Project Concept
This project focuses on training a generative neural network on the likeness of an original character—a black cat situated within the visual atmosphere of St. Petersburg. The main goal is to teach the model to reproduce the hero’s recognizable traits and create new images featuring him in various urban scenarios.
Examples of source images
Visual Concept
The main character is a black cat with yellow eyes, a red scarf, and dark clothing. His image is complemented by the historical architecture of St. Petersburg, narrow streets, streetlamps, snow, and a red-pink color palette. The project’s visual language is based on the combination of stylized illustration, expressive lighting, and a cinematic atmosphere.
Examples of source images (AI)
Stable Diffusion XL and the LoRA (Low-Rank Adaptation) method were used to train the character. Training was conducted in Google Colab using the Diffusers library on a dataset of original images of a black cat. After 100 training steps, the LoRA weights were saved for further generation. The resulting model was used to create new images of the character set against the visual backdrop of St. Petersburg, maintaining its core characteristics: black fur, yellow eyes, and a red scarf.
Preparing source images for model training
Generating a series of images using a trained LoRA model
Program Completion and Model Weight Saving
Stable Diffusion XL · LoRA · 512 × 512 px · 20 steps
Image Generation
After training the LoRA model, new images were created featuring a black cat with a red scarf set in the atmosphere of St. Petersburg. Using text prompts, the character was placed in various urban scenes. The series demonstrates the capabilities of the trained model and its ability to reproduce the hero’s image in new contexts.
Project Outcome
Through training the LoRA model, we successfully created a series of images featuring an original character—a black cat in a red scarf—set within the atmosphere of St. Petersburg. Generating images from various text prompts allowed us to showcase the character in several urban scenes and evaluate the model’s potential for adapting to our unique visual identity.
