Concept
I wanted to test whether a generative neural network could learn to reproduce my illustrations and adopt the distinctive features of my style. To do this, I trained the model on my own work so that I could later compare the generations with the originals. The main goal of the project is to investigate how accurately the neural network can convey the characteristic traits of the illustrations and maintain the recognizability of the characters.
code snippet
code snippet
Generations
Prompt Examples: «flwrstyle, lush botanical illustration with abundant blue and dusty pink flowers, „
„many delicate green leaves, thin winding stems, tiny buds, „
„soft pastel gradients, fine pink outlines, warm cream background, „
„airy elegant composition, decorative hand-drawn floral artwork“
Summary
The neural network successfully conveyed the overall atmosphere of my illustration: the color palette, the botanical motifs, and the soft mood. It retained the core principles of stylization by using simplified shapes, smooth lines, and pastel shades.
Additionally, the network performed well in rendering the female character, preserving her emotional tone.
It was slightly less successful in capturing the minimalism and flatness of the illustration, but overall, it managed to translate the general stylistic characteristics.
