AI Projects — 01

An AI pipeline that generates complete, layered 3D game environments from a text prompt — in minutes. Built for indie developers who carry the creative ambition of AAA studios with a fraction of the resources.

Year

2025

Institution

IAAC — MAAI

Team

Elena Papakosta · Mısra Özgök Genç · Nouhaila Elmalouli · Ramy Anka · Salvador Cantuarias

Output

Blender Plugin · Open Dataset · Alpha Community

GG Worlds — 3D generated environment

The Pipeline

01

Topography

Text prompt → diffusion model → greyscale heightmap. LoRA fine-tuned on 1000+ real cities sourced from OpenStreetMap and Google Earth Engine.

02

Streets

Topography-aware ControlNet generates a street network that respects terrain — highways follow valleys, settlements avoid steep slopes.

03

Buildings

A second ControlNet places building footprints aligned to the street grid, varying density by zone type extracted from the prior layer.

04

Landmarks

Three modes — existing landmarks via Google Places API, style-blended hybrids, or fully diffusion-generated structures — converted to 3D via TRELLIS.

Assembled Layers

Assembled 3D city — topography, streets and buildings
Topography and street layer
Generated landmarks

Stack

LoRA Fine-tuning ControlNet TRELLIS img→3D Stable Diffusion Blender OpenStreetMap Google Earth Engine Google Places API 1000+ City Dataset

Links