In this project, I explored how a character—specifically, my cat, Ashley—would look as a representative of different subcultures.
Ashley in her natural habitat
The original dataset consists of 38 photographs gathered from a personal archive. It includes images of a cat from various angles, under different lighting conditions, and against various backgrounds.
Result
METAL / GRUNGE
PUNK / CYBERPUNK
SKATER / Y2K
RAVE / GOTH
I had to edit the prompts for a long time to make the neural network convey Ashley’s facial features, and even in the final generations, not all details look realistic; nevertheless, the generations often capture the cat’s favorite poses found in the photos.
Process
Example where the placeholder for the subculture image turned out to be a portrait
Initially, Stable Diffusion XL 1.0 (SDXL) was chosen as the base model during the process, and LoRA (Low-Rank Adaptation) was used to fine-tune the character. However, SDXL proved to be too resource-intensive and was replaced with Stable Diffusion 1.5.Another set of issues arose during the creation phase: the neural network was transforming the cat into a human and generating portraits without backgrounds or details. This was resolved by transferring the key features of Ashley’s subculture onto the background.Overall, I am satisfied with the result and hope Ashley’s cyber-doubles had a good time.
Top Attempts