Original size 1059x1486

Training an AI on a character. Virus\_Bear

PROTECT STATUS: not protected
Longread translated automatically

For this project, I aimed to test the neural network’s ability to memorize and reproduce a visual style based on a series of my own illustrations. As the dataset, I used a collection of images featuring my character—an anthropomorphic bear. While the character’s recognizable head shape and muzzle remain consistent across all works, the clothing, accessories, glasses, headwear, color palettes, and background change.

Source Data


The source data utilized NFT images developed by me personally. A total of 100 images were used. These same cards, in a 1:1 format, are attached via the Google Drive link at the end of the study.

0
big
Original size 512x512
0

Result

Prompts:
Virus_BEAR style, a cartoon bear wearing a hat and sunglassesVirus_BEAR style, a cartoon bear relaxing on a tropical beach, wearing a summer shirt, sunglasses and a hat, holding a cocktail, sunny weather, ocean in backgroundVirus_BEAR style, a cartoon bear in a neon megapolis at night, wearing a different hat and different glasses, urban streetwear, glowing city lights, futuristic city backgroundVirus_BEAR style, a cartoon bear sitting on a rooftop in a big city at sunset, wearing a cap and round glasses, purple jacket, skyline in backgroundVirus_BEAR style, a cartoon bear playing video games in a neon arcade, wearing a cap, angular glasses, hoodie and varsity jacket, arcade machines in backgroundVirus_BEAR style, a cartoon bear in an artist studio, wearing a beret and round glasses, holding a paintbrush and palette, colorful paintings in backgroundVirus_BEAR style, a cartoon bear in futuristic streetwear, wearing a bucket hat and reflective sunglasses, graffiti style background

Original size 1254x1254
Original size 1254x1254

During the generation process, the key features of the original visual style were successfully preserved: the character’s recognizable shape, distinctive contour graphics, the combination of accessories, stylized clothing, and overall color aesthetic. Furthermore, the neural network did not simply replicate existing images but adapted the character to new settings and scenarios. As a result, new variations of the character were obtained in various situations—in an urban environment, on the beach, and in a metropolis. This demonstrated that even when the surroundings, clothing, glasses, and headwear were changed, the character maintained its integrity and visual recognizability.

Training an AI on a character. Virus\_Bear
Project created at 26.09.2026