URBAN JUNGLES v1.0 for FLUX

Trigger word: urjungle

Recommended strength: 1

I'm European, and I'm a chaos puppy to boot. Flux loves creating very rectilinear cities, big wide roads, long straight lines, all very... regularised. So I trained this LORA on 40 images of more chaotic, organic urban landscapes. I wanted something that would produce maze-like cityscapes, places you can imagine getting lost in. I'm pleased with the result.

All the images above are just the raw output of Flux + this LORA. Most of them were generated using the following phrase:

urjungle photograph of sprawling XX city rooftops and buildings seen YY taken from a nearby rooftop.

I replaced XX with a location -- 'Tokyo' or 'Siena', for example, or 'random collection of buildings from all times and places'. I also replaced YY with a time of day, a type of light, or some other visual style modifier -- 'at golden hour', 'in the snow', etc.

The end part, 'taken from a nearby rooftop', reliably brings the image down from a fairly distant aerial panorama to a more immediate urban sprawl. You can experiment with all sorts of view-related ideas though.

Note that if you just use the term 'urjungle' without any reason to make Flux think otherwise, it will default to jungles. My bad. However, as the last images in the showcase demonstrate, you can use this LORA to produce other things with a similar sprawling pattern, and you can force it into artistic styles if you fight it a bit or turn the strength down. It should play fine with other LORAs too, particularly at lower strengths, if you want to avoid photographic images.

As a final note, I'll say that the last image in the reel is a (fairly half-hearted) homage to Argus's excellent novel THE DAILY GRIND. If a good-natured, idealistic story featuring, amongst other things, an infinite office cubicle dungeon populated by terrifying cameracondas and sentient staplers appeals to you, definitely go check it out.