Retrocat: Stable Diffusion Training

The project aimed to train the generative neural network Stable Diffusion to create images featuring a unique character—an astronaut cat—rendered in a specific retro-futurist style. The foundation was set by the aesthetic of flat vector illustration, characterized by a distinct orange-cyan color palette and clean geometric lines.The goal was to test whether the neural network could not merely draw a cat, but actually «learn» a specific authorial style while maintaining the character’s recognizability across entirely different scenes and contexts.

Examples of source images generated for dataset formation