side)
• Python + FastAPI for screenshot analysis
• Capacitor wrapper for the Android app

Links

• Live solver: https://blockblastsolversi.com
• GitHub: https://github.com/buzdarama/block-blast-solver