arcki is a world building playground: fly around a 3d globe, wipe out anything you don’t like, and drop in your own vision instead. if you can describe it, you can generate it, place it, and see it live on the map.
what if you could change the world around you, delete existing buildings and insert anything you imagine. arcki starts with that feeling and turns it into a real, navigable space.
arcki lets you travel the world on a 3d globe and explore any location. see a building you hate? delete it and replace it with a 3d model of your choice. you can upload your own glb file or generate one by describing what you want and choosing a style. we return a 3d model plus four perspective shots (top, down, front, side) so you can decide what fits. it’s a tool for anyone with a vision, from sustainability tweaks to fully reimagined skylines.
change the weather and time of day. gizmos let you move and rotate models precisely. a library + asset manager keeps your models organized and reusable.
the frontend is next.js with mapbox powering the globe and map interactions. the backend is a pipeline: prompt → dalle for four viewpoints → trellis for a 3d model. redis and postgresql cache and store generations so good models don’t get lost.
the map is the canvas. mapbox handles navigation, tiles, and the globe camera while our layer manages model placement and transforms. user actions like delete, place, move, and rotate update a scene graph stored in the backend. generation runs as a small pipeline: prompt goes to dalle for four views, trellis reconstructs the mesh, then we store the model and its metadata. we cache generation results in redis and persist models in postgresql so the same idea never has to be generated twice.