Project Concept
Urman—in Bashkir and Tatar, «dense forest, ” taiga. This is a hybrid universe where Soviet panel housing developments and the forests of Siberia, the Urals, and Bashkortostan merge into a single ecosystem.
With this project, I wanted to understand what nature will do with what remains after we are gone, when everything on earth has vanished. In my view, what will remain won’t be ruins, but a symbiosis of nature and cities. Concrete becomes bark, the seams between panels become overgrown with moss, and bus stops and courtyards transform into part of the forest. The city doesn’t disappear; it grows into the urban landscape.
Source images
The Learning Process
The process consisted of four steps:
1. Dataset Collection. The code automatically searches Wikimedia Commons for images using the queries «Khrushchevka, ” „panel apartment building, ” „taiga, ” „Bashkortostan forests, ” „Ural, ” and others. It checks the license, crops the images to a 1:1 square, and resizes them to 1024×1024.
2. Captioning. The BLIP model describes each image, and the keyword URMAN is added to the description. Example: „photo in URMAN style, a lake surrounded by trees and mountains.“
3. Series Generation and Comparison of Training Stages
Resulting Series
City and Nature
A Soviet-era bus stop crafted from birch bark and moss, set against the taiga
An old Soviet trolleybus, overgrown with thick moss, its body resembling tree bark
A row of Soviet garages overgrown with moss, birch trees growing on the roofs
A courtyard bench grown from a fallen mossy tree, panel apartment blocks in the background
A metro station where the columns are pine trunks and the ceiling is a forest canopy
Fusion of Textures
Lichen pattern on a rock forming a grid of Soviet apartment windows
Close-up of a concrete panel wall covered with thick moss and lichen
A Soviet-era panel apartment block made of pine bark
Facade of a Soviet panel building made of birch bark
Forest floor of cracked concrete panels dotted with moss, ferns, and lingonberries
Detailed results commentary
Using Generative AI
Stable Diffusion XL 1.0 — the base model for training and generation.
BLIP — automated captions for the dataset.
