Most "AI LEGO generators" I found online do the same thing: generate a picture of something brick-shaped and call it done. Cool render, but you can't actually buy those bricks and build it - there's no real model behind the image.
So that's what I built. The prompt-to-image-to-3D-mesh part is the easy, already existing half. The hard problem, and the part I actually spent most of my time on, is turning an arbitrary 3D shape into something physically buildable:
- Voxelize the mesh into a brick-scale grid
- Tile it with real, purchasable parts - staggering seams across layers so it doesn't crack apart like a stack of loose bricks would
- Run structural analysis on the result (a connectivity graph, gravity load propagation) to find anything that's not actually attached to the main structure
- Auto-repair whatever's weak - reconnect it if there's a hidden path, otherwise remove it, before it ever reaches you
- Refine the surface with slopes and tiles
Where it's at right now: a few real users are trying it, real generations are completing, but $0 in revenue so far. Working on understanding that part next.
If you want to poke at it: brickforgerai.com
(Not affiliated with the LEGO Group - LEGO® is a trademark of its owner. Part geometry from the LDraw parts library, CCAL 2.0.)
The $0 revenue with real generations is the interesting part.
Are users treating it as a fun generator, or asking for outputs they’d actually want to build?