00024 0005 0001.0

 

Date - December - 2024

Research - Flexible urban generation and scalable damage models

​

Tools

​

Photoshop - Layouts and editing

Stable Diffusion - Forge/ComfyUI Toolkits

Handmade Lora Models - Urban and Industrial Photography

I set out to create a near-future city and then deconstruct it. I wanted to test whether an internal offline AI model could help me quickly mutate building shapes using specific style loras I developed.

​

Lighting and layouts were steered by prompts and specific guide sketches added using ControlNet. Building design and damage was a blend of lora models.

​

AI handled the art style and city setting with ease, delivering a fast and highly flexible workflow that let me quickly explore many variables.

 

Applying damage using a specific lora was very simple and scalable. This is an area where AI excels, leveraging its chaotic side effects to great advantage. Sadly, video rendering still struggles to understand a lot of the damage because most models probably lack the relevant reference data.