The neural network was trained on examples of drawings with a distinct fantasy-themed style.
Concept
Drawings taken from board games (author is me). The idea is to understand how an AI interprets a specific style and reproduce it.
Source Images
Model Training
Base model—SDXL 1.0, fine-tuned using DreamBooth + LoRA in Google Colab on a Tesla T4 GPU. Square images were assembled from the source materials; images with a blank background are 1:1.
Image Series
Stone golem, gray stone, golem, greenery in the background
big and scary fish
Outcomes
Neural networks struggle to perceive a distinct approach or style in characters, which supports the idea that the human sets the genre, the concept, the style, and the meaning. The neural network executes it, but it cannot replace it.
1) I received assistance with code correction from ChatGPT.2) The trained model is Stable Diffusion.
